Documentation: Custom models #2463
Merged
Conversation
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
yifanmai
requested review from
percyliang,
farzaank and
JosselinSomervilleRoberts
percyliang
reviewed
Mar 14, 2024
docs/custom_models.md
Outdated
|
|
||
| To create a local model deployments configuration file, determine the location of your local configuration folder. If you are using the `--local-path` flag with `helm-run`, the specified folder of the flag is the local configuration folder. Otherwise, the local configuration folder is `./prod_env/` under your current working directory by default. | ||
|
|
||
| Create a file called `model_deployments.yaml` underneath that directory (e.g. `./prod_env/model_deployments.yaml`). |
There was a problem hiding this comment.
Maybe call this custom_model_deployments.yaml or something, since model_deployements.yaml is already a filename, just to avoid a clash.
There was a problem hiding this comment.
This requires a code change, so I'll defer it.
percyliang
reviewed
Mar 14, 2024
| # Custom Models | ||
|
|
||
| HELM comes with more than a hundred built-in models. If you want to run a HELM evaluation on a model that is not built-in, you can configure HELM to add your own model. This also allows you to evaluate private models that not publicly accessible, such as a model checkpoint on local disk, or a model server on a private network | ||
|
|
There was a problem hiding this comment.
Give a concrete example where you would need to override the class.
percyliang
reviewed
Mar 14, 2024
docs/custom_models.md
Outdated
|
|
||
| ## Creating a model deployments configuration | ||
|
|
||
| To create a local model deployments configuration file, determine the location of your local configuration folder. If you are using the `--local-path` flag with `helm-run`, the specified folder of the flag is the local configuration folder. Otherwise, the local configuration folder is `./prod_env/` under your current working directory by default. |
There was a problem hiding this comment.
Maybe give some motivation for why create model deployment and what they are.
There was a problem hiding this comment.
Added definition of deployment.
docs/custom_models.md
Outdated
| @@ -0,0 +1,104 @@ | |||
| # Custom Models | |||
|
|
|||
| HELM comes with more than a hundred built-in models. If you want to run a HELM evaluation on a model that is not built-in, you can configure HELM to add your own model. This also allows you to evaluate private models that not publicly accessible, such as a model checkpoint on local disk, or a model server on a private network | |||
There was a problem hiding this comment.
that not publicly -> that are not publicly
There was a problem hiding this comment.
Good catch, fixed.
docs/custom_models.md
Outdated
|
|
||
| If you wish to evaluate a model not covered by an existing `Client` and `Tokenizer`, you can implement your own `Client` and `Tokenizer` subclasses. Instructions for adding custom `Client` and `Tokenizer` subclasses will be added to the documentation in the future. | ||
|
|
||
| ## Creating a model deployments configuration |
There was a problem hiding this comment.
deployments should be singular here?
docs/custom_models.md
Outdated
|
|
||
| If you wish to evaluate a model not covered by an existing `Client` and `Tokenizer`, you can implement your own `Client` and `Tokenizer` subclasses. Instructions for adding custom `Client` and `Tokenizer` subclasses will be added to the documentation in the future. | ||
|
|
||
| ## Creating a model deployments configuration |
There was a problem hiding this comment.
Is it clear what a deployment is?
There was a problem hiding this comment.
Added definition of deployment.
docs/custom_models.md
Outdated
|
|
||
| Note: If `together_model` is omitted, the Together model with `model_name` (_not_ `name`) will be used by default. | ||
|
|
||
| Note: This model may not be currently available on Together AI. Consult [Together AI's Inference Models documentation](https://docs.together.ai/docs/inference-models) for a list of currently available models and corresponding model strings. |
There was a problem hiding this comment.
Add a note to remember to register Together credentials?
farzaank
reviewed
Mar 19, 2024
docs/custom_models.md
Outdated
| helm-run --run-entries mmlu:subject=anatomy,model=eleutherai/pythia-70m --suite my_suite --max-eval-instances 5 | ||
| ``` | ||
|
|
||
| Note: This uses Hugging Face local inference. It will attempt to use GPU inference if available, and use CPU inference otherwise. It is only able to use the first GPU. Multi-GPU inference is not supported. Every model needed by `helm-run` will be loaded on the same GPU - if evaluating multiple models, it is prudent to evaluate each model with a separate `helm-run` invocation. |
There was a problem hiding this comment.
"Hugging Face" -> "Hugging Face's" or "the Hugging Face"
There was a problem hiding this comment.
Fixed: "local inference with Hugging Face".
yifanmai
force-pushed
the
yifanmai/docs-adding-models
branch
from
667860a
to
e39cea9
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This will eventually replace Adding new models because it documents the new way of adding models that does not require forking HELM.
Fixes #2379