Quantization improvements for k_quants #2707

ikawrakow · 2023-08-22T09:58:22Z

This PR improves k_quants perplexity scores by tweaking the quantization approach and quantization mixes. It is fully backward compatible (but obviously one needs to re-quantize the models to take advantage of these improvements).

The most significant gains are for LLAMA_FTYPE_MOSTLY_Q2_K, where perplexity is reduced by a significant margin while slightly reducing the model size (e.g., from 2.67 GiB to 2.63 GiB for 7B). See graphs below.

Significant improvements are also observed for LLAMA_FTYPE_MOSTLY_Q3_K_M and LLAMA_FTYPE_MOSTLY_Q4_K_S for LLaMA-v2-7B. This comes at the expense of a slightly increased model size (e.g., at 7B, 3.59 GiB vs 3.56 GiB for Q4_K_S and 3.07 GiB vs 3.06 GiB for Q3_K_M).

Other quantization types / models are slightly better for LLaMA-v2 (but the change is much smaller compared to those mentioned above), or basically the same for LLaMA-v1.

Note on LLAMA_FTYPE_MOSTLY_Q2_K: strictly speaking, this is now mostly a Q3_K quantization. All tensors are quantized using Q3_K, except for attention K and Q, which are Q2_K, and output.weight, which is Q6_K as usual. I considered naming it LLAMA_FTYPE_MOSTLY_Q3_K_XS or similar, but given that this model is smaller and better than the previous LLAMA_FTYPE_MOSTLY_Q2_K, so the existing Q2_K model would have been useless in comparison, I decided that it is simpler to just re-use the LLAMA_FTYPE_MOSTLY_Q2_K designation for this new quantization mix.

The following graph shows perplexity vs model size for the LLaMA-v2-7B model and a context length of 512. Black dots/lines are for current master (i.e., after the merge of the GGUF related changes). Red dots/lines depict the results of this PR. Results for Q4_0, Q4_1, Q5_0 and Q5_1 on current master are shown in blue for comparison. The perplexity of the fp16 model is 5.7963. The new Q6_K quantization arrives at 5.8067 (so, 0.18% higher) compared to 5.8118 (0.27% higher) on Master.

The following graph is the same as the above, but with a smaller plot range to better appreciate the perplexity differences in the 4-6 bit quantization range.

Similar to the above graphs, but for the LLaMA-v1-7B model.

* Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above.

For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138.

With it, we get PPL = 5.8828 for L2-7B Q4_K_S.

Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K

Green-Sky · 2023-08-22T10:33:59Z

would it be possible to add some kind of meta information the the gguf to be able to determine if it was generated using the improvements of this pr. maybe generic like date/builnumber or more specific like k-quants-v1.1 or something. (whatever makes sense, but gguf now has easy extensibility)

The following graph is the same as the above, but with a smaller plot range to better appreciate the perplexity differences in the 4-6 bit quantization range.

I loled

klosax · 2023-08-22T11:27:58Z

would it be possible to add some kind of meta information the the gguf to be able to determine if it was generated using the improvements of this pr

Currently main will print the number of tensors of each quantization format.

ikawrakow · 2023-08-22T11:51:40Z

Currently main will print the number of tensors of each quantization format.

Yes, but one might decide to change the quantization strategy, so even though all tensors are quantized with the same type, result is still different. For instance, in this PR I have changed Q4_K and Q5_K to use the newly added function make_qkx2_quants() instead of the previous approach in make_qkx1_quants(). Hence, a date (or any other version identifier such as the commit hash) as requested by @Green-Sky will be a very useful thing to have in the meta data (and be printed, so people know what kind of quantized model they are using).

As it stands, when running with this PR, LLAMA_FTYPE_MOSTLY_Q2_K to LLAMA_FTYPE_MOSTLY_Q3_K_L get all reported as Mostly Q3_K - Medium, LLAMA_FTYPE_MOSTLY_Q4_K_S and LLAMA_FTYPE_MOSTLY_Q4_K_M are both reported as Mostly Q4_K - Medium, etc. This will be really confusing if it remains this way.

ggerganov · 2023-08-22T11:55:51Z

#2710 adds the ftype field back to the meta data.

Feel free to extend further the meta info with version/date/commit/etc. As long as the added KV info is optional, we can extend it anyway we like

KerfuffleV2 · 2023-08-22T12:22:45Z

@TheBloke - in case you didn't see this. Might be a reason to hold off on conversion for a bit if you haven't started yet.

IgnacioFDM · 2023-08-22T12:57:21Z

Any benchmarks on models larger than 7B?

TheBloke · 2023-08-22T13:28:37Z

Looks fantastic! 6.0x for Q3 is amazing

ikawrakow · 2023-08-22T13:35:38Z

Any benchmarks on models larger than 7B?

Yes, I did comparison for both 13B LLaMA's. Bud development was done on a branch that did not have the GGUF changes. When I was ready to submit the PR, I rebased on master, which brought in the GGUF changes, which changes the perplexity results. The change is actually quite dramatic for LLaMA-v2-13B: fp16 perplexity on current master is 5.1195, while what I had before was PPL = 5.1000. This is due to the change in rms_eps. Before GGUF, the default rms_eps was 5e-6. After GGUF, it is taken from the model meta data and there is no way to modify it, so it ends up being 1e-5. It was discussed e.g. in #2384 that 5e-6 gives lower perplexities than 1e-5, despite 1e-5 having been used during training. In any case, I'm re-running all calculations for the 13B models and will post when they become ready.

ikawrakow · 2023-08-22T15:49:37Z

OK, here is the LLaMA-v2-13B result.

First an overview:

Then with focus on the 4-6 bit quantization range:

IgnacioFDM · 2023-08-22T15:57:22Z

Massive improvement with Q2_K it looks like

cebtenzzre · 2023-08-22T17:43:19Z

The part of the graphs that surprises me the most is that q4_1 and q5_1 have higher perplexity than their q4_0/q5_0 counterparts on LLaMA-v2.

@TheBloke It makes me wonder whether you should even bother providing q4_1/q5_1 quantizations for LLaMA-v2 models, since they are bigger, slower, and lower quality. Maybe you could at least make a note on the READMEs that they are probably not useful.

Green-Sky · 2023-08-22T18:00:41Z

the values in the help of the quantization tool where not updated. @ikawrakow

mirek190 · 2023-08-22T22:38:17Z

q4.x or q5.x should be banned already as qk models are just better in everything ....

cebtenzzre · 2023-08-23T01:49:24Z

I ran some performance tests. The most noticeable change is Q2_K, which is now 40% slower.

GPU	Model	Test	t/s before	t/s PR	Speedup
P40	7b q4_0	tg128	55.94	55.91	no change
P40	7b q2_K	tg128	55.71	33.38	0.599
P40	7b q3_K_S	tg128	30.84	30.81	no change
P40	7b q3_K_M	tg128	37.35	37.20	0.996
P40	7b q3_K_L	tg128	34.79	34.77	no change
P40	7b q4_K_S	tg128	53.97	53.32	0.988
P40	7b q4_K_M	tg128	49.12	49.13	no change
P40	7b q5_K_S	tg128	42.90	42.90	no change
P40	7b q5_K_M	tg128	41.06	41.07	no change
P40	7b q6_K	tg128	33.50	33.50	no change

KerfuffleV2 · 2023-08-23T08:24:00Z

The most noticeable change is Q2_K, which is now 40% slower.

That seems surprising since this is a backward compatible change. You should be able to quantize with this version and then test with a version before the pull was committed - if you do that, you still see a large performance difference?

ikawrakow · 2023-08-23T09:04:38Z

I cannot confirm a change in performance for Q2_K on RTX-4080. I get 147.7 t/s for Q2_K quantized the old and the new way. On an older GTX-1660, I get 41.1 t/s using the old Q2_K quantization, and 33.2 t/s using the new, so 0.808X. This is due to the fact that now there are a lot of Q3_K quantized tensors, and Q3_K performance is not as good as the others on older GPU's. But a 40% drop in inference performance seems too much. @cebtenzzre Can you share details of your tests (GPU, CUDA settings)? Thanks.

TheBloke · 2023-08-23T14:26:27Z

Hey guys, a couple of quick questions:

When I run ./quantize -h I see this table:

   2  or  Q4_0   :  3.56G, +0.2166 ppl @ LLaMA-v1-7B
   3  or  Q4_1   :  3.90G, +0.1585 ppl @ LLaMA-v1-7B
   8  or  Q5_0   :  4.33G, +0.0683 ppl @ LLaMA-v1-7B
   9  or  Q5_1   :  4.70G, +0.0349 ppl @ LLaMA-v1-7B
  10  or  Q2_K   :  2.63G, +0.6717 ppl @ LLaMA-v1-7B
  12  or  Q3_K   : alias for Q3_K_M
  11  or  Q3_K_S :  2.75G, +0.5551 ppl @ LLaMA-v1-7B
  12  or  Q3_K_M :  3.07G, +0.2496 ppl @ LLaMA-v1-7B
  13  or  Q3_K_L :  3.35G, +0.1764 ppl @ LLaMA-v1-7B
  15  or  Q4_K   : alias for Q4_K_M
  14  or  Q4_K_S :  3.59G, +0.0992 ppl @ LLaMA-v1-7B
  15  or  Q4_K_M :  3.80G, +0.0532 ppl @ LLaMA-v1-7B
  17  or  Q5_K   : alias for Q5_K_M
  16  or  Q5_K_S :  4.33G, +0.0400 ppl @ LLaMA-v1-7B
  17  or  Q5_K_M :  4.45G, +0.0122 ppl @ LLaMA-v1-7B
  18  or  Q6_K   :  5.15G, -0.0008 ppl @ LLaMA-v1-7B
   7  or  Q8_0   :  6.70G, +0.0004 ppl @ LLaMA-v1-7B
   1  or  F16    : 13.00G              @ 7B
   0  or  F32    : 26.00G              @ 7B

Is it correct that Q6_K has better perplexity than Q8_0? In which case there'd no reason to include Q8_0 any more?

Also I assume it must be a measurement error that Q6_K has better perplexity than FP16? :) Like one figure is from before GGUF and one after or something? Would that also affect the Q6_K vs Q8_0 figures?

There used to be some text information displayed when ./quantize -h was run, explaining the different formats and giving recommended/not recommended. Any reason that was removed? I was going to use it for writing my new GGUF README description of the quant methods.

klosax · 2023-08-23T14:44:26Z

I suggest all +0.0004 ppl should be the difference from F32 of the PTH model which gives the lowest possible ppl. The original PTH models are using BF16 which will be cut when converting to F16.

ikawrakow · 2023-08-23T14:47:56Z

A PPL difference of +/- 0.001 is within the statistical noise for the amount of tokens in Wikitext. In the case of LLaMA-v1-7B it happens that Q6_K by chance arrives at a better PPL than fp16 (and Q8_0). But this will not be the case in general. For instance, for LLaMA-v2-7B, Q6_K has PPL = 5.8067, which is 0.18% higher than the PPL = 5.7963 for fp16 (and this is likely outside of the statistical uncertainty of the result, but one needs a proper uncertainty estimate added to the perplexity tool to confirm that). When editing the numbers I just took what was currently on master and adapted the number. Not sure why the previous explanations were removed. I was considering to update with the LLaMA-v2-7B numbers, but then decided to go for continuity and keep LLaMA-v1-7B results in the help.

klosax · 2023-08-23T15:07:06Z

The +/- ppl statistic may be is confusing for normal users to understand. Printing the real ppl may be better?

KerfuffleV2 · 2023-08-23T15:11:33Z

Printing the real ppl may be better?

As a user, what could you do with the raw ppl number except for subtracting it from some other value (like unquantized) to get a relative value?

klosax · 2023-08-23T15:19:23Z

At least print the real PPL value of the unquantized F32.

TheBloke · 2023-08-23T15:22:23Z

OK thanks for the explanations!

What is the feeling regarding Q6_K vs Q8_0? Is there enough of a statistically significant difference between Q8_0 vs Q6_K to make it worthwhile including Q8_0 still?

For example do you have a Q8_0 figure for the Llama V2 7B case you mentioned?

KerfuffleV2 · 2023-08-23T15:32:52Z

At least print the real PPL value of the unquantized F32.

Seems pretty reasonable, though I think it's still kind of hard for the user to do anything with.

I was actually the one that added the additional information to the quantize tool and my first pass included a lot more stuff. Some of the stuff from this post: #406 (comment) (note, the values are outdated)

One metric I think is actually pretty useful is % PPL increase relative to going from a 13B to 7B model. I think users that have messed with LLMs a bit will have some conception of the difference between a 13B and 7B model, so saying "this increases perplexity 50% as much as going from 13B to 7B" means more than +0.68 ppl.

What is the feeling regarding Q6_K vs Q8_0? Is there enough of a statistically significant difference between Q8_0 vs Q6_K to make it worthwhile including Q8_0 still?

The main use case I can think of is people who want to keep a high quality version of the model to requantize but don't want to keep the full 16bit model around. I.E. When using the quantize tool with --allow-requantize. The difference seems too small to care about for just running inference.

IgnacioFDM · 2023-08-23T15:38:30Z

OK thanks for the explanations!

What is the feeling regarding Q6_K vs Q8_0? Is there enough of a statistically significant difference between Q8_0 vs Q6_K to make it worthwhile including Q8_0 still?

For example do you have a Q8_0 figure for the Llama V2 7B case you mentioned?

Q8_0 could potentially be significantly faster than Q6_K if properly optimized I'd think (especially if we did INT8 activations instead of converting to FP16). But I might be mistaken.

ikawrakow · 2023-08-23T15:38:50Z

For example do you have a Q8_0 figure for the Llama V2 7B case you mentioned?

Yes. Q8_0 PPL for LLaMA-v2-7B is 5.7986, so 0.14% better than Q6_K.

While experimenting with the k_quants refinement (PR #2707), at some point I tried using Q8_0 instead of Q6_K for the output.weight tensor. This improved the PPL for all quantization types by ~0.003 for LLaMA-v2-7B, but made it worse by ~0.003 for LLaMA-v1-7B. So, basically, it very much depends on the model and the distribution of weight values in the tensors. Q6_K has only 6 bits available, but it does some extra work to minimize the difference to the float weights, so in some cases this can end up being better than the 8-bit round-to-nearest used in Q8_0. But it is unlikely this to be true in general, 2 extra bits are going to be beneficial more often than not. Overall I think Q6_K is within 0.2% of fp16 or better, so most likely indistinguishable from fp16 for most practical purposes. But I wouldn't quire remove Q8_0 yet.

KerfuffleV2 · 2023-08-23T15:41:51Z

Q6_K has only 6 bits available, but it does some extra work to minimize the difference to the float weights, so in some cases this can end up being better than the 8-bit round-to-nearest used in Q8_0.

Theoretically a new Q8_something could be added that does this extra work and is always better than Q6_K and Q8_0. Correct?

ikawrakow · 2023-08-23T15:51:28Z

Q8_0 could potentially be significantly faster than Q6_K if properly optimized I'd think (especially if we did INT8 activations instead of converting to FP16). But I might be mistaken.

Q8_0 will never be faster than Q6_K for token prediction, which, at least on current hardware, is totally memory bound. The ~30% difference in size cannot be recovered by the fewer computations needed in Q8_0 matrix multiplications. For prompt processing, yes, Q8_0 is faster than Q6_K. To give specific numbers: TG-128 on RTX-4080 is 91.3 tokens/second for Q6_K vs 78.3 t/s for Q8_0. Perplexity takes 134.6 seconds for Q8_0 and 146.9 seconds for Q6_K.

TheBloke · 2023-08-23T16:04:23Z

OK thanks very much! I will keep making Q8_0s then.

I'm definitely dropping Q4_0, Q4_1, Q5_0 and Q5_1.

ikawrakow · 2023-08-23T16:04:26Z

Theoretically a new Q8_something could be added that does this extra work and is always better than Q6_K and Q8_0. Correct?

