-
Notifications
You must be signed in to change notification settings - Fork 168
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
🦙 Llama2 optimization with multi-ep managed env (#1087)
## Describe your changes 1. Llama2 optimization with multi-ep 2. Add info-level log to tell the system creation time. 3. Use device and ep but not only device to determine the default value of `use_gpu`. ## Checklist before requesting a review - [ ] Add unit tests for this change. - [ ] Make sure all tests can pass. - [ ] Update documents if necessary. - [ ] Lint and apply fixes to your code by running `lintrunner -a` - [ ] Is this a user-facing change? If yes, give a description of this change to be included in the release notes. - [ ] Is this PR including examples changes? If yes, please remember to update [example documentation](https://github.com/microsoft/Olive/blob/main/docs/source/examples.md) in a follow-up PR. ## (Optional) Issue link
- Loading branch information
Showing
8 changed files
with
1,497 additions
and
1 deletion.
There are no files selected for viewing
172 changes: 172 additions & 0 deletions
172
examples/llama2/notebook/llama2_multiep/config_cpu.json
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,172 @@ | ||
| { | ||
| "input_model": { | ||
| "type": "PyTorchModel", | ||
| "config": { | ||
| "model_script": "user_script.py", | ||
| "io_config": "get_merged_decoder_with_past_io_config", | ||
| "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs", | ||
| "hf_config": { | ||
| "model_name": "meta-llama/Llama-2-7b-hf", | ||
| "model_class": "LlamaForCausalLM", | ||
| "from_pretrained_args": { | ||
| "_attn_implementation": "eager" | ||
| }, | ||
| "task": "text-generation" | ||
| } | ||
| } | ||
| }, | ||
| "data_configs": { | ||
| "wikitext2_train": { | ||
| "name": "wikitext2_train", | ||
| "type": "HuggingfaceContainer", | ||
| "params_config": { | ||
| "data_name": "wikitext", | ||
| "subset": "wikitext-2-raw-v1", | ||
| "split": "train", | ||
| "component_kwargs": { | ||
| "pre_process_data": { | ||
| "text_cols": [ | ||
| "text" | ||
| ], | ||
| "corpus_strategy": "join", | ||
| "add_special_tokens": false, | ||
| "source_max_len": 2048, | ||
| "max_samples": 128, | ||
| "joiner": "\n\n" | ||
| } | ||
| } | ||
| } | ||
| } | ||
| }, | ||
| "systems": { | ||
| "local_system": { | ||
| "type": "LocalSystem", | ||
| "config": { | ||
| "accelerators": [ | ||
| { | ||
| "device": "CPU", | ||
| "execution_providers": [ | ||
| "CPUExecutionProvider" | ||
| ] | ||
| } | ||
| ] | ||
| } | ||
| } | ||
| }, | ||
| "evaluators": { | ||
| "merged_evaluator": { | ||
| "metrics": [ | ||
| { | ||
| "name": "latency_prompt_processing", | ||
| "type": "latency", | ||
| "sub_types": [ | ||
| { | ||
| "name": "avg", | ||
| "priority": 1 | ||
| } | ||
| ], | ||
| "user_config": { | ||
| "user_script": "user_script.py", | ||
| "dataloader_func": "dataloader_func_for_merged", | ||
| "func_kwargs": { | ||
| "dataloader_func": { | ||
| "model_id": "meta-llama/Llama-2-7b-hf", | ||
| "past_seq_length": 0, | ||
| "seq_length": 8, | ||
| "max_seq_length": 2048 | ||
| } | ||
| }, | ||
| "batch_size": 2, | ||
| "io_bind": true | ||
| } | ||
| }, | ||
| { | ||
| "name": "latency_token_generation", | ||
| "type": "latency", | ||
| "sub_types": [ | ||
| { | ||
| "name": "avg" | ||
| } | ||
| ], | ||
| "user_config": { | ||
| "user_script": "user_script.py", | ||
| "dataloader_func": "dataloader_func_for_merged", | ||
| "func_kwargs": { | ||
| "dataloader_func": { | ||
| "model_id": "meta-llama/Llama-2-7b-hf", | ||
| "past_seq_length": 8, | ||
| "seq_length": 1, | ||
| "max_seq_length": 2048 | ||
| } | ||
| }, | ||
| "batch_size": 2, | ||
| "io_bind": true | ||
| } | ||
| } | ||
| ] | ||
| } | ||
| }, | ||
| "passes": { | ||
| "conversion_merged": { | ||
| "type": "OnnxConversion", | ||
| "config": { | ||
| "target_opset": 17, | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true | ||
| } | ||
| }, | ||
| "transformers_optimization_fp32": { | ||
| "type": "OrtTransformersOptimization", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "model_type": "gpt2", | ||
| "opt_level": 0, | ||
| "only_onnxruntime": false, | ||
| "keep_io_types": false, | ||
| "float16": false, | ||
| "use_gqa": false | ||
| } | ||
| }, | ||
| "transformers_optimization_fp16": { | ||
| "type": "OrtTransformersOptimization", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "model_type": "gpt2", | ||
| "opt_level": 0, | ||
| "only_onnxruntime": false, | ||
| "keep_io_types": false, | ||
| "float16": true, | ||
| "use_gqa": true | ||
| } | ||
| }, | ||
| "blockwise_quant_int4": { | ||
| "type": "OnnxMatMul4Quantizer", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "block_size": 32, | ||
| "is_symmetric": true | ||