Theoretically, yes. In practice, not so easy to make sure that it always beats (or is at least the same) as Q6_K, given how small the difference is between Q6_K and fp16.

IgnacioFDM · 2023-08-23T16:18:22Z

Q8_0 will never be faster than Q6_K for token prediction, which, at least on current hardware, is totally memory bound. The ~30% difference in size cannot be recovered by the fewer computations needed in Q8_0 matrix multiplications.

Isn't that only the case for consumer hardware? I'd expect tensor core INT8 inference to be significantly faster on A100 than the current setup with quantized mulmat.

cebtenzzre · 2023-08-23T16:59:33Z

I'm definitely dropping Q4_0, Q4_1, Q5_0 and Q5_1.

On my P40, Q5_0 is about 9% faster at token generation than Q5_K_S for a negligible difference in perplexity and file size on LLaMA-v2-7b. Could you keep that one at least?

Dampfinchen · 2023-08-23T19:36:21Z

Q8_0 will never be faster than Q6_K for token prediction, which, at least on current hardware, is totally memory bound. The ~30% difference in size cannot be recovered by the fewer computations needed in Q8_0 matrix multiplications.

Isn't that only the case for consumer hardware? I'd expect tensor core INT8 inference to be significantly faster on A100 than the current setup with quantized mulmat.

Consumer GPUs support INT4 and INT8 inference on tensor cores as well.

ikawrakow · 2023-08-24T05:47:18Z

OK, I have most of the LLaMA-v2-70B results now. Did not (yet) do Q5_1. Cannot run Q8_0 and fp16 on the computers I have available (not enough RAM).

As table:

Quantization	Model size (GiB)	Perplexity	Delta to fp16
Q4_0	36.20	3.5550	3.61%
Q4_1	40.20	3.5125	2.37%
Q5_0	44.20	3.4744	1.26%
Q2_K	27.11	3.8164	11.2%
Q3_K_S	27.70	3.7800	10.2%
Q3_K_M	30.83	3.5932	4.72%
Q3_K_L	33.67	3.5617	3.80%
Q4_K_S	36.31	3.4923	1.78%
Q4_K_M	38.54	3.4725	1.20%
Q5_K_S	44.20	3.4483	0.50%
Q5_K_M	45.41	3.4451	0.40%
Q6_K	52.70	3.4367	0.16%
fp16	128.5	3.4313	-

As graph:

KerfuffleV2 · 2023-08-24T07:21:01Z

I have most of the LLaMA-v2-70B results now.

Based on the 13B results, I guess we can expect the difference between the previous version and this pull to be very small so not really worth comparing?

ggerganov · 2023-08-24T07:30:41Z

The PPL for LLaMA v2 70B F16 is 3.4313

Here is full Metal run. Not sure why the estimated time is so off (~4 hours). It took just 1.2 hours

main: build = 1044 (44d5462)
main: seed  = 1692857864
llama_model_loader: loaded meta data with 15 key-value pairs and 723 tensors from models/llama-70b-v2/ggml-model-f16.gguf (version GGUF V1 (latest))
llama_model_loader: - tensor    0:                token_embd.weight f16      [  8192, 32000,     1,     1 ]
llama_model_loader: - tensor    1:               output_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor    2:                    output.weight f16      [  8192, 32000,     1,     1 ]
llama_model_loader: - tensor    3:              blk.0.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor    4:              blk.0.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor    5:              blk.0.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor    6:         blk.0.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor    7:            blk.0.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor    8:            blk.0.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor    9:              blk.0.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   10:           blk.0.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   11:            blk.0.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   12:              blk.1.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   13:              blk.1.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   14:              blk.1.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   15:         blk.1.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   16:            blk.1.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   17:            blk.1.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   18:              blk.1.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   19:           blk.1.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   20:            blk.1.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   21:              blk.2.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   22:              blk.2.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   23:              blk.2.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   24:         blk.2.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   25:            blk.2.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   26:            blk.2.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   27:              blk.2.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   28:           blk.2.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   29:            blk.2.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   30:              blk.3.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   31:              blk.3.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   32:              blk.3.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   33:         blk.3.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   34:            blk.3.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   35:            blk.3.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   36:              blk.3.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   37:           blk.3.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   38:            blk.3.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   39:              blk.4.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   40:              blk.4.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   41:              blk.4.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   42:         blk.4.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   43:            blk.4.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   44:            blk.4.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   45:              blk.4.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   46:           blk.4.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   47:            blk.4.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   48:              blk.5.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   49:              blk.5.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   50:              blk.5.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   51:         blk.5.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   52:            blk.5.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   53:            blk.5.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   54:              blk.5.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   55:           blk.5.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   56:            blk.5.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   57:              blk.6.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   58:              blk.6.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   59:              blk.6.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   60:         blk.6.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   61:            blk.6.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   62:            blk.6.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   63:              blk.6.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   64:           blk.6.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   65:            blk.6.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   66:              blk.7.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   67:              blk.7.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   68:              blk.7.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   69:         blk.7.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   70:            blk.7.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   71:            blk.7.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   72:              blk.7.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   73:           blk.7.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   74:            blk.7.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   75:              blk.8.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   76:              blk.8.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   77:              blk.8.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   78:         blk.8.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   79:            blk.8.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   80:            blk.8.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   81:              blk.8.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   82:           blk.8.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   83:            blk.8.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   84:              blk.9.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   85:              blk.9.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   86:              blk.9.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   87:         blk.9.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   88:            blk.9.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   89:            blk.9.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   90:              blk.9.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   91:           blk.9.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   92:            blk.9.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor   93:             blk.10.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   94:             blk.10.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   95:             blk.10.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor   96:        blk.10.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor   97:           blk.10.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor   98:           blk.10.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor   99:             blk.10.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  100:          blk.10.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  101:           blk.10.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  102:             blk.11.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  103:             blk.11.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  104:             blk.11.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  105:        blk.11.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  106:           blk.11.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  107:           blk.11.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  108:             blk.11.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  109:          blk.11.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  110:           blk.11.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  111:             blk.12.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  112:             blk.12.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  113:             blk.12.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  114:        blk.12.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  115:           blk.12.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  116:           blk.12.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  117:             blk.12.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  118:          blk.12.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  119:           blk.12.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  120:             blk.13.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  121:             blk.13.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  122:             blk.13.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  123:        blk.13.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  124:           blk.13.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  125:           blk.13.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  126:             blk.13.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  127:          blk.13.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  128:           blk.13.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  129:             blk.14.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  130:             blk.14.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  131:             blk.14.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  132:        blk.14.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  133:           blk.14.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  134:           blk.14.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  135:             blk.14.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  136:          blk.14.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  137:           blk.14.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  138:             blk.15.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  139:             blk.15.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  140:             blk.15.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  141:        blk.15.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  142:           blk.15.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  143:           blk.15.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  144:             blk.15.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  145:          blk.15.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  146:           blk.15.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  147:             blk.16.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  148:             blk.16.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  149:             blk.16.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  150:        blk.16.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  151:           blk.16.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  152:           blk.16.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  153:             blk.16.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  154:          blk.16.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  155:           blk.16.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  156:             blk.17.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  157:             blk.17.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  158:             blk.17.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  159:        blk.17.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  160:           blk.17.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  161:           blk.17.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  162:             blk.17.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  163:          blk.17.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  164:           blk.17.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  165:             blk.18.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  166:             blk.18.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  167:             blk.18.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  168:        blk.18.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  169:           blk.18.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  170:           blk.18.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  171:             blk.18.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  172:          blk.18.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  173:           blk.18.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  174:             blk.19.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  175:             blk.19.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  176:             blk.19.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  177:        blk.19.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  178:           blk.19.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  179:           blk.19.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  180:             blk.19.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  181:          blk.19.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  182:           blk.19.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  183:             blk.20.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  184:             blk.20.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  185:             blk.20.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  186:        blk.20.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  187:           blk.20.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  188:           blk.20.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  189:             blk.20.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  190:          blk.20.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  191:           blk.20.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  192:             blk.21.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  193:             blk.21.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  194:             blk.21.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  195:        blk.21.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  196:           blk.21.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  197:           blk.21.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  198:             blk.21.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  199:          blk.21.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  200:           blk.21.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  201:             blk.22.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  202:             blk.22.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  203:             blk.22.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  204:        blk.22.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  205:           blk.22.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  206:           blk.22.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  207:             blk.22.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  208:          blk.22.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  209:           blk.22.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  210:             blk.23.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  211:             blk.23.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  212:             blk.23.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  213:        blk.23.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  214:           blk.23.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  215:           blk.23.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  216:             blk.23.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  217:          blk.23.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  218:           blk.23.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  219:             blk.24.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  220:             blk.24.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  221:             blk.24.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  222:        blk.24.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  223:           blk.24.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  224:           blk.24.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  225:             blk.24.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  226:          blk.24.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  227:           blk.24.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  228:             blk.25.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  229:             blk.25.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  230:             blk.25.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  231:        blk.25.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  232:           blk.25.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  233:           blk.25.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  234:             blk.25.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  235:          blk.25.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  236:           blk.25.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  237:             blk.26.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  238:             blk.26.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  239:             blk.26.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  240:        blk.26.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  241:           blk.26.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  242:           blk.26.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  243:             blk.26.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  244:          blk.26.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  245:           blk.26.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  246:             blk.27.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  247:             blk.27.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  248:             blk.27.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  249:        blk.27.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  250:           blk.27.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  251:           blk.27.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  252:             blk.27.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  253:          blk.27.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  254:           blk.27.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  255:             blk.28.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  256:             blk.28.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  257:             blk.28.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  258:        blk.28.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  259:           blk.28.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  260:           blk.28.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  261:             blk.28.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  262:          blk.28.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  263:           blk.28.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  264:             blk.29.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  265:             blk.29.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  266:             blk.29.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  267:        blk.29.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  268:           blk.29.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  269:           blk.29.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  270:             blk.29.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  271:          blk.29.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  272:           blk.29.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  273:             blk.30.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  274:             blk.30.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  275:             blk.30.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  276:        blk.30.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  277:           blk.30.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  278:           blk.30.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  279:             blk.30.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  280:          blk.30.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  281:           blk.30.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  282:             blk.31.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  283:             blk.31.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  284:             blk.31.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  285:        blk.31.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  286:           blk.31.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  287:           blk.31.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  288:             blk.31.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  289:          blk.31.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  290:           blk.31.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  291:             blk.32.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  292:             blk.32.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  293:             blk.32.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  294:        blk.32.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  295:           blk.32.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  296:           blk.32.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  297:             blk.32.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  298:          blk.32.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  299:           blk.32.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  300:             blk.33.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  301:             blk.33.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  302:             blk.33.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  303:        blk.33.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  304:           blk.33.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  305:           blk.33.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  306:             blk.33.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  307:          blk.33.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  308:           blk.33.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  309:             blk.34.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  310:             blk.34.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  311:             blk.34.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  312:        blk.34.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  313:           blk.34.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  314:           blk.34.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  315:             blk.34.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  316:          blk.34.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  317:           blk.34.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  318:             blk.35.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  319:             blk.35.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  320:             blk.35.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  321:        blk.35.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  322:           blk.35.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  323:           blk.35.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  324:             blk.35.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  325:          blk.35.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  326:           blk.35.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  327:             blk.36.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  328:             blk.36.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  329:             blk.36.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  330:        blk.36.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  331:           blk.36.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  332:           blk.36.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  333:             blk.36.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  334:          blk.36.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  335:           blk.36.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  336:             blk.37.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  337:             blk.37.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  338:             blk.37.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  339:        blk.37.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  340:           blk.37.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  341:           blk.37.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  342:             blk.37.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  343:          blk.37.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  344:           blk.37.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  345:             blk.38.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  346:             blk.38.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  347:             blk.38.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  348:        blk.38.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  349:           blk.38.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  350:           blk.38.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  351:             blk.38.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  352:          blk.38.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  353:           blk.38.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  354:             blk.39.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  355:             blk.39.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  356:             blk.39.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  357:        blk.39.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  358:           blk.39.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  359:           blk.39.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  360:             blk.39.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  361:          blk.39.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  362:           blk.39.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  363:             blk.40.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  364:             blk.40.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  365:             blk.40.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  366:        blk.40.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  367:           blk.40.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  368:           blk.40.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  369:             blk.40.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  370:          blk.40.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  371:           blk.40.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  372:             blk.41.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  373:             blk.41.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  374:             blk.41.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  375:        blk.41.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  376:           blk.41.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  377:           blk.41.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  378:             blk.41.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  379:          blk.41.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  380:           blk.41.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  381:             blk.42.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  382:             blk.42.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  383:             blk.42.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  384:        blk.42.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  385:           blk.42.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  386:           blk.42.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  387:             blk.42.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  388:          blk.42.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  389:           blk.42.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  390:             blk.43.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  391:             blk.43.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  392:             blk.43.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  393:        blk.43.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  394:           blk.43.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  395:           blk.43.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  396:             blk.43.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  397:          blk.43.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  398:           blk.43.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  399:             blk.44.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  400:             blk.44.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  401:             blk.44.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  402:        blk.44.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  403:           blk.44.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  404:           blk.44.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  405:             blk.44.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  406:          blk.44.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  407:           blk.44.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  408:             blk.45.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  409:             blk.45.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  410:             blk.45.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  411:        blk.45.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  412:           blk.45.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  413:           blk.45.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  414:             blk.45.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  415:          blk.45.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  416:           blk.45.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  417:             blk.46.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  418:             blk.46.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  419:             blk.46.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  420:        blk.46.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  421:           blk.46.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  422:           blk.46.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  423:             blk.46.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  424:          blk.46.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  425:           blk.46.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  426:             blk.47.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  427:             blk.47.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  428:             blk.47.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  429:        blk.47.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  430:           blk.47.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  431:           blk.47.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  432:             blk.47.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  433:          blk.47.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  434:           blk.47.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  435:             blk.48.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  436:             blk.48.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  437:             blk.48.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  438:        blk.48.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  439:           blk.48.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  440:           blk.48.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  441:             blk.48.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  442:          blk.48.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  443:           blk.48.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  444:             blk.49.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  445:             blk.49.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  446:             blk.49.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  447:        blk.49.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  448:           blk.49.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  449:           blk.49.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  450:             blk.49.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  451:          blk.49.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  452:           blk.49.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  453:             blk.50.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  454:             blk.50.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  455:             blk.50.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  456:        blk.50.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  457:           blk.50.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  458:           blk.50.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  459:             blk.50.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  460:          blk.50.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  461:           blk.50.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  462:             blk.51.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  463:             blk.51.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  464:             blk.51.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  465:        blk.51.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  466:           blk.51.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  467:           blk.51.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  468:             blk.51.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  469:          blk.51.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  470:           blk.51.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  471:             blk.52.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  472:             blk.52.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  473:             blk.52.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  474:        blk.52.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  475:           blk.52.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  476:           blk.52.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  477:             blk.52.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  478:          blk.52.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  479:           blk.52.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  480:             blk.53.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  481:             blk.53.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  482:             blk.53.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  483:        blk.53.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  484:           blk.53.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  485:           blk.53.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  486:             blk.53.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  487:          blk.53.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  488:           blk.53.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  489:             blk.54.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  490:             blk.54.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  491:             blk.54.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  492:        blk.54.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  493:           blk.54.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  494:           blk.54.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  495:             blk.54.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  496:          blk.54.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  497:           blk.54.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  498:             blk.55.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  499:             blk.55.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  500:             blk.55.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  501:        blk.55.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  502:           blk.55.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  503:           blk.55.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  504:             blk.55.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  505:          blk.55.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  506:           blk.55.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  507:             blk.56.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  508:             blk.56.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  509:             blk.56.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  510:        blk.56.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  511:           blk.56.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  512:           blk.56.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  513:             blk.56.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  514:          blk.56.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  515:           blk.56.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  516:             blk.57.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  517:             blk.57.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  518:             blk.57.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  519:        blk.57.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  520:           blk.57.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  521:           blk.57.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  522:             blk.57.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  523:          blk.57.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  524:           blk.57.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  525:             blk.58.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  526:             blk.58.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  527:             blk.58.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  528:        blk.58.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  529:           blk.58.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  530:           blk.58.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  531:             blk.58.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  532:          blk.58.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  533:           blk.58.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  534:             blk.59.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  535:             blk.59.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  536:             blk.59.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  537:        blk.59.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  538:           blk.59.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  539:           blk.59.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  540:             blk.59.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  541:          blk.59.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  542:           blk.59.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  543:             blk.60.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  544:             blk.60.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  545:             blk.60.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  546:        blk.60.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  547:           blk.60.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  548:           blk.60.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  549:             blk.60.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  550:          blk.60.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  551:           blk.60.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  552:             blk.61.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  553:             blk.61.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  554:             blk.61.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  555:        blk.61.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  556:           blk.61.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  557:           blk.61.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  558:             blk.61.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  559:          blk.61.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  560:           blk.61.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  561:             blk.62.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  562:             blk.62.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  563:             blk.62.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  564:        blk.62.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  565:           blk.62.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  566:           blk.62.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  567:             blk.62.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  568:          blk.62.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  569:           blk.62.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  570:             blk.63.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  571:             blk.63.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  572:             blk.63.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  573:        blk.63.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  574:           blk.63.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  575:           blk.63.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  576:             blk.63.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  577:          blk.63.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  578:           blk.63.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  579:             blk.64.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  580:             blk.64.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  581:             blk.64.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  582:        blk.64.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  583:           blk.64.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  584:           blk.64.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  585:             blk.64.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  586:          blk.64.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  587:           blk.64.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  588:             blk.65.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  589:             blk.65.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  590:             blk.65.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  591:        blk.65.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  592:           blk.65.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  593:           blk.65.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  594:             blk.65.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  595:          blk.65.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  596:           blk.65.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  597:             blk.66.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  598:             blk.66.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  599:             blk.66.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  600:        blk.66.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  601:           blk.66.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  602:           blk.66.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  603:             blk.66.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  604:          blk.66.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  605:           blk.66.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  606:             blk.67.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  607:             blk.67.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  608:             blk.67.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  609:        blk.67.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  610:           blk.67.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  611:           blk.67.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  612:             blk.67.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  613:          blk.67.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  614:           blk.67.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  615:             blk.68.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  616:             blk.68.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  617:             blk.68.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  618:        blk.68.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  619:           blk.68.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  620:           blk.68.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  621:             blk.68.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  622:          blk.68.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  623:           blk.68.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  624:             blk.69.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  625:             blk.69.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  626:             blk.69.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  627:        blk.69.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  628:           blk.69.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  629:           blk.69.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  630:             blk.69.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  631:          blk.69.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  632:           blk.69.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  633:             blk.70.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  634:             blk.70.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  635:             blk.70.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  636:        blk.70.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  637:           blk.70.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  638:           blk.70.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  639:             blk.70.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  640:          blk.70.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  641:           blk.70.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  642:             blk.71.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  643:             blk.71.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  644:             blk.71.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  645:        blk.71.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  646:           blk.71.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  647:           blk.71.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  648:             blk.71.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  649:          blk.71.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  650:           blk.71.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  651:             blk.72.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  652:             blk.72.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  653:             blk.72.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  654:        blk.72.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  655:           blk.72.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  656:           blk.72.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  657:             blk.72.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  658:          blk.72.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  659:           blk.72.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  660:             blk.73.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  661:             blk.73.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  662:             blk.73.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  663:        blk.73.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  664:           blk.73.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  665:           blk.73.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  666:             blk.73.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  667:          blk.73.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  668:           blk.73.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  669:             blk.74.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  670:             blk.74.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  671:             blk.74.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  672:        blk.74.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  673:           blk.74.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  674:           blk.74.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  675:             blk.74.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  676:          blk.74.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  677:           blk.74.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  678:             blk.75.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  679:             blk.75.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  680:             blk.75.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  681:        blk.75.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  682:           blk.75.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  683:           blk.75.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  684:             blk.75.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  685:          blk.75.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  686:           blk.75.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  687:             blk.76.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  688:             blk.76.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  689:             blk.76.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  690:        blk.76.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  691:           blk.76.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  692:           blk.76.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  693:             blk.76.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  694:          blk.76.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  695:           blk.76.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  696:             blk.77.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  697:             blk.77.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  698:             blk.77.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  699:        blk.77.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  700:           blk.77.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  701:           blk.77.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  702:             blk.77.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  703:          blk.77.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  704:           blk.77.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  705:             blk.78.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  706:             blk.78.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  707:             blk.78.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  708:        blk.78.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  709:           blk.78.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  710:           blk.78.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  711:             blk.78.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  712:          blk.78.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  713:           blk.78.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  714:             blk.79.attn_q.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  715:             blk.79.attn_k.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  716:             blk.79.attn_v.weight f16      [  8192,  1024,     1,     1 ]
llama_model_loader: - tensor  717:        blk.79.attn_output.weight f16      [  8192,  8192,     1,     1 ]
llama_model_loader: - tensor  718:           blk.79.ffn_gate.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  719:           blk.79.ffn_down.weight f16      [ 28672,  8192,     1,     1 ]
llama_model_loader: - tensor  720:             blk.79.ffn_up.weight f16      [  8192, 28672,     1,     1 ]
llama_model_loader: - tensor  721:          blk.79.attn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - tensor  722:           blk.79.ffn_norm.weight f32      [  8192,     1,     1,     1 ]
llama_model_loader: - kv   0:                       general.architecture str     
llama_model_loader: - kv   1:                               general.name str     
llama_model_loader: - kv   2:                       llama.context_length u32     
llama_model_loader: - kv   3:                     llama.embedding_length u32     
llama_model_loader: - kv   4:                          llama.block_count u32     
llama_model_loader: - kv   5:                  llama.feed_forward_length u32     
llama_model_loader: - kv   6:                 llama.rope.dimension_count u32     
llama_model_loader: - kv   7:                 llama.attention.head_count u32     
llama_model_loader: - kv   8:              llama.attention.head_count_kv u32     
llama_model_loader: - kv   9:     llama.attention.layer_norm_rms_epsilon f32     
llama_model_loader: - kv  10:                          general.file_type u32     
llama_model_loader: - kv  11:                       tokenizer.ggml.model str     
llama_model_loader: - kv  12:                      tokenizer.ggml.tokens arr     
llama_model_loader: - kv  13:                      tokenizer.ggml.scores arr     
llama_model_loader: - kv  14:                  tokenizer.ggml.token_type arr     
llama_model_loader: - type  f32:  161 tensors
llama_model_loader: - type  f16:  562 tensors
llm_load_print_meta: format         = GGUF V1 (latest)
llm_load_print_meta: arch           = llama
llm_load_print_meta: vocab type     = SPM
llm_load_print_meta: n_vocab        = 32000
llm_load_print_meta: n_merges       = 0
llm_load_print_meta: n_ctx_train    = 4096
llm_load_print_meta: n_ctx          = 512
llm_load_print_meta: n_embd         = 8192
llm_load_print_meta: n_head         = 64
llm_load_print_meta: n_head_kv      = 8
llm_load_print_meta: n_layer        = 80
llm_load_print_meta: n_rot          = 128
llm_load_print_meta: n_gqa          = 8
llm_load_print_meta: f_norm_eps     = 1.0e-05
llm_load_print_meta: f_norm_rms_eps = 1.0e-05
llm_load_print_meta: n_ff           = 28672
llm_load_print_meta: freq_base      = 10000.0
llm_load_print_meta: freq_scale     = 1
llm_load_print_meta: model type     = 70B
llm_load_print_meta: model ftype    = mostly F16
llm_load_print_meta: model size     = 68.98 B
llm_load_print_meta: general.name   = LLaMA v2
llm_load_print_meta: BOS token = 1 '<s>'
llm_load_print_meta: EOS token = 2 '</s>'
llm_load_print_meta: UNK token = 0 '<unk>'
llm_load_print_meta: LF token  = 13 '<0x0A>'
llm_load_tensors: ggml ctx size = 131565.25 MB
llm_load_tensors: mem required  = 131565.25 MB (+  160.00 MB per state)
....................................................................................................
llama_new_context_with_model: kv self size  =  160.00 MB
ggml_metal_init: allocating
ggml_metal_init: loading '/Users/ggerganov/development/github/llama.cpp/ggml-metal.metal'
ggml_metal_init: loaded kernel_add                            0x113106ee0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_add_row                        0x113107620 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_mul                            0x113107b60 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_mul_row                        0x1131081b0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_scale                          0x1131086f0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_silu                           0x113108c30 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_relu                           0x113109170 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_gelu                           0x1131096b0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_soft_max                       0x113109d80 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_diag_mask_inf                  0x11310a400 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_f16                   0x113205590 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q4_0                  0x113205ef0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q4_1                  0x1132065c0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q2_K                  0x113206c90 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q3_K                  0x113207360 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q4_K                  0x113207a30 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q5_K                  0x113208100 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_get_rows_q6_K                  0x113306f40 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_rms_norm                       0x113307770 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_norm                           0x1133080d0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_f16_f32                0x113704530 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q4_0_f32               0x113704dd0 | th_max =  896 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q4_1_f32               0x113705550 | th_max =  896 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q2_K_f32               0x113705e50 | th_max =  640 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q3_K_f32               0x1137065d0 | th_max =  704 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q4_K_f32               0x11310aa60 | th_max =  576 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q5_K_f32               0x11310b300 | th_max =  576 | th_width =   32
ggml_metal_init: loaded kernel_mul_mat_q6_K_f32               0x11310bc80 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_f16_f32                 0x1132088a0 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q4_0_f32                0x113209180 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q4_1_f32                0x113706c70 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q2_K_f32                0x113308770 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q3_K_f32                0x113308e10 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q4_K_f32                0x1133095d0 | th_max =  768 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q5_K_f32                0x113309d90 | th_max =  704 | th_width =   32
ggml_metal_init: loaded kernel_mul_mm_q6_K_f32                0x11310c320 | th_max =  704 | th_width =   32
ggml_metal_init: loaded kernel_rope                           0x11310c980 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_alibi_f32                      0x1137071b0 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_cpy_f32_f16                    0x113707b80 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_cpy_f32_f32                    0x113708430 | th_max = 1024 | th_width =   32
ggml_metal_init: loaded kernel_cpy_f16_f16                    0x113708ce0 | th_max = 1024 | th_width =   32
ggml_metal_init: recommendedMaxWorkingSetSize  = 147456.00 MB
ggml_metal_init: hasUnifiedMemory              = true
ggml_metal_init: maxTransferRate               = built-in GPU
llama_new_context_with_model: compute buffer total size =  145.41 MB
llama_new_context_with_model: max tensor size =   500.00 MB
ggml_metal_add_buffer: allocated 'data            ' buffer, size = 110592.00 MB, offs =            0
ggml_metal_add_buffer: allocated 'data            ' buffer, size = 21473.28 MB, offs = 115439812608, (132065.72 / 147456.00)
ggml_metal_add_buffer: allocated 'eval            ' buffer, size =     1.42 MB, (132067.14 / 147456.00)
ggml_metal_add_buffer: allocated 'kv              ' buffer, size =   162.00 MB, (132229.14 / 147456.00)
ggml_metal_add_buffer: allocated 'alloc           ' buffer, size =   144.02 MB, (132373.16 / 147456.00)

system_info: n_threads = 16 / 24 | AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | VSX = 0 | 
perplexity: tokenizing the input ..
perplexity: calculating perplexity over 655 chunks, batch_size=512
perplexity: 21.47 seconds per pass - ETA 3 hours 54.40 minutes
[1]3.0177,[2]3.3428,[3]3.8718,[4]3.3837,[5]3.1850,[6]3.0186,[7]3.0333,[8]3.0344,[9]2.9744,[10]2.9192,[11]2.8739,[12]2.9128,[13]3.0101,[14]3.1590,[15]3.3160,[16]3.2632,[17]3.3276,[18]3.4259,[19]3.3624,[20]3.4237,[21]3.4344,[22]3.3608,[23]3.3745,[24]3.3487,[25]3.3381,[26]3.2516,[27]3.1746,[28]3.1237,[29]3.0685,[30]2.9913,[31]2.9330,[32]2.9428,[33]2.9190,[34]2.9255,[35]2.9385,[36]2.9692,[37]2.9755,[38]2.9795,[39]3.0035,[40]3.0395,[41]3.0563,[42]3.0874,[43]3.0924,[44]3.1373,[45]3.1631,[46]3.1706,[47]3.2040,[48]3.2140,[49]3.2255,[50]3.2191,[51]3.2372,[52]3.2472,[53]3.2885,[54]3.2967,[55]3.2900,[56]3.2434,[57]3.2204,[58]3.2464,[59]3.2732,[60]3.3094,[61]3.3189,[62]3.3675,[63]3.3935,[64]3.4052,[65]3.4332,[66]3.4374,[67]3.4546,[68]3.4731,[69]3.5022,[70]3.5340,[71]3.5619,[72]3.5952,[73]3.6369,[74]3.6527,[75]3.6676,[76]3.6844,[77]3.7020,[78]3.7016,[79]3.7280,[80]3.7360,[81]3.7569,[82]3.7490,[83]3.7277,[84]3.7284,[85]3.7362,[86]3.7348,[87]3.6951,[88]3.6524,[89]3.6128,[90]3.5817,[91]3.5583,[92]3.5413,[93]3.5320,[94]3.5054,[95]3.5123,[96]3.4905,[97]3.4703,[98]3.4532,[99]3.4390,[100]3.4341,[101]3.4333,[102]3.4269,[103]3.4249,[104]3.4189,[105]3.4138,[106]3.4103,[107]3.4050,[108]3.4076,[109]3.3945,[110]3.3808,[111]3.3814,[112]3.3858,[113]3.3781,[114]3.3634,[115]3.3566,[116]3.3469,[117]3.3343,[118]3.3498,[119]3.3640,[120]3.3877,[121]3.3983,[122]3.4198,[123]3.4489,[124]3.4667,[125]3.4725,[126]3.5028,[127]3.5299,[128]3.5527,[129]3.5316,[130]3.5422,[131]3.5481,[132]3.5520,[133]3.5500,[134]3.5624,[135]3.5695,[136]3.5720,[137]3.5776,[138]3.5749,[139]3.5751,[140]3.5788,[141]3.5666,[142]3.5685,[143]3.5555,[144]3.5500,[145]3.5472,[146]3.5493,[147]3.5592,[148]3.5662,[149]3.5694,[150]3.5756,[151]3.5854,[152]3.5889,[153]3.5875,[154]3.5901,[155]3.5994,[156]3.6032,[157]3.6160,[158]3.6200,[159]3.6256,[160]3.6348,[161]3.6465,[162]3.6372,[163]3.6361,[164]3.6274,[165]3.6174,[166]3.6078,[167]3.5933,[168]3.5751,[169]3.5644,[170]3.5612,[171]3.5509,[172]3.5486,[173]3.5446,[174]3.5332,[175]3.5237,[176]3.5201,[177]3.5141,[178]3.5061,[179]3.5017,[180]3.5009,[181]3.4939,[182]3.4867,[183]3.4847,[184]3.4878,[185]3.4811,[186]3.4831,[187]3.4899,[188]3.4845,[189]3.4960,[190]3.4979,[191]3.5121,[192]3.5233,[193]3.5342,[194]3.5410,[195]3.5583,[196]3.5687,[197]3.5830,[198]3.5937,[199]3.5990,[200]3.5866,[201]3.5713,[202]3.5632,[203]3.5575,[204]3.5494,[205]3.5488,[206]3.5483,[207]3.5440,[208]3.5426,[209]3.5441,[210]3.5511,[211]3.5609,[212]3.5684,[213]3.5777,[214]3.5840,[215]3.5887,[216]3.6001,[217]3.6130,[218]3.6250,[219]3.6287,[220]3.6308,[221]3.6298,[222]3.6332,[223]3.6331,[224]3.6303,[225]3.6294,[226]3.6438,[227]3.6478,[228]3.6567,[229]3.6647,[230]3.6657,[231]3.6782,[232]3.6755,[233]3.6708,[234]3.6656,[235]3.6534,[236]3.6529,[237]3.6508,[238]3.6572,[239]3.6547,[240]3.6530,[241]3.6562,[242]3.6604,[243]3.6618,[244]3.6590,[245]3.6593,[246]3.6547,[247]3.6516,[248]3.6513,[249]3.6506,[250]3.6561,[251]3.6534,[252]3.6535,[253]3.6470,[254]3.6439,[255]3.6408,[256]3.6328,[257]3.6293,[258]3.6271,[259]3.6275,[260]3.6238,[261]3.6225,[262]3.6224,[263]3.6213,[264]3.6074,[265]3.6105,[266]3.6119,[267]3.6111,[268]3.6160,[269]3.6169,[270]3.6214,[271]3.6292,[272]3.6259,[273]3.6268,[274]3.6286,[275]3.6335,[276]3.6377,[277]3.6476,[278]3.6562,[279]3.6634,[280]3.6675,[281]3.6749,[282]3.6812,[283]3.6927,[284]3.7010,[285]3.7093,[286]3.7185,[287]3.7157,[288]3.7206,[289]3.7187,[290]3.7072,[291]3.6944,[292]3.6817,[293]3.6710,[294]3.6658,[295]3.6677,[296]3.6693,[297]3.6683,[298]3.6676,[299]3.6635,[300]3.6532,[301]3.6469,[302]3.6400,[303]3.6353,[304]3.6298,[305]3.6259,[306]3.6197,[307]3.6140,[308]3.6111,[309]3.6038,[310]3.5991,[311]3.5946,[312]3.5918,[313]3.5887,[314]3.5865,[315]3.5774,[316]3.5714,[317]3.5638,[318]3.5534,[319]3.5606,[320]3.5698,[321]3.5747,[322]3.5744,[323]3.5705,[324]3.5709,[325]3.5781,[326]3.5811,[327]3.5834,[328]3.5871,[329]3.5912,[330]3.5929,[331]3.6018,[332]3.6007,[333]3.6069,[334]3.6029,[335]3.6013,[336]3.6042,[337]3.6050,[338]3.6043,[339]3.5963,[340]3.5952,[341]3.6003,[342]3.6043,[343]3.6083,[344]3.6092,[345]3.6125,[346]3.6134,[347]3.6171,[348]3.6215,[349]3.6254,[350]3.6257,[351]3.6280,[352]3.6296,[353]3.6273,[354]3.6264,[355]3.6289,[356]3.6330,[357]3.6325,[358]3.6403,[359]3.6388,[360]3.6378,[361]3.6391,[362]3.6445,[363]3.6518,[364]3.6558,[365]3.6591,[366]3.6621,[367]3.6685,[368]3.6691,[369]3.6719,[370]3.6750,[371]3.6745,[372]3.6803,[373]3.6834,[374]3.6843,[375]3.6844,[376]3.6899,[377]3.6891,[378]3.6923,[379]3.6947,[380]3.6918,[381]3.6910,[382]3.6879,[383]3.6873,[384]3.6886,[385]3.6886,[386]3.6889,[387]3.6905,[388]3.6888,[389]3.6865,[390]3.6838,[391]3.6814,[392]3.6801,[393]3.6785,[394]3.6819,[395]3.6828,[396]3.6802,[397]3.6852,[398]3.6895,[399]3.6949,[400]3.6952,[401]3.6963,[402]3.6967,[403]3.6994,[404]3.7043,[405]3.6937,[406]3.6836,[407]3.6760,[408]3.6734,[409]3.6793,[410]3.6862,[411]3.6941,[412]3.7035,[413]3.7092,[414]3.7121,[415]3.7162,[416]3.7198,[417]3.7259,[418]3.7247,[419]3.7277,[420]3.7313,[421]3.7381,[422]3.7396,[423]3.7413,[424]3.7459,[425]3.7499,[426]3.7534,[427]3.7555,[428]3.7584,[429]3.7607,[430]3.7640,[431]3.7717,[432]3.7735,[433]3.7725,[434]3.7708,[435]3.7730,[436]3.7739,[437]3.7801,[438]3.7859,[439]3.7847,[440]3.7831,[441]3.7758,[442]3.7734,[443]3.7743,[444]3.7747,[445]3.7743,[446]3.7756,[447]3.7775,[448]3.7776,[449]3.7763,[450]3.7771,[451]3.7755,[452]3.7665,[453]3.7577,[454]3.7497,[455]3.7409,[456]3.7409,[457]3.7332,[458]3.7271,[459]3.7194,[460]3.7099,[461]3.7004,[462]3.6912,[463]3.6846,[464]3.6765,[465]3.6681,[466]3.6596,[467]3.6504,[468]3.6435,[469]3.6346,[470]3.6274,[471]3.6192,[472]3.6116,[473]3.6049,[474]3.5978,[475]3.5906,[476]3.5834,[477]3.5800,[478]3.5788,[479]3.5735,[480]3.5676,[481]3.5622,[482]3.5558,[483]3.5482,[484]3.5399,[485]3.5362,[486]3.5299,[487]3.5255,[488]3.5242,[489]3.5169,[490]3.5089,[491]3.5027,[492]3.4950,[493]3.4898,[494]3.4846,[495]3.4782,[496]3.4741,[497]3.4668,[498]3.4591,[499]3.4557,[500]3.4489,[501]3.4410,[502]3.4359,[503]3.4306,[504]3.4231,[505]3.4184,[506]3.4189,[507]3.4168,[508]3.4165,[509]3.4189,[510]3.4212,[511]3.4254,[512]3.4295,[513]3.4328,[514]3.4375,[515]3.4333,[516]3.4341,[517]3.4344,[518]3.4343,[519]3.4366,[520]3.4383,[521]3.4399,[522]3.4420,[523]3.4431,[524]3.4478,[525]3.4504,[526]3.4515,[527]3.4526,[528]3.4502,[529]3.4523,[530]3.4518,[531]3.4534,[532]3.4586,[533]3.4624,[534]3.4618,[535]3.4633,[536]3.4607,[537]3.4606,[538]3.4616,[539]3.4626,[540]3.4626,[541]3.4602,[542]3.4614,[543]3.4636,[544]3.4660,[545]3.4664,[546]3.4683,[547]3.4676,[548]3.4661,[549]3.4672,[550]3.4669,[551]3.4664,[552]3.4669,[553]3.4661,[554]3.4661,[555]3.4660,[556]3.4669,[557]3.4682,[558]3.4674,[559]3.4689,[560]3.4656,[561]3.4671,[562]3.4666,[563]3.4673,[564]3.4720,[565]3.4745,[566]3.4763,[567]3.4754,[568]3.4776,[569]3.4783,[570]3.4813,[571]3.4831,[572]3.4845,[573]3.4864,[574]3.4851,[575]3.4846,[576]3.4854,[577]3.4852,[578]3.4850,[579]3.4854,[580]3.4837,[581]3.4831,[582]3.4845,[583]3.4869,[584]3.4894,[585]3.4869,[586]3.4847,[587]3.4865,[588]3.4902,[589]3.4949,[590]3.4980,[591]3.5007,[592]3.5012,[593]3.4948,[594]3.4906,[595]3.4851,[596]3.4863,[597]3.4850,[598]3.4815,[599]3.4825,[600]3.4827,[601]3.4779,[602]3.4753,[603]3.4750,[604]3.4737,[605]3.4732,[606]3.4733,[607]3.4733,[608]3.4728,[609]3.4737,[610]3.4760,[611]3.4758,[612]3.4776,[613]3.4766,[614]3.4747,[615]3.4722,[616]3.4749,[617]3.4733,[618]3.4719,[619]3.4706,[620]3.4649,[621]3.4628,[622]3.4585,[623]3.4587,[624]3.4598,[625]3.4611,[626]3.4618,[627]3.4638,[628]3.4650,[629]3.4657,[630]3.4680,[631]3.4705,[632]3.4742,[633]3.4743,[634]3.4769,[635]3.4769,[636]3.4751,[637]3.4701,[638]3.4666,[639]3.4616,[640]3.4558,[641]3.4519,[642]3.4488,[643]3.4436,[644]3.4391,[645]3.4348,[646]3.4321,[647]3.4272,[648]3.4254,[649]3.4251,[650]3.4270,[651]3.4297,[652]3.4299,[653]3.4331,[654]3.4314,[655]3.4313,
llama_print_timings:        load time = 57976.93 ms
llama_print_timings:      sample time =     0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
llama_print_timings: prompt eval time = 4176637.30 ms / 335360 tokens (   12.45 ms per token,    80.29 tokens per second)
llama_print_timings:        eval time =     0.00 ms /     1 runs   (    0.00 ms per token,      inf tokens per second)
llama_print_timings:       total time = 4242444.38 ms
ggml_metal_free: deallocating