| } | ||
| } | ||
| }, | ||
| "engine": { | ||
| "evaluator": "merged_evaluator", | ||
| "host": "local_system", | ||
| "target": "local_system", | ||
| "cache_dir": "cache", | ||
| "output_dir": "models/llama2_cpu/meta-llama/Llama-2-7b-hf" | ||
| }, | ||
| "pass_flows": [ | ||
| [ | ||
| "conversion_merged", | ||
| "transformers_optimization_fp32", | ||
| "blockwise_quant_int4" | ||
| ], | ||
| [ | ||
| "conversion_merged", | ||
| "transformers_optimization_fp16" | ||
| ] | ||
| ] | ||
| } |
172 changes: 172 additions & 0 deletions
172
examples/llama2/notebook/llama2_multiep/config_gpu.json
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,172 @@ | ||
| { | ||
| "input_model": { | ||
| "type": "PyTorchModel", | ||
| "config": { | ||
| "model_script": "user_script.py", | ||
| "io_config": "get_merged_decoder_with_past_io_config", | ||
| "dummy_inputs_func": "get_merged_decoder_with_past_dummy_inputs", | ||
| "hf_config": { | ||
| "model_name": "meta-llama/Llama-2-7b-hf", | ||
| "model_class": "LlamaForCausalLM", | ||
| "from_pretrained_args": { | ||
| "_attn_implementation": "eager" | ||
| }, | ||
| "task": "text-generation" | ||
| } | ||
| } | ||
| }, | ||
| "data_configs": { | ||
| "wikitext2_train": { | ||
| "name": "wikitext2_train", | ||
| "type": "HuggingfaceContainer", | ||
| "params_config": { | ||
| "data_name": "wikitext", | ||
| "subset": "wikitext-2-raw-v1", | ||
| "split": "train", | ||
| "component_kwargs": { | ||
| "pre_process_data": { | ||
| "text_cols": [ | ||
| "text" | ||
| ], | ||
| "corpus_strategy": "join", | ||
| "add_special_tokens": false, | ||
| "source_max_len": 2048, | ||
| "max_samples": 128, | ||
| "joiner": "\n\n" | ||
| } | ||
| } | ||
| } | ||
| } | ||
| }, | ||
| "systems": { | ||
| "local_system": { | ||
| "type": "LocalSystem", | ||
| "config": { | ||
| "accelerators": [ | ||
| { | ||
| "device": "GPU", | ||
| "execution_providers": [ | ||
| "CUDAExecutionProvider" | ||
| ] | ||
| } | ||
| ] | ||
| } | ||
| } | ||
| }, | ||
| "evaluators": { | ||
| "merged_evaluator": { | ||
| "metrics": [ | ||
| { | ||
| "name": "latency_prompt_processing", | ||
| "type": "latency", | ||
| "sub_types": [ | ||
| { | ||
| "name": "avg", | ||
| "priority": 1 | ||
| } | ||
| ], | ||
| "user_config": { | ||
| "user_script": "user_script.py", | ||
| "dataloader_func": "dataloader_func_for_merged", | ||
| "func_kwargs": { | ||
| "dataloader_func": { | ||
| "model_id": "meta-llama/Llama-2-7b-hf", | ||
| "past_seq_length": 0, | ||
| "seq_length": 8, | ||
| "max_seq_length": 2048 | ||
| } | ||
| }, | ||
| "batch_size": 2, | ||
| "io_bind": true | ||
| } | ||
| }, | ||
| { | ||
| "name": "latency_token_generation", | ||
| "type": "latency", | ||
| "sub_types": [ | ||
| { | ||
| "name": "avg" | ||
| } | ||
| ], | ||
| "user_config": { | ||
| "user_script": "user_script.py", | ||
| "dataloader_func": "dataloader_func_for_merged", | ||
| "func_kwargs": { | ||
| "dataloader_func": { | ||
| "model_id": "meta-llama/Llama-2-7b-hf", | ||
| "past_seq_length": 8, | ||
| "seq_length": 1, | ||
| "max_seq_length": 2048 | ||
| } | ||
| }, | ||
| "batch_size": 2, | ||
| "io_bind": true | ||
| } | ||
| } | ||
| ] | ||
| } | ||
| }, | ||
| "passes": { | ||
| "conversion_merged": { | ||
| "type": "OnnxConversion", | ||
| "config": { | ||
| "target_opset": 17, | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true | ||
| } | ||
| }, | ||
| "transformers_optimization_fp32": { | ||
| "type": "OrtTransformersOptimization", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "model_type": "gpt2", | ||
| "opt_level": 0, | ||
| "only_onnxruntime": false, | ||
| "keep_io_types": false, | ||
| "float16": false, | ||
| "use_gqa": false | ||
| } | ||
| }, | ||
| "transformers_optimization_fp16": { | ||
| "type": "OrtTransformersOptimization", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "model_type": "gpt2", | ||
| "opt_level": 0, | ||
| "only_onnxruntime": false, | ||
| "keep_io_types": false, | ||
| "float16": true, | ||
| "use_gqa": true | ||
| } | ||
| }, | ||
| "blockwise_quant_int4": { | ||
| "type": "OnnxMatMul4Quantizer", | ||
| "config": { | ||
| "save_as_external_data": true, | ||
| "all_tensors_to_one_file": true, | ||
| "block_size": 32, | ||
| "is_symmetric": true | ||
| } | ||
| } | ||
| }, | ||
| "engine": { | ||
| "evaluator": "merged_evaluator", | ||
| "host": "local_system", | ||
| "target": "local_system", | ||
| "cache_dir": "cache", | ||
| "output_dir": "models/llama2_gpu/meta-llama/Llama-2-7b-hf" | ||
| }, | ||
| "pass_flows": [ | ||
| [ | ||
| "conversion_merged", | ||
| "transformers_optimization_fp32", | ||
| "blockwise_quant_int4" | ||
| ], | ||
| [ | ||
| "conversion_merged", | ||
| "transformers_optimization_fp16" | ||
| ] | ||
| ] | ||
| } |
Oops, something went wrong.