KerfuffleV2 · 2023-08-24T07:37:08Z

Not sure why the estimated time is so off (~4 hours). It took just 1.2 hours

Anything like a random CPU usage spike or swapping while the first block is running will throw off the whole time estimation calculation since it's only based on the first block time. I've seen it happen from time to time.

ikawrakow · 2023-08-24T08:34:10Z

Not sure why the estimated time is so off (~4 hours). It took just 1.2 hours

I'm observing this on a regular basis. The very first time you load a model, the time estimate is off by a sizable margin. If you stop the process after getting the time estimate, on next run with the same model the time estimate is fairly reliable.

commit 3416c986d9d9a31c3cdefd7e7bd4d9438d72ba35 Merge: 5eb17f0 4c4e435 Author: YellowRoseCx <[email protected]> Date: Fri Aug 25 13:46:56 2023 -0500 Merge remote-tracking branch 'upstream/concedo' commit 5eb17f02c8638e003bb91bddf95ccf54d2ad0c12 Author: YellowRoseCx <[email protected]> Date: Fri Aug 25 13:38:21 2023 -0500 ROCm Port update * use hipblas based on cublas * Update Makefile for the Cuda kernels * Expand arch list and make it overrideable * Fix multi GPU on multiple amd architectures with rocblas_initialize() (#5) * add hipBLAS to README * new build arg LLAMA_CUDA_MMQ_Y * fix half2 decomposition * Add intrinsics polyfills for AMD * AMD assembly optimized __dp4a * Allow overriding CC_TURING * use "ROCm" instead of "CUDA" * ignore all build dirs * Add Dockerfiles * fix llama-bench * fix -nommq help for non CUDA/HIP --------- Co-Authored-By: YellowRoseCx <[email protected]> Co-Authored-By: ardfork <[email protected]> Co-Authored-By: funnbot <[email protected]> Co-Authored-By: Engininja2 <[email protected]> Co-Authored-By: Kerfuffle <[email protected]> Co-Authored-By: jammm <[email protected]> Co-Authored-By: jdecourval <[email protected]> commit 4c4e4358ed54c397d3f0f5bc268f1ac59d909f57 Author: Concedo <[email protected]> Date: Thu Aug 24 22:12:56 2023 +0800 fixed linux build error commit 661bede62fe216632d099678a9dac08de7a68a4e Author: Concedo <[email protected]> Date: Thu Aug 24 21:16:16 2023 +0800 optimize tokenize method commit b95a4ccb228ebfac12e5ce4b445f073ca67b99d2 Author: Concedo <[email protected]> Date: Thu Aug 24 20:41:49 2023 +0800 added a token counting endpoint, set mmq as default commit 81a0ef342ce1e583f6a5b060252565dbd59e1d8d Author: Concedo <[email protected]> Date: Thu Aug 24 16:26:38 2023 +0800 updated lite, switched to unminified source commit 598d4d89ab3aaa539ddf36784306071f1411814a Author: Concedo <[email protected]> Date: Thu Aug 24 15:45:33 2023 +0800 fix for config file loading. from kcpp settings file commit a3b994962673e681aafd9503781c7470acdcc63f Merge: b8372d4 2d86b2e Author: Concedo <[email protected]> Date: Thu Aug 24 15:22:17 2023 +0800 Merge remote-tracking branch 'pop/add_config_arg' into concedo_experimental commit b8372d44666531f5d17cbe264912fbe5548fd54b Merge: 8263fd7 6e91a1b Author: Concedo <[email protected]> Date: Thu Aug 24 15:21:24 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # .gitignore # README.md # tests/CMakeLists.txt commit 6e91a1b0706c2e0e52b9d9be7ee82d3c1e7a33c1 Author: Evan Jones <[email protected]> Date: Thu Aug 24 00:07:13 2023 -0400 llama : fix grammar sometimes generating null char (#2756) commit 44d5462b5cddc1c5cbcd7647646f7b55b175b01f Author: Georgi Gerganov <[email protected]> Date: Wed Aug 23 23:44:19 2023 +0300 readme : fix link commit c7868b075377c8c3fa916ea7c1aca600f44bed55 Author: Georgi Gerganov <[email protected]> Date: Wed Aug 23 23:43:00 2023 +0300 minor : fix trailing whitespace commit 79da24b58c1ea72340e64f799a4717d372207676 Author: Georgi Gerganov <[email protected]> Date: Wed Aug 23 23:41:16 2023 +0300 readme : update hot topics commit cf658adc832badaaa2ca119fe86070e5a830f8f6 Author: Georgi Gerganov <[email protected]> Date: Wed Aug 23 23:08:04 2023 +0300 llm : add Falcon support (#2717) * llama : refactor GGUF constants into static maps * llama : check if model architecture is known * llama : refactor llama_model_load_internal() * gguf : add KV constant maps * llm : read arch-specific KVs * convert : add dummy scores + types * falcon : load tensor data (CPU only) * llama : fix loading progress bar * llama : add arch member to llama_model * falcon : CPU inference working * falcon : support non-40B models * falcon : minor * llama : minor updates ggml-ci * convert-falcon-hf-to-gguf.py : fix special token mapping * llama.cpp : llama default UNK token = id 0 * llama.cpp : fix bpe tokenizer * llama.cpp : fix the fix of bpe tokenizer * ggml : pass eps to ggml_norm * metal : implement RoPE (mode = 2) + avoid ggml_repeat * ggml : ggml_repeat always creates new tensor * falcon : copy-paste self-attention from LLaMA * metal : print extra compute pipeline info * falcon : minor changes (still chasing the Metal problem) * llama.cpp : fix linefeed token * metal : fix GELU kernel numerical stability by using precise::tanh * metal : temporary workaround for the concurrency optimization bug * falcon : add CUDA offloading (#2739) * llama : better model naming and size reporting * llama : prep new tokenizer support * llama : advanced BPE tokenizer based on ggllm.cpp imlpementation * llama : remove oboslete comment ggml-ci * common : remove obsolete BPE API + disable test-tokenizer-1 * llama : revert BPE special-case in llama_byte_to_token() * cuda : add TODOs for RoPE NeoX implementation * llama : default special tokens based on vocab type * perplexity : add log for start of tokenization --------- Co-authored-by: klosax <[email protected]> Co-authored-by: slaren <[email protected]> commit a192860cfec89a38d59a943623bf595b1fe4495b Author: Georgi Gerganov <[email protected]> Date: Wed Aug 23 22:37:39 2023 +0300 minor : fix trailing whitespace commit 95385241a91a616788a3bb76d12c9b7b2379ca2d Author: Olivier Chafik <[email protected]> Date: Wed Aug 23 20:33:05 2023 +0100 examples : restore the functionality to import llama2.c models (#2685) * Fix import of llama2.c models that don't share weights between embedding layers * llama2c: reinstate ggmlv3 conversion output + update readme w/ gguf conv * llama2.c: comment out legacy "load from ggml model" logic * llama2.c: convert special-cased "<0xXX>" single byte tokens from tokenizer.bin commit 335acd2ffd7b04501c6d8773ab9fcee6e7bf8639 Author: slaren <[email protected]> Date: Wed Aug 23 16:46:54 2023 +0200 fix convert-lora-to-ggml.py (#2738) commit 5290c38e6e9b66ee2b543e560e301c1a1a90929c Author: klosax <[email protected]> Date: Wed Aug 23 16:46:03 2023 +0200 main : insert bos if no tokens (#2727) * main.cpp : insert bos if no tokens * Update examples/main/main.cpp * Update examples/main/main.cpp --------- Co-authored-by: Georgi Gerganov <[email protected]> commit cc34dbda9681418a2b18382446b90cdcec398d82 Author: akawrykow <[email protected]> Date: Wed Aug 23 07:31:34 2023 -0700 gitignore : fix for windows (#2729) commit 7c2227a1972a4add4b5c118e4914c086513d0382 Author: Cebtenzzre <[email protected]> Date: Wed Aug 23 10:29:09 2023 -0400 chmod : make scripts executable (#2675) commit f19dca04ea5fbf9a0b2753091d93464585d5c73b Author: JohnnyB <[email protected]> Date: Wed Aug 23 15:28:22 2023 +0100 devops : RPM Specs (#2723) * Create llama-cpp.srpm * Rename llama-cpp.srpm to llama-cpp.srpm.spec Correcting extension. * Tested spec success. * Update llama-cpp.srpm.spec * Create lamma-cpp-cublas.srpm.spec * Create lamma-cpp-clblast.srpm.spec * Update lamma-cpp-cublas.srpm.spec Added BuildRequires * Moved to devops dir commit 8263fd7bdb247f2c3ff21debb50b22bd9b030339 Author: askmyteapot <[email protected]> Date: Thu Aug 24 00:15:48 2023 +1000 Update llama_v3.cpp (#393) Fixing C2065 compiler error. Missed '3' on 3 separate identifiers (kB > kB3, MB > MB3) commit bfdc596d58fbd9bbadd2352705af4373005e1411 Author: Concedo <[email protected]> Date: Wed Aug 23 19:19:52 2023 +0800 gguf reader in file format detection commit 8207214b6a37a46526cee9e72d4c9092b9d1872f Author: Kawrakow <[email protected]> Date: Wed Aug 23 12:57:12 2023 +0300 Fix values shown in the quantize tool help (#2735) Co-authored-by: Iwan Kawrakow <[email protected]> commit 62959e740e8759d246ac8d09036950efde09981c Author: Kawrakow <[email protected]> Date: Wed Aug 23 12:56:42 2023 +0300 Strided perplexity (#2714) * Implementing strided computation of perplexity * Alternative way to output PPL results --------- Co-authored-by: Iwan Kawrakow <[email protected]> commit 7f7ddd5002040804e33fcdbde44aa22f8635f57d Author: IgnacioFDM <[email protected]> Date: Wed Aug 23 06:31:09 2023 -0300 Fix ggml to gguf conversion on Windows (#2733) This fixes `RuntimeWarning: overflow encountered in long_scalars` Credit: anon (not mine) commit af170fc2db1186d3002b602d909c52c22de4a076 Merge: 981c913 b8ad1b6 Author: Concedo <[email protected]> Date: Wed Aug 23 17:08:09 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # README.md # llama.cpp # scripts/sync-ggml.sh # tests/test-tokenizer-0.cpp commit 981c9131f0f20c10099735c1e353534b5bfe1e59 Author: Concedo <[email protected]> Date: Wed Aug 23 16:07:07 2023 +0800 gguf for llama is working commit b8ad1b66b23f9b2e6e4531e9a62753323036a556 Author: Xiao-Yong Jin <[email protected]> Date: Wed Aug 23 02:12:12 2023 -0500 server : allow json array in prompt or content for direct token input (#2306) * server: allow json array in prompt or content We accept an array of strings and numbers representing tokens, in addition to the current string valued prompt or content. This allows direct token input, so that any special tokens can be processed and used at the frontend during the construction of the json data, before sending to the server. And the server does not need to know or parse special tokens from textual input. With this, we can use EOS and BOS used in llama-2-chat models. * server: use tokenizePrompt(json) and default "" if empty prompt * server: fix prompt check * server: tokenize endpoint no longer adds BOS commit f5fe98d11bdf9e7797bcfb05c0c3601ffc4b9d26 Author: Evan Jones <[email protected]> Date: Tue Aug 22 21:01:57 2023 -0400 docs : add grammar docs (#2701) * docs : add grammar docs * tweaks to grammar guide * rework GBNF example to be a commented grammar commit 777f42ba18b29f25c71ff8de3ecf97b8017304c0 Author: Kerfuffle <[email protected]> Date: Tue Aug 22 17:39:39 2023 -0600 Improve handling of special tokens in GGML to GGUF converter (#2725) * Improve UNK, BOS, EOS token handling when converting without metadata. * Allow importing as a module. * Remove some obsolete code and minor cleanups. * Set default UNK token mapping from -1 to 0 in llama.cpp * Try to handle overflow due to buggy Windows Python with a better error message commit 46ef5b5fcf4c366e1fb27726b6394adbbf8fd0ea Author: goerch <[email protected]> Date: Tue Aug 22 23:10:42 2023 +0200 llama : fix whitespace escaping in tokenizer (#2724) commit c63bb1d16a70c03440671b76954bb767513cead8 Author: Johannes Gäßler <[email protected]> Date: Tue Aug 22 22:47:05 2023 +0200 CUDA: use mul_mat_q kernels by default (#2683) commit 3b6cfe7c927df178ca3c11643c3ec93e143471c9 Author: Alex Petenchea <[email protected]> Date: Tue Aug 22 21:58:16 2023 +0300 convert.py : clarifying error message (#2718) commit 800c9635b4a9390126f397870f3a825fc7455bd1 Author: Jiahao Li <[email protected]> Date: Wed Aug 23 02:27:06 2023 +0800 Fix CUDA softmax by subtracting max value before exp (#2665) commit deb7dfca4b9725cd295d1426db75fe8e0a6d5312 Author: Georgi Gerganov <[email protected]> Date: Tue Aug 22 20:05:59 2023 +0300 gguf : add ftype meta info to the model (#2710) * llama : add ftype meta info to the model ggml-ci * convert.py : add ftype when converting (does not work) * convert.py : fix Enum to IntEnum ggml-ci commit bac66994cf356cf488078c056831396eb4ce31d5 Author: Kawrakow <[email protected]> Date: Tue Aug 22 19:14:09 2023 +0300 Quantization imrovements for k_quants (#2707) * Improve LLaMA-2 2-, 3- and 4-bit quantization * Q3_K_S: use Q5_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q4_K_S: use Q6_K for 1st 2 layers of attention.wv and feed_forward.w2 * Q2_K and Q3_K_M: use Q5_K instead of Q4_K for 1st 2 layers of attention.wv and feed_forward.w2 This leads to a slight model sized increase as follows: Q2_K : 2.684G vs 2.670G Q3_K_S: 2.775G vs 2.745G Q3_K_M: 3.071G vs 3.057G Q4_K_S: 3.592G vs 3.563G LLaMA-2 PPL for context 512 changes as follows: Q2_K : 6.6691 vs 6.8201 Q3_K_S: 6.2129 vs 6.2584 Q3_K_M: 6.0387 vs 6.1371 Q4_K_S: 5.9138 vs 6.0041 There are improvements for LLaMA-1 as well, but they are way smaller than the above. * Minor 4-bit quantization improvement For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138. * Some more fine tuning * Adding make_qkx2_quants With it, we get PPL = 5.8828 for L2-7B Q4_K_S. * Another minor improvement * Q2_K improvement Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K * Iterating * Revert Q5_K back to make_qkx1_quants * Better Q6_K * make_qkx2_quants is better for Q5_K after all * Fix after rebasing on master * Fix for changed tensor names --------- Co-authored-by: Iwan Kawrakow <[email protected]> commit 39cc83e8c9fafe1494c4996b07f97afed29c9f27 Merge: 2d17c22 6381d4e Author: Concedo <[email protected]> Date: Tue Aug 22 23:12:47 2023 +0800 incomplete merge, compiles but generates rubbish commit 519c981f8b65ee6c87c2965539685ced0a17223b Author: slaren <[email protected]> Date: Tue Aug 22 16:03:12 2023 +0200 embedding : evaluate prompt in batches (#2713) commit 1123f7fbdfb8012e46f05e903e6f675922916378 Author: slaren <[email protected]> Date: Tue Aug 22 15:25:19 2023 +0200 ggml-cuda : use graph allocator (#2684) use a different function for no_alloc to avoid breaking backwards compat, fixes lora remove 512 n_batch limit fixed 2048 batch size cleanup Co-authored-by: Johannes Gäßler <[email protected]> commit ef3f333d3775600d1646a9fa249aca532d15fb89 Author: Georgi Gerganov <[email protected]> Date: Tue Aug 22 14:22:08 2023 +0300 ggml : sync latest (SAM + SD operators, CUDA alibi) (#2709) * ggml : sync latest (SAM + SD operators, CUDA alibi) ggml-ci * ggml : fix tabs commit 2d17c224376c0fb2d6cfce8726de5a5f7b666bfe Merge: 36b0c5b dadbed9 Author: Concedo <[email protected]> Date: Tue Aug 22 18:20:06 2023 +0800 functional commit before gguf merge commit 8e4364f2af9cd5d57240f23e83c0e29bc068bc02 Author: slaren <[email protected]> Date: Tue Aug 22 09:56:03 2023 +0200 llama-bench : minor fixes (#2695) commit 1e3bc523d8053a77df3ac7126a84d0297ee97ef6 Author: Kylin <[email protected]> Date: Tue Aug 22 15:14:23 2023 +0800 ggml : support CUDA's half type for aarch64(#1455) (#2670) * ggml: support CUDA's half type for aarch64(#1455) support CUDA's half type for aarch64 in ggml_fp16_t definition * ggml: use __CUDACC__ to recognise nvcc compiler commit 14b1d7e6f720dee41ce5a826376df738096d9033 Author: Shouzheng Liu <[email protected]> Date: Tue Aug 22 02:18:40 2023 -0400 metal : add missing barriers for mul-mat (#2699) commit 226255b44ef2c2794bfac48d101d35a9c2dbb965 Author: Jhen-Jie Hong <[email protected]> Date: Tue Aug 22 08:32:00 2023 +0800 server : fallback to default if client param is null (#2688) * server : fallback to default if client param is null * server : do not overwrite 404 if status is 500 from exception_handler commit 930523c8e1cbbee5449c055daa894917fac6805e Author: Kerfuffle <[email protected]> Date: Mon Aug 21 18:01:34 2023 -0600 Fix convert-llama-ggmlv3-to-gguf.py vocab conversion (#2698) When converting without metadata, the hex value for bytes entries weren't 0 padded to 2 digits. commit 2d86b2e219ef988878bdea7e33a534aad3a744da Author: Pontus Mårdnäs <[email protected]> Date: Mon Aug 21 23:46:56 2023 +0200 Add --config argument commit c8dba409e6d6a754090f08e6a862c5ffdd52e421 Author: Georgi Gerganov <[email protected]> Date: Mon Aug 21 23:40:22 2023 +0300 py : remove obsolete script commit 6381d4e110bd0ec02843a60bbeb8b6fc37a9ace9 Author: Georgi Gerganov <[email protected]> Date: Mon Aug 21 23:07:43 2023 +0300 gguf : new file format with flexible meta data (beta) (#2398) * gguf : first API pass * gguf : read header + meta data * gguf : read tensor info * gguf : initial model loading - not tested * gguf : add gguf_get_tensor_name() * gguf : do not support passing existing ggml_context to gguf_init * gguf : simplify gguf_get_val * gguf : gguf.c is now part of ggml.c * gguf : read / write sample models * gguf : add comments * refactor : reduce code duplication and better API (#2415) * gguf : expose the gguf_type enum through the API for now * gguf : add array support * gguf.py : some code style changes * convert.py : start a new simplified implementation by removing old stuff * convert.py : remove GGML vocab + other obsolete stuff * GGUF : write tensor (#2426) * WIP: Write tensor * GGUF : Support writing tensors in Python * refactor : rm unused import and upd todos * fix : fix errors upd writing example * rm example.gguf * gitignore *.gguf * undo formatting * gguf : add gguf_find_key (#2438) * gguf.cpp : find key example * ggml.h : add gguf_find_key * ggml.c : add gguf_find_key * gguf : fix writing tensors * gguf : do not hardcode tensor names to read * gguf : write sample tensors to read * gguf : add tokenization constants * quick and dirty conversion example * gguf : fix writing gguf arrays * gguf : write tensors one by one and code reuse * gguf : fix writing gguf arrays * gguf : write tensors one by one * gguf : write tensors one by one * gguf : write tokenizer data * gguf : upd gguf conversion script * Update convert-llama-h5-to-gguf.py * gguf : handle already encoded string * ggml.h : get array str and f32 * ggml.c : get arr str and f32 * gguf.py : support any type * Update convert-llama-h5-to-gguf.py * gguf : fix set is not subscriptable * gguf : update convert-llama-h5-to-gguf.py * constants.py : add layer norm eps * gguf.py : add layer norm eps and merges * ggml.h : increase GGML_MAX_NAME to 64 * ggml.c : add gguf_get_arr_n * Update convert-llama-h5-to-gguf.py * add gptneox gguf example * Makefile : add gptneox gguf example * Update convert-llama-h5-to-gguf.py * add gptneox gguf example * Update convert-llama-h5-to-gguf.py * Update convert-gptneox-h5-to-gguf.py * Update convert-gptneox-h5-to-gguf.py * Update convert-llama-h5-to-gguf.py * gguf : support custom alignment value * gguf : fix typo in function call * gguf : mmap tensor data example * fix : update convert-llama-h5-to-gguf.py * Update convert-llama-h5-to-gguf.py * convert-gptneox-h5-to-gguf.py : Special tokens * gptneox-main.cpp : special tokens * Update gptneox-main.cpp * constants.py : special tokens * gguf.py : accumulate kv and tensor info data + special tokens * convert-gptneox-h5-to-gguf.py : accumulate kv and ti + special tokens * gguf : gguf counterpart of llama-util.h * gguf-util.h : update note * convert-llama-h5-to-gguf.py : accumulate kv / ti + special tokens * convert-llama-h5-to-gguf.py : special tokens * Delete gptneox-common.cpp * Delete gptneox-common.h * convert-gptneox-h5-to-gguf.py : gpt2bpe tokenizer * gptneox-main.cpp : gpt2 bpe tokenizer * gpt2 bpe tokenizer (handles merges and unicode) * Makefile : remove gptneox-common * gguf.py : bytesarray for gpt2bpe tokenizer * cmpnct_gpt2bpe.hpp : comments * gguf.py : use custom alignment if present * gguf : minor stuff * Update gptneox-main.cpp * map tensor names * convert-gptneox-h5-to-gguf.py : map tensor names * convert-llama-h5-to-gguf.py : map tensor names * gptneox-main.cpp : map tensor names * gguf : start implementing libllama in GGUF (WIP) * gguf : start implementing libllama in GGUF (WIP) * rm binary commited by mistake * upd .gitignore * gguf : calculate n_mult * gguf : inference with 7B model working (WIP) * gguf : rm deprecated function * gguf : start implementing gguf_file_saver (WIP) * gguf : start implementing gguf_file_saver (WIP) * gguf : start implementing gguf_file_saver (WIP) * gguf : add gguf_get_kv_type * gguf : add gguf_get_kv_type * gguf : write metadata in gguf_file_saver (WIP) * gguf : write metadata in gguf_file_saver (WIP) * gguf : write metadata in gguf_file_saver * gguf : rm references to old file formats * gguf : shorter name for member variable * gguf : rm redundant method * gguf : get rid of n_mult, read n_ff from file * Update gguf_tensor_map.py * Update gptneox-main.cpp * gguf : rm references to old file magics * gguf : start implementing quantization (WIP) * gguf : start implementing quantization (WIP) * gguf : start implementing quantization (WIP) * gguf : start implementing quantization (WIP) * gguf : start implementing quantization (WIP) * gguf : start implementing quantization (WIP) * gguf : quantization is working * gguf : roper closing of file * gguf.py : no need to convert tensors twice * convert-gptneox-h5-to-gguf.py : no need to convert tensors twice * convert-llama-h5-to-gguf.py : no need to convert tensors twice * convert-gptneox-h5-to-gguf.py : simplify nbytes * convert-llama-h5-to-gguf.py : simplify nbytes * gptneox-main.cpp : n_layer --> n_block * constants.py : n_layer --> n_block * gguf.py : n_layer --> n_block * convert-gptneox-h5-to-gguf.py : n_layer --> n_block * convert-llama-h5-to-gguf.py : n_layer --> n_block * gptneox-main.cpp : n_layer --> n_block * Update gguf_tensor_map.py * convert-gptneox-h5-to-gguf.py : load model in parts to save memory * convert-llama-h5-to-gguf.py : load model in parts to save memory * convert : write more metadata for LLaMA * convert : rm quantization version * convert-gptneox-h5-to-gguf.py : add file_type key * gptneox-main.cpp : add file_type key * fix conflicts * gguf : add todos and comments * convert-gptneox-h5-to-gguf.py : tensor name map changes * Create gguf_namemap.py : tensor name map changes * Delete gguf_tensor_map.py * gptneox-main.cpp : tensor name map changes * convert-llama-h5-to-gguf.py : fixes * gguf.py : dont add empty strings * simple : minor style changes * gguf : use UNIX line ending * Create convert-llama-7b-pth-to-gguf.py * llama : sync gguf-llama.cpp with latest llama.cpp (#2608) * llama : sync gguf-llama.cpp with latest llama.cpp * minor : indentation + assert * llama : refactor gguf_buffer and gguf_ctx_buffer * llama : minor * gitignore : add gptneox-main * llama : tokenizer fixes (#2549) * Merge tokenizer fixes into the gguf branch. * Add test vocabularies * convert : update convert-new.py with tokenizer fixes (#2614) * Merge tokenizer fixes into the gguf branch. * Add test vocabularies * Adapt convert-new.py (and fix a clang-cl compiler error on windows) * llama : sync gguf-llama with llama (#2613) * llama : sync gguf-llama with llama * tests : fix build + warnings (test-tokenizer-1 still fails) * tests : fix wstring_convert * convert : fix layer names * llama : sync gguf-llama.cpp * convert : update HF converter to new tokenizer voodoo magics * llama : update tokenizer style * convert-llama-h5-to-gguf.py : add token types * constants.py : add token types * gguf.py : add token types * convert-llama-7b-pth-to-gguf.py : add token types * gguf-llama.cpp : fix n_head_kv * convert-llama-h5-to-gguf.py : add 70b gqa support * gguf.py : add tensor data layout * convert-llama-h5-to-gguf.py : add tensor data layout * convert-llama-7b-pth-to-gguf.py : add tensor data layout * gptneox-main.cpp : add tensor data layout * convert-llama-h5-to-gguf.py : clarify the reverse permute * llama : refactor model loading code (#2620) * llama : style formatting + remove helper methods * llama : fix quantization using gguf tool * llama : simplify gguf_file_saver * llama : fix method names * llama : simplify write_header() * llama : no need to pass full file loader to the file saver just gguf_ctx * llama : gguf_file_saver write I32 * llama : refactor tensor names (#2622) * gguf: update tensor names searched in quantization * gguf : define tensor names as constants * gguf : initial write API (not tested yet) * gguf : write to file API (not tested) * gguf : initial write API ready + example * gguf : fix header write * gguf : fixes + simplify example + add ggml_nbytes_pad() * gguf : minor * llama : replace gguf_file_saver with new gguf write API * gguf : streaming support when writing files * gguf : remove oboslete write methods * gguf : remove obosolete gguf_get_arr_xxx API * llama : simplify gguf_file_loader * llama : move hparams and vocab from gguf_file_loader to llama_model_loader * llama : merge gguf-util.h in llama.cpp * llama : reorder definitions in .cpp to match .h * llama : minor simplifications * llama : refactor llama_model_loader (WIP) wip : remove ggml_ctx from llama_model_loader wip : merge gguf_file_loader in llama_model_loader * llama : fix shape prints * llama : fix Windows build + fix norm_rms_eps key * llama : throw error on missing KV paris in model meta data * llama : improve printing + log meta data * llama : switch print order of meta data --------- Co-authored-by: M. Yusuf Sarıgöz <[email protected]> * gguf : deduplicate (#2629) * gguf : better type names * dedup : CPU + Metal is working * ggml : fix warnings about unused results * llama.cpp : fix line feed and compiler warning * llama : fix strncpy warning + note token_to_str does not write null * llama : restore the original load/save session implementation Will migrate this to GGUF in the future * convert-llama-h5-to-gguf.py : support alt ctx param name * ggml : assert when using ggml_mul with non-F32 src1 * examples : dedup simple --------- Co-authored-by: klosax <[email protected]> * gguf.py : merge all files in gguf.py * convert-new.py : pick #2427 for HF 70B support * examples/gguf : no need to keep q option for quantization any more * llama.cpp : print actual model size * llama.cpp : use ggml_elements() * convert-new.py : output gguf (#2635) * convert-new.py : output gguf (WIP) * convert-new.py : add gguf key-value pairs * llama : add hparams.ctx_train + no longer print ftype * convert-new.py : minor fixes * convert-new.py : vocab-only option should work now * llama : fix tokenizer to use llama_char_to_byte * tests : add new ggml-vocab-llama.gguf * convert-new.py : tensor name mapping * convert-new.py : add map for skipping tensor serialization * convert-new.py : convert script now works * gguf.py : pick some of the refactoring from #2644 * convert-new.py : minor fixes * convert.py : update to support GGUF output * Revert "ci : disable CI temporary to not waste energy" This reverts commit 7e82d25f40386540c2c15226300ad998ecd871ea. * convert.py : n_head_kv optional and .gguf file extension * convert.py : better always have n_head_kv and default it to n_head * llama : sync with recent PRs on master * editorconfig : ignore models folder ggml-ci * ci : update ".bin" to ".gguf" extension ggml-ci * llama : fix llama_model_loader memory leak * gptneox : move as a WIP example * llama : fix lambda capture ggml-ci * ggml : fix bug in gguf_set_kv ggml-ci * common.h : .bin --> .gguf * quantize-stats.cpp : .bin --> .gguf * convert.py : fix HF tensor permuting / unpacking ggml-ci * llama.cpp : typo * llama : throw error if gguf fails to init from file ggml-ci * llama : fix tensor name grepping during quantization ggml-ci * gguf.py : write tensors in a single pass (#2644) * gguf : single pass for writing tensors + refactoring writer * gguf : single pass for writing tensors + refactoring writer * gguf : single pass for writing tensors + refactoring writer * gguf : style fixes in simple conversion script * gguf : refactor gptneox conversion script * gguf : rename h5 to hf (for HuggingFace) * gguf : refactor pth to gguf conversion script * gguf : rm file_type key and method * gguf.py : fix vertical alignment * gguf.py : indentation --------- Co-authored-by: Georgi Gerganov <[email protected]> * convert-gptneox-hf-to-gguf.py : fixes * gguf.py : gptneox mapping * convert-llama-hf-to-gguf.py : fixes * convert-llama-7b-pth-to-gguf.py : fixes * ggml.h : reverse GGUF_MAGIC * gguf.py : reverse GGUF_MAGIC * test-tokenizer-0.cpp : fix warning * llama.cpp : print kv general.name * llama.cpp : get special token kv and linefeed token id * llama : print number of tensors per type + print arch + style * tests : update vocab file with new magic * editorconfig : fix whitespaces * llama : re-order functions * llama : remove C++ API + reorganize common source in /common dir * llama : minor API updates * llama : avoid hardcoded special tokens * llama : fix MPI build ggml-ci * llama : introduce enum llama_vocab_type + remove hardcoded string constants * convert-falcon-hf-to-gguf.py : falcon HF --> gguf conversion, not tested * falcon-main.cpp : falcon inference example * convert-falcon-hf-to-gguf.py : remove extra kv * convert-gptneox-hf-to-gguf.py : remove extra kv * convert-llama-7b-pth-to-gguf.py : remove extra kv * convert-llama-hf-to-gguf.py : remove extra kv * gguf.py : fix for falcon 40b * falcon-main.cpp : fix for falcon 40b * convert-falcon-hf-to-gguf.py : update ref * convert-falcon-hf-to-gguf.py : add tensor data layout * cmpnct_gpt2bpe.hpp : fixes * falcon-main.cpp : fixes * gptneox-main.cpp : fixes * cmpnct_gpt2bpe.hpp : remove non-general stuff * Update examples/server/README.md Co-authored-by: slaren <[email protected]> * cmpnct_gpt2bpe.hpp : cleanup * convert-llama-hf-to-gguf.py : special tokens * convert-llama-7b-pth-to-gguf.py : special tokens * convert-permute-debug.py : permute debug print * convert-permute-debug-master.py : permute debug for master * convert-permute-debug.py : change permute type of attn_q * convert.py : 70b model working (change attn_q permute) * Delete convert-permute-debug-master.py * Delete convert-permute-debug.py * convert-llama-hf-to-gguf.py : fix attn_q permute * gguf.py : fix rope scale kv * convert-llama-hf-to-gguf.py : rope scale and added tokens * convert-llama-7b-pth-to-gguf.py : rope scale and added tokens * llama.cpp : use rope scale kv * convert-llama-7b-pth-to-gguf.py : rope scale fix * convert-llama-hf-to-gguf.py : rope scale fix * py : fix whitespace * gguf : add Python script to convert GGMLv3 LLaMA models to GGUF (#2682) * First pass at converting GGMLv3 LLaMA models to GGUF * Cleanups, better output during conversion * Fix vocab space conversion logic * More vocab conversion fixes * Add description to converted GGUF files * Improve help text, expand warning * Allow specifying name and description for output GGUF * Allow overriding vocab and hyperparams from original model metadata * Use correct params override var name * Fix wrong type size for Q8_K Better handling of original style metadata * Set default value for gguf add_tensor raw_shape KW arg * llama : improve token type support (#2668) * Merge tokenizer fixes into the gguf branch. * Add test vocabularies * Adapt convert-new.py (and fix a clang-cl compiler error on windows) * Improved tokenizer test But does it work on MacOS? * Improve token type support - Added @klosax code to convert.py - Improved token type support in vocabulary * Exclude platform dependent tests * More sentencepiece compatibility by eliminating magic numbers * Restored accidentally removed comment * llama : add API for token type ggml-ci * tests : use new tokenizer type API (#2692) * Merge tokenizer fixes into the gguf branch. * Add test vocabularies * Adapt convert-new.py (and fix a clang-cl compiler error on windows) * Improved tokenizer test But does it work on MacOS? * Improve token type support - Added @klosax code to convert.py - Improved token type support in vocabulary * Exclude platform dependent tests * More sentencepiece compatibility by eliminating magic numbers * Restored accidentally removed comment * Improve commentary * Use token type API in test-tokenizer-1.cpp * py : cosmetics * readme : add notice about new file format ggml-ci --------- Co-authored-by: M. Yusuf Sarıgöz <[email protected]> Co-authored-by: klosax <[email protected]> Co-authored-by: goerch <[email protected]> Co-authored-by: slaren <[email protected]> Co-authored-by: Kerfuffle <[email protected]> commit dadbed99e65252d79f81101a392d0d6497b86caa Author: Shouzheng Liu <[email protected]> Date: Mon Aug 21 06:59:29 2023 -0400 metal : fix synchronization in new matrix multiplication kernel (#2686) commit cb1c0727bd59803b439b6a3af121c99e6393ff3d Author: Kawrakow <[email protected]> Date: Mon Aug 21 11:11:31 2023 +0300 HellaSwag: split token evaluation into batches if needed (#2681) Co-authored-by: Iwan Kawrakow <[email protected]> commit 9e232f0234073358e7031c1b8d7aa45020469a3b Author: slaren <[email protected]> Date: Sun Aug 20 22:17:53 2023 +0200 ggml : move all type info to ggml_type_traits (#2663) commit 5e9ff54a675d163d9f42aad1b5b3e734f17b2701 Author: Kawrakow <[email protected]> Date: Sun Aug 20 16:44:46 2023 +0300 More efficient Hellaswag implementation (#2677) Co-authored-by: Iwan Kawrakow <[email protected]> commit b34f4bd2724733e188ec4f6074042f66a5ed28c9 Author: YellowRoseCx <[email protected]> Date: Sat Aug 19 17:12:52 2023 -0500 Update README.md commit 1f0bccb27929e261744c979bc75114955da49e98 Author: Georgi Gerganov <[email protected]> Date: Sat Aug 19 00:45:36 2023 +0300 server : better default prompt (#2646) commit f63564adfaa157ca387071d6b9a06cfaef0ef576 Author: Jhen-Jie Hong <[email protected]> Date: Sat Aug 19 05:41:32 2023 +0800 server : update xxd usage for older versions compatibility (#2649) * server : update xxd usage for older versions compatibility * remove unused $func commit 2d8b76a110d76ff6b5728ff0af8477531e4db60e Author: Adrian <[email protected]> Date: Fri Aug 18 12:39:22 2023 -0700 Add link to clojure bindings to Readme. (#2659) commit 7af633aec339367e36c867ae709088d6a801aa75 Author: Georgi Gerganov <[email protected]> Date: Fri Aug 18 17:48:31 2023 +0300 readme : incoming BREAKING CHANGE commit 097e121e2f17ed3541cf02c55ff7e9febc091b19 Author: slaren <[email protected]> Date: Fri Aug 18 12:44:58 2023 +0200 llama : add benchmark example (#2626) * llama : add benchmark example * add to examples CMakeLists.txt * fix msvc build * add missing include * add Bessel's correction to stdev calculation Co-authored-by: Johannes Gäßler <[email protected]> * improve markdown formatting * add missing include * print warning is NDEBUG is not defined * remove n_prompt and n_gen from the matrix, use each value separately instead * better checks for non-optimized builds * llama.cpp : fix MEM_REQ_SCRATCH0 reusing the value of n_ctx of the first call * fix json formatting * add sql output * add basic cpu and gpu info (linx/cuda only) * markdown: also show values that differ from the default * markdown: add build id * cleanup * improve formatting * formatting --------- Co-authored-by: Johannes Gäßler <[email protected]> commit eaf98c2649d7da705de255712f0038ac7e47c610 Author: mdrokz <[email protected]> Date: Fri Aug 18 15:47:58 2023 +0530 readme : add link to Rust bindings (#2656) commit e9b12c332ec6e215fbac4b2ef165353acbcd8319 Author: Georgi Gerganov <[email protected]> Date: Fri Aug 18 12:48:55 2023 +0300 perplexity : more meaningful ETA number - 2 decimal points commit 604b8bdfa6320bbcb018eebcc1252dfede603c6b Author: Evan Jones <[email protected]> Date: Thu Aug 17 19:54:44 2023 -0400 Fix unicode in grammars (fixes #2501) (#2553) * Fix unicode in grammars (fixes #2501) * add more comments * fix test-llama-grammar commit 10151bee2e38b5711335c4a38f6ca93b50223acf Author: staviq <[email protected]> Date: Thu Aug 17 23:34:01 2023 +0000 server : support for saving templates in browser LocalStorage (#2486) * support for templates in browser LocalStorage * sync accepted #2409 fix from upstream * convert autosave invocation to useEffect * Apply suggestions from code review Co-authored-by: Jhen-Jie Hong <[email protected]> * Regen index.html.cpp, suggested from code review --------- Co-authored-by: Jhen-Jie Hong <[email protected]> commit 0992a7b8b18a89e29a205efb48ceb559c9a08203 Author: Johannes Gäßler <[email protected]> Date: Thu Aug 17 23:57:59 2023 +0200 README: fix LLAMA_CUDA_MMV_Y documentation (#2647) commit 6ddeefad9b634c5c79e6bcf046523493ff1fdf7d Author: Henri Vasserman <[email protected]> Date: Thu Aug 17 23:11:18 2023 +0300 [Zig] Fixing Zig build and improvements (#2554) * Fix zig after console.o was split * Better include and flag management * Change LTO to option commit 36b0c5b39816c039a5235733cfcd2b4e32386ff9 Author: Concedo <[email protected]> Date: Thu Aug 17 22:45:49 2023 +0800 fix for incorrect missing backends displayed commit 8dae7ce68437faf1fa96ec0e7687b8700956ef20 Author: Kerfuffle <[email protected]> Date: Thu Aug 17 07:29:44 2023 -0600 Add --cfg-negative-prompt-file option for examples (#2591) Add --cfg-negative-prompt-file option for examples commit a73ccf1aa34de49f61bfeb7f8a679c3bfdb3abe3 Author: Georgi Gerganov <[email protected]> Date: Thu Aug 17 10:47:09 2023 +0300 llama : replace (permute + reshape + view_1d) with (view_3d) (#2538) ggml-ci commit 7cf54e1f746941279d81d485796777c01f88049c Author: drbh <[email protected]> Date: Thu Aug 17 03:41:01 2023 -0400 tests : adds simple llama grammar tests (#2618) * adds simple llama grammar tests * fix lint and add Makefile * 0 terminate code_points * avoid dangling pointers in candidate cleanup * cleanup grammar at end of test commit a872a2b28eaefc8d464eaa535c94deeb501666f9 Author: Shouzheng Liu <[email protected]> Date: Thu Aug 17 03:35:53 2023 -0400 ggml-alloc : fix discrepency between measure&eval (#2639) The GGML memory allocator consistently places a tensor within the optimal-fit memory block, which is the smallest block capable of accommodating the tensor's size. During the measurement phase, the final block is generously sized, ensuring it never qualifies as the optimal-fit block as long as there exists another block capable of accommodating the tensor. Nevertheless, in the evaluation phase, the last block is constrained in size and could potentially qualify as the optimal-fit block. Consequently, there exists the possibility of a tensor being allocated to a different region during evaluation, leading to more memory fragmentation in our scratch buffer. This recent commit guarantees uniform behavior of the allocator across both the measurement and evaluation phases, eliminating discrepancies between the two. commit 0919a0f73d95cfb93a1646a1d1741a0615fe2c5e Author: Kolen Cheung <[email protected]> Date: Wed Aug 16 21:09:49 2023 +0100 cmake : install ggml-meta.metal if LLAMA_METAL (#2449) commit ed53db86c3b0e0815331a96d7a379edb5e62472c Author: Jhen-Jie Hong <[email protected]> Date: Thu Aug 17 04:09:03 2023 +0800 metal : print error of load pipeline state (#2564) * metal : print error of load pipeline state * metal : return null if load pipeline failed commit fc8ef549e50087762a0b4f901cd74b2defcc6ae3 Author: Shouzheng Liu <[email protected]> Date: Wed Aug 16 16:08:28 2023 -0400 metal : enable ggml-alloc (#2627) * metal: enable ggml-alloc Make ggml-alloc work with concurrently dispatch. * style-fix Co-authored-by: slaren <[email protected]> --------- Co-authored-by: slaren <[email protected]> Co-authored-by: Georgi Gerganov <[email protected]> commit bf83bff6742c0f1795b4c18695a13a34ac7adf62 Author: Shouzheng Liu <[email protected]> Date: Wed Aug 16 16:07:04 2023 -0400 metal : matrix-matrix multiplication kernel (#2615) * metal: matrix-matrix multiplication kernel This commit removes MPS and uses custom matrix-matrix multiplication kernels for all quantization types. This commit also adds grouped-query attention to support llama2 70B. * metal: fix performance degradation from gqa Integers are slow on the GPU, and 64-bit divides are extremely slow. In the context of GQA, we introduce a 64-bit divide that cannot be optimized out by the compiler, which results in a decrease of ~8% in inference performance. This commit fixes that issue by calculating a part of the offset with a 32-bit divide. Naturally, this limits the size of a single matrix to ~4GB. However, this limitation should suffice for the near future. * metal: fix bugs for GQA and perplexity test. I mixed up ne02 and nb02 in previous commit. commit 075d079a72c741050a4c31a27530c8af19df70a6 Merge: 469d70b b5ffb28 Author: Concedo <[email protected]> Date: Wed Aug 16 10:43:06 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # CMakeLists.txt # Makefile # ggml-cuda.cu # llama-util.h # tests/CMakeLists.txt commit b5ffb2849d23afe73647f68eec7b68187af09be6 Author: Georgi Gerganov <[email protected]> Date: Tue Aug 15 10:04:58 2023 +0300 scripts : add helper script to get wikitext commit 469d70be45dfdac4d926c1326b579e88d0f0e040 Author: Concedo <[email protected]> Date: Tue Aug 15 13:49:05 2023 +0800 add support for precompiled binaries, used as a fallback commit 7d1196108ad330b32845546fb3472c2172a0b6b8 Author: YellowRoseCx <[email protected]> Date: Mon Aug 14 23:03:12 2023 -0500 remove force DMMV commit 3ebb00935f3f0522b75df49c2769ab1774b91380 Author: Jhen-Jie Hong <[email protected]> Date: Tue Aug 15 06:14:14 2023 +0800 server : add missing /json-schema-to-grammar.mjs (#2616) fixes #2611 commit d783f7982e0e823a2626a9956359c0d36c1a7e21 Author: Jhen-Jie Hong <[email protected]> Date: Mon Aug 14 21:37:39 2023 +0800 metal : return null instead of exit(1) (#2573) commit d75561df207d22790609ee0ad924302f66ac2599 Author: Cheng Shao <[email protected]> Date: Mon Aug 14 15:36:42 2023 +0200 server : add --numa support (#2524) commit 348acf188c9fbe66396990f2dc83229df367969b Author: Kamil Tomšík <[email protected]> Date: Mon Aug 14 15:35:16 2023 +0200 llama : add missing enum keyword in function signatures (#2610) commit 1cd06fa25eb859b14b3427a1d815a48f25fc3c34 Author: Johannes Gäßler <[email protected]> Date: Mon Aug 14 10:41:22 2023 +0200 CUDA: launch_bounds, small q4_K, q5_K mmq refactor (#2596) commit 2feb8934eb75ca63f3c42724229cce1df9579c8e Author: Jhen-Jie Hong <[email protected]> Date: Mon Aug 14 16:20:17 2023 +0800 server : fix default grammar by use empty string in the UI (#2604) commit 5517d6e69214cdead000a76983b9fe175c3f8329 Author: Jhen-Jie Hong <[email protected]> Date: Mon Aug 14 15:16:54 2023 +0800 server : implement json-schema-to-grammar.mjs & add grammar param in the UI (#2588) * server : implement json-schema-to-grammar.mjs by follow python impl * server : add grammar support in chat.mjs * server : implement grammer param in the UI * server : generate .hpp * server : remove trailing whitespaces * server : generate .hpp * server : fix sort of prop pairs * server : optimize regex & iteration commit f31b5397143009d682db90fd2a6cde83f1ef00eb Author: vxiiduu <[email protected]> Date: Mon Aug 14 13:59:16 2023 +1000 Enhance Windows 7 and below compatibility. (#2592) * Enhance Windows 7 compatibility. * Clean away unnecessary preprocessor conditional commit ee77efea2a1e3f7d153976b0934522b6bbaa62e6 Author: drbh <[email protected]> Date: Sun Aug 13 10:00:48 2023 -0400 test : add simple grammar parsing tests (#2594) * adds simple grammar parsing tests * adds cassert header commit f64d44a9b9581cd58f7ec40f4fa1c3ca5ca18e1e Author: Johannes Gäßler <[email protected]> Date: Sun Aug 13 00:24:45 2023 +0200 CUDA: Fixed OpenLLaMA 3b mmq, reduced compile time (#2590) commit cd61aa0d9e16627935c7978adf488a679ddfa745 Author: YellowRoseCx <[email protected]> Date: Sat Aug 12 17:24:31 2023 -0500 restore main_gpu parameter commit 4a042f326830271a4c31104051b7b08e08ac234e Author: Henri Vasserman <[email protected]> Date: Sat Aug 12 10:51:46 2023 +0300 gfx1100 support --------- Co-authored-by: ardfork <[email protected]> Co-authored-by: jammm <[email protected]> Co-authored-by: jdecourval <[email protected]> commit 8913bc6fea97d3cb860937b0461f455c6abe3ea1 Author: Henri Vasserman <[email protected]> Date: Fri Aug 11 10:16:02 2023 +0300 Allow overriding CC_TURING commit e77a4c37a756c002e97173f4122e088fb304e18a Author: Henri Vasserman <[email protected]> Date: Fri Aug 11 10:00:07 2023 +0300 Merge 'origin/master' into hipblas commit cc4c4e355cd553b1557d5fba2562e824db93f9b4 Author: Engininja2 <[email protected]> Date: Fri Aug 11 09:43:14 2023 +0300 New __dp4a assembly Now compatible with gfx900 and faster as well. commit 1a03b709848ce68d5bf5966237756167e2cac540 Author: Henri Vasserman <[email protected]> Date: Fri Aug 11 09:30:28 2023 +0300 Undo mess --------- Co-authored-by: ardfork <[email protected]> commit 4366ff9ba1b1f12e494118ef9b5198479022fcc5 Author: DannyDaemonic <[email protected]> Date: Thu Aug 10 13:11:36 2023 -0700 Handle `ENABLE_VIRTUAL_TERMINAL_PROCESSING` more gracefully on earlier versions of Windows. commit 811ff855a24323cafddc95c1b8aca711fef05f76 Author: Christian Demsar <[email protected]> Date: Thu Aug 10 10:28:27 2023 -0400 Add --n-predict -2 for stopping generation on full context (#2565) commit 37c9717aaa6815b6a5be21aaab970212f20fe6bf Author: Martin Krasser <[email protected]> Date: Thu Aug 10 12:16:38 2023 +0200 Fix grammar-based sampling issue in server (#2566) commit 9483288e0318a4dcc2e08eb817dfdd09c6552533 Merge: dae9dff b19edd5 Author: Concedo <[email protected]> Date: Sat Aug 12 16:04:11 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # Makefile commit b19edd54d51cef5e3616c18b1d0d8626895b2cba Author: byte-6174 <[email protected]> Date: Fri Aug 11 19:17:25 2023 -0400 Adding support for llama2.c models (#2559) commit 53dc399472d5bd35ee739b865e843b1996bd3814 Author: Equim <[email protected]> Date: Sat Aug 12 06:35:14 2023 +0800 server: fixed wrong variable name in timing json (#2579) * server: fixed wrong variable name in timing json * remove redunct entry commit dae9dffa6aa53923cfbb09ac5de7e08f34920733 Author: Concedo <[email protected]> Date: Fri Aug 11 14:54:27 2023 +0800 rename koboldcpp.dll to koboldcpp_default.dll commit 9ca4abed893685692f90413e4d43153af12342d9 Author: DannyDaemonic <[email protected]> Date: Thu Aug 10 13:11:36 2023 -0700 Handle `ENABLE_VIRTUAL_TERMINAL_PROCESSING` more gracefully on earlier versions of Windows. commit d18ecd5b9e5dde58ae08a3eef1637406159ddaca Author: YellowRoseCx <[email protected]> Date: Thu Aug 10 13:19:41 2023 -0500 make mmq gen faster for amd commit 243894a952147a4fac5b6aee748861a0df6cc2c6 Author: Henri Vasserman <[email protected]> Date: Thu Aug 10 12:14:40 2023 +0300 ws fix commit ac2f14da445ea87d73539adbd29d19ff2c9eba58 Author: Engininja2 <[email protected]> Date: Thu Aug 10 12:11:27 2023 +0300 AMD assembly optimized __dp4a Doesn't seem to work for gfx900, so commented out. commit 9dba0c985f140ddded8cbb671f139e81fff82eed Author: Henri Vasserman <[email protected]> Date: Thu Aug 10 12:09:28 2023 +0300 Fix merge --------- Co-authored-by: ardfork <[email protected]> Co-authored-by: Kerfuffle <[email protected]> commit e59fcb2bc129881f4a269fee748fb38bce0a64de Author: Christian Demsar <[email protected]> Date: Thu Aug 10 10:28:27 2023 -0400 Add --n-predict -2 for stopping generation on full context (#2565) commit 886f4eed7948f494e3da1d48d4f6f844e2f9a2c2 Author: Concedo <[email protected]> Date: Thu Aug 10 22:01:33 2023 +0800 updated lite, up ver, remove bell commit 1638757767072a4957f52b9e3594f0b67610631b Author: Martin Krasser <[email protected]> Date: Thu Aug 10 12:16:38 2023 +0200 Fix grammar-based sampling issue in server (#2566) commit c5f5209d37b09325377e36f39eab0b0f0c0d006e Author: Concedo <[email protected]> Date: Thu Aug 10 16:30:02 2023 +0800 globalize args commit f570b5cb1070591527a82d94bba408927b37778d Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 22:11:20 2023 -0500 Revert "revert cuda changes as they are bugggy" This reverts commit 1541bf879772aeeed8ff646bfc52185c2a88b79b. commit 1541bf879772aeeed8ff646bfc52185c2a88b79b Author: Concedo <[email protected]> Date: Wed Aug 9 22:36:41 2023 +0800 revert cuda changes as they are bugggy commit bacc20203efb1839aa313858a04d75255bb4b7f4 Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 20:37:17 2023 -0500 Merge remote-tracking branch 'upstream/concedo' commit b7cb4cfd109986bd66e8fd382d1e2516eaddfebb Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 20:00:52 2023 -0500 additional fixes commit fadae727baa3735ad3e0667384d6e05ca056b3ef Merge: 518eb2a 8f8ab6c Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 18:45:50 2023 -0500 Merge branch 'hipblas' into develop4Main commit 518eb2af9225f8300a108c4244c7eb0a2217c3bc Merge: bda0215 cae6a84 Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 18:32:10 2023 -0500 Merge remote-tracking branch 'upstream/concedo' into develop2Main commit bda0215b413bafc49890aa23fc35f96a191fb3e0 Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 18:17:54 2023 -0500 update makefile to multisystem path commit 8f8ab6c4c049df501e9a5ed8fef3aa0fc0691421 Author: YellowRoseCx <[email protected]> Date: Wed Aug 9 18:05:03 2023 -0500 hipLDFLAG Path change Unix to multisystem in Makefile changed the hardcoded linux distro hipblas LD path from -L/opt/rocm/lib to use the defined ROCM_PATH variable to be flexible with ROCm on non-Linux OS commit 610ba4cfc460ed65c4adc32d3365a216690384d5 Merge: 4024f91 25d43e0 Author: Henri Vasserman <[email protected]> Date: Wed Aug 9 23:54:58 2023 +0300 Merge 'origin/master' into hipblas commit 916a9acdd0a411426690400ebe2bb7ce840a6bba Author: Sam Spilsbury <[email protected]> Date: Wed Aug 9 23:47:42 2023 +0300 ggml-alloc: Don't try to re-use buffers of external tensors (#2562) * ggml-alloc: Don't try to re-use buffers of external tensors They might be weights that came from another context, so we have no control over them (and they might be re-used elsewhere so writing to them would be a bad idea). * ggml-alloc: >= when checking for out-of-bounds Co-authored-by: slaren <[email protected]> --------- Co-authored-by: slaren <[email protected]> commit ea04a4ca1940d92becc0ee26523aa2c4a18cf938 Author: grahameth <[email protected]> Date: Wed Aug 9 22:46:40 2023 +0200 add log_callback to llama_context_params for custom logging. (#2234) * add log_callback to llama_context_params for custom logging. * Fix macro expansion on gcc * Add struct llama_state for global variables and move log_callback there * Turn log level into enum and some minor changes. * Remove model_for_logging parameter (not needed anymore) * Convert remaining fprintf(stderr, ...) calls to use new macros. * Fix enum and initialize g_state * Fix log calls after merge * Fix missing static * Add back all the new lines in the logging strings * Add comment for llama_log_callback and replace remaining printf calls --------- Co-authored-by: grahameth <-> Co-authored-by: Helmut <[email protected]> commit a07e6dd3ad1a622f08c3187799879d4f1c49bad4 Author: Concedo <[email protected]> Date: Wed Aug 9 22:36:41 2023 +0800 revert cuda changes as they are bugggy commit f8376c7e610f68d07e079ff91f6988fb7a8399e2 Author: Concedo <[email protected]> Date: Wed Aug 9 21:23:33 2023 +0800 up ver, fixed compile (+1 squashed commits) Squashed commits: [ca51aa9e] up ver commit ba09f1c807956c59d8c64988626e95459f627ced Merge: 3a7853d 25d43e0 Author: Concedo <[email protected]> Date: Wed Aug 9 21:18:34 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # README.md # ggml-cuda.cu commit 3a7853d259c242d4977e9f4dc7627a799d5812b4 Author: Concedo <[email protected]> Date: Wed Aug 9 21:07:57 2023 +0800 handle stablecode-completion-alpha-3b commit 25d43e0eb578b6e73046d9d6644a3a14d460600d Author: Johannes Gäßler <[email protected]> Date: Wed Aug 9 09:42:34 2023 +0200 CUDA: tuned mul_mat_q kernels (#2546) commit 90058d96b0c6ab77802e153c23fad66d2f21a438 Author: Concedo <[email protected]> Date: Wed Aug 9 15:28:07 2023 +0800 sleep longer before exit commit 19cf2a8663938c424407544c13749f371104517b Author: Concedo <[email protected]> Date: Wed Aug 9 12:42:59 2023 +0800 add idle field and up ver commit 4b8a354895e078d3f0cafdf53430d72d3af8bb99 Author: Concedo <[email protected]> Date: Wed Aug 9 12:25:21 2023 +0800 cudatoolkit version commit 159ad9269d95bc07720c79debc23b5c466357b53 Author: Concedo <[email protected]> Date: Wed Aug 9 11:50:12 2023 +0800 up ver, set the cuda pool malloc lookahead back to 5% instead of 2% (+1 squashed commits) Squashed commits: [e0f65278] up ver, set the cuda pool malloc lookahead back to 5% instead of 2% commit 4024f91a665d83b6de8658d45ec9d004c5d90c79 Author: Henri Vasserman <[email protected]> Date: Wed Aug 9 01:56:44 2023 +0300 Add intrinsics polyfills for AMD --------- Co-authored-by: ardfork <[email protected]> Co-authored-by: funnbot <[email protected]> Co-authored-by: Engininja2 <[email protected]> commit ab6212864ce8e9af200bcedb3e0126ee49aa8d0a Merge: d91456a f5bfea0 Author: Henri Vasserman <[email protected]> Date: Wed Aug 9 00:37:01 2023 +0300 Merge 'origin/master' into hipblas commit 926d90fbabe836d16a5326eb99bdcb89ca0fc042 Merge: 793cfd1 f5bfea0 Author: Concedo <[email protected]> Date: Wed Aug 9 01:09:04 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # Makefile commit 793cfd136cc721884f79d09036b748e4f176cdb4 Author: Concedo <[email protected]> Date: Wed Aug 9 01:05:00 2023 +0800 fixed 70B detection again, try fix horde issues, fixed lite unicode issue, fixed cmake for cuda commit f5bfea0580e417f99850d5456ca541d871a3e48c Author: Martin Krasser <[email protected]> Date: Tue Aug 8 15:29:19 2023 +0200 Allow passing grammar to completion endpoint (#2532) * Allow passing grammar to completion endpoint commit acfc5478ff3446ca3b54553967a3dea09b7c771a Author: Johannes Gäßler <[email protected]> Date: Tue Aug 8 14:38:16 2023 +0200 CUDA: tighter VRAM scratch size for 65b/70b (#2551) commit 7ed8d1fe7f8cbe6a6763e6b46759795ac8d21e12 Author: chaihahaha <[email protected]> Date: Tue Aug 8 20:07:02 2023 +0800 llm.vim : multiline autocompletion, get rid of "^@" (#2543) commit e7f94d6fdc83b41ba449b4b8c80821673dd12ffc Author: Georgi Gerganov <[email protected]> Date: Tue Aug 8 15:05:30 2023 +0300 vim : bring back simple llm.vim example commit 2d7baaf50f3277e65cf71071f61ea34823d14c30 Author: AustinMroz <[email protected]> Date: Tue Aug 8 06:44:48 2023 -0500 vim : streaming and more (#2495) * Update Vim plugin * Remove getbufoneline usage, Add input bind example. getbufoneline() appears to be a recently added function and has been replaced with getbufline for compatibility. An additional example that explains how to add a keybind that works in insert mode was added. commit f3c3b4b1672d860800639c87d3b5d17564692469 Author: klosax <[email protected]> Date: Mon Aug 7 19:07:19 2023 +0200 Add --rope-scale parameter (#2544) * common.cpp : Add --rope-scale parameter * README.md : Add info about using linear rope scaling commit 3554080502cb050ccc3ae11d7a67df866ac3bd07 Author: Concedo <[email protected]> Date: Tue Aug 8 00:41:02 2023 +0800 fixed blasbatchmul multiplier commit 28ad80b6e4d38dde9e395fc5d4ebf19dc4aa4b66 Merge: 3c7d938 93356bd Author: Concedo <[email protected]> Date: Tue Aug 8 00:34:10 2023 +0800 Merge branch 'master' into concedo_experimental commit 3c7d938d95fd51780be37f10cdddb2f26a770adf Author: Concedo <[email protected]> Date: Tue Aug 8 00:32:51 2023 +0800 update lite, resize scratch buffers for blasbatch 2048 commit 93356bdb7a324a8f6570f99d02af392cd4c45796 Author: Georgi Gerganov <[email protected]> Date: Mon Aug 7 14:25:58 2023 +0300 ggml : mul mat tweaks (#2372) * ggml : mul mat wip ggml-ci * ggml : alternative thread distribution for mul_mat ggml-ci * ggml : mul_mat block tiling attempt * ggml : mul_mat threads yield ggml-ci commit 60baff7c8584ec369e53469cad5f92e102b1efe4 Author: Georgi Gerganov <[email protected]> Date: Mon Aug 7 14:24:42 2023 +0300 ggml : pad result of ggml_nbytes() commit 9082b5dfbfae01243a0b822dcd2812877e63bf1b Author: Georgi Gerganov <[email protected]> Date: Mon Aug 7 13:55:18 2023 +0300 ggml : change params pointer (style change) (#2539) ggml-ci commit 99d29c0094476c4962023036ecd61a3309d0e16b Author: Georgi Gerganov <[email protected]> Date: Mon Aug 7 13:20:09 2023 +0300 ggml : sync (custom ops) (#2537) ggml-ci commit 9133e456d2d52b05c6c7f92cd94a0d2564ddb2f7 Merge: cae6a84 3d9a551 Author: Concedo <[email protected]> Date: Mon Aug 7 17:33:42 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # Makefile # build.zig commit cae6a847ada88e415b0beda09d70d79b51762618 Author: Concedo <[email protected]> Date: Mon Aug 7 16:40:13 2023 +0800 cuda free only for non mmq (+2 squashed commit) Squashed commit: [3aca763a] only cuda free for non mmq [e69a8c9f] revert to pool alloc to try again commit 3d9a55181603e85a26378a850a14068034e5002d Author: Johannes Gäßler <[email protected]> Date: Mon Aug 7 10:09:40 2023 +0200 Fixed mmap prefetch for GPU offloading (#2529) commit f6f9896ac3d2ff207e18f87dab85d126ceef5236 Author: Georgi Gerganov <[email protected]> Date: Mon Aug 7 10:52:57 2023 +0300 metal : fix out-of-bounds access + inc concurrency nodes (#2416) * metal : fix out-of-bounds access + style changes * metal : increase concurrency nodes to 2*GGML_MAX_NODES commit 9f16a4c4efc5cca845e027c1dbad615612b9248c Author: Concedo <[email protected]> Date: Mon Aug 7 15:16:37 2023 +0800 switch to upstream implementation of pool malloc commit 34a14b28ff7f3c98730339bacee035091b2a812a Author: GiviMAD <[email protected]> Date: Sun Aug 6 23:21:46 2023 -0700 [Makefile] Move ARM CFLAGS before compilation (#2536) commit 7297128db8159c7b12db4c28a4532b993025c2e5 Author: Henri Vasserman <[email protected]> Date: Mon Aug 7 08:35:53 2023 +0300 [Zig] Rewrite build for Zig 0.11 (#2514) * zig build fixes * Disable LTO on Windows. commit 6659652c9fd1853dcb2d1882efc8f14b159d5d43 Author: Concedo <[email protected]> Date: Mon Aug 7 11:05:06 2023 +0800 lower actual temp used when temp=0 commit 0e41b94f40e1d10893d6ac29c727482573ef1652 Author: Concedo <[email protected]> Date: Mon Aug 7 10:43:06 2023 +0800 improve detection for 70B. commit fb44d72a78a81790d238ffd2453cf66d02eed688 Merge: 559c0e2 d9024df Author: Concedo <[email protected]> Date: Mon Aug 7 10:17:43 2023 +0800 Merge remote-tracking branch 'johannes/cuda-fix-mmap-prefetch' into concedo_experimental commit 559c0e2d1f621402d410944b5291da647243ab33 Author: Concedo <[email protected]> Date: Mon Aug 7 10:15:20 2023 +0800 updated lite again, fix for wi commit d9024df759b25d030fc8266d399c565fe7be9a04 Author: JohannesGaessler <[email protected]> Date: Sun Aug 6 10:18:05 2023 +0200 Fixed mmap prefetch for GPU offloading commit d442888626f11335e0c9e3b8555d2429b3262580 Merge: 198cc82 86c3219 Author: Concedo <[email protected]> Date: Sun Aug 6 22:47:33 2023 +0800 Merge branch 'master' into concedo_experimental # Conflicts: # Makefile commit 198cc826fcb9…

cebtenzzre · 2023-09-01T03:37:49Z

@cebtenzzre Can you share details of your tests (GPU, CUDA settings)? Thanks.

Sorry, I missed this. I have a Tesla P40 24GB. I was comparing commit bac6699 ("PR") with commit 519c981 ("before"). I used ehartford/dolphin-llama2-7b because that's what I had on hand at the time. I compiled with make LLAMA_CUBLAS=1 and benchmarked with this command, which runs main three times so the cache is warm:

{ for i in 0 1 2; do CUDA_VISIBLE_DEVICES=0 ./main -n 128 -m dolphin-llama2-7b.q2_k.gguf -ngl 100 -mmq --ignore-eos -t 1; done } |& tail

I then wrote down the "eval time" t/s.

I just re-quantized and re-tested just to make sure, and for Q2_K I get 55.67 t/s before this PR and 33.32 t/s after - a difference of less than 0.2% from the previously provided numbers.

Johannes has a few P40s, so he should be able to reproduce my results.

Kawrakow added 12 commits August 22, 2023 08:47

Minor 4-bit quantization improvement

77aea72

For the same model size as previus commit, we get PPL = 5.9069 vs 5.9138.

Some more fine tuning

ec9cb75

Adding make_qkx2_quants

4f8dcb1

With it, we get PPL = 5.8828 for L2-7B Q4_K_S.

Another minor improvement

e9f1340

Q2_K improvement

1c1f985

Smaller model, lower perplexity. 7B: file size = 2.632G, PPL = 6.3772 vs original 2.670G PPL = 6.8201 12B: file size = 5.056G, PPL = 5.4577 vs original 5.130G PPL = 5.7178 It is mostly Q3_K except for tok_embeddings, attention.wq, attention.wk, which are Q2_K

Iterating

404e43c

Revert Q5_K back to make_qkx1_quants

9f78d4c

Better Q6_K

e2af308

make_qkx2_quants is better for Q5_K after all

b706339

Fix after rebasing on master

35a0b97

Fix for changed tensor names

fdf73db

ggerganov approved these changes Aug 22, 2023

View reviewed changes

Green-Sky changed the title ~~Quantization imrovements for k_quants~~ Quantization improvements for k_quants Aug 22, 2023

ikawrakow merged commit bac6699 into master Aug 22, 2023
25 checks passed

ikawrakow deleted the ik/better_q234_k branch August 22, 2023 16:14

akawrykow mentioned this pull request Aug 23, 2023

Add date and commit hash to gguf metadata #2728

Open

KerfuffleV2 mentioned this pull request Aug 23, 2023

Minor: fix values shown in the quantize tool help #2735

Merged

ikawrakow mentioned this pull request Jan 11, 2024

Restore intended k-quants quantization mixes for MoE models #4872

Merged

cebtenzzre mentioned this pull request Feb 23, 2024

IQ3_S: a much better alternative to Q3_K #5676

Merged

Quantization improvements for k_quants #2707

Quantization improvements for k_quants #2707

Conversation

ikawrakow commented Aug 22, 2023

Green-Sky commented Aug 22, 2023

klosax commented Aug 22, 2023

ikawrakow commented Aug 22, 2023

ggerganov commented Aug 22, 2023

KerfuffleV2 commented Aug 22, 2023

IgnacioFDM commented Aug 22, 2023

TheBloke commented Aug 22, 2023 • edited Loading

ikawrakow commented Aug 22, 2023

ikawrakow commented Aug 22, 2023

IgnacioFDM commented Aug 22, 2023

cebtenzzre commented Aug 22, 2023

Green-Sky commented Aug 22, 2023

mirek190 commented Aug 22, 2023

cebtenzzre commented Aug 23, 2023

KerfuffleV2 commented Aug 23, 2023

ikawrakow commented Aug 23, 2023

TheBloke commented Aug 23, 2023

klosax commented Aug 23, 2023

ikawrakow commented Aug 23, 2023 • edited Loading

klosax commented Aug 23, 2023

KerfuffleV2 commented Aug 23, 2023

klosax commented Aug 23, 2023

TheBloke commented Aug 23, 2023 • edited Loading

KerfuffleV2 commented Aug 23, 2023

IgnacioFDM commented Aug 23, 2023

ikawrakow commented Aug 23, 2023

KerfuffleV2 commented Aug 23, 2023

ikawrakow commented Aug 23, 2023

TheBloke commented Aug 23, 2023

ikawrakow commented Aug 23, 2023

IgnacioFDM commented Aug 23, 2023

cebtenzzre commented Aug 23, 2023 • edited Loading

Dampfinchen commented Aug 23, 2023

ikawrakow commented Aug 24, 2023 • edited Loading

KerfuffleV2 commented Aug 24, 2023

ggerganov commented Aug 24, 2023

KerfuffleV2 commented Aug 24, 2023

ikawrakow commented Aug 24, 2023

cebtenzzre commented Sep 1, 2023 • edited Loading

TheBloke commented Aug 22, 2023 •

edited

Loading

ikawrakow commented Aug 23, 2023 •

edited

Loading

TheBloke commented Aug 23, 2023 •

edited

Loading

cebtenzzre commented Aug 23, 2023 •

edited

Loading

ikawrakow commented Aug 24, 2023 •

edited

Loading

cebtenzzre commented Sep 1, 2023 •

edited

Loading