Update docs

allenai · Jun 22, 2024 · 9a40a83 · 9a40a83
1 parent 0fe8008
commit 9a40a83
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 1 deletion.
diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -24,6 +24,7 @@ shared memory implementation can be used by passing `use_legacy_shared_mem_impl`
 ### Fixed
 
 - Changed from `ignored_index` to `ignore_index` for `cross_entropy_loss` when `flash-attn>=2.5.8`.
+- Make `hf_olmo` support `AutoModelForCasualLM` and similar HF methods again.
 
 ## [v0.3.0](https://github.com/allenai/OLMo/releases/tag/v0.3.0) - 2024-04-25
 

diff --git a/docs/Checkpoints.md b/docs/Checkpoints.md
@@ -51,7 +51,9 @@ python scripts/convert_olmo_to_hf_new.py --input_dir /path/to/olmo/checkpoint --
 HF OLMo checkpoints
 ---
 
-These checkpoints can be used with the Transformers-style OLMo implementation in the `hf_olmo` package. This implementation has only partial support for Transformers functionality. In particular, using "auto" methods like `AutoModelForCausalLM` with these checkpoints is not supported. Consequently, we recommend using Transformers checkpoints over these if available. These checkpoints cannot be used with the pretraining/fine-tuning script provided in this repo.
+These checkpoints can be used with the Transformers-style OLMo implementation in the `hf_olmo` package. This implementation has only partial support for Transformers functionality. Consequently, we recommend using Transformers checkpoints over these if available.
+"auto" methods like `AutoModelForCausalLM` are now supported for these checkpoints.
+These checkpoints cannot be used with the pretraining/fine-tuning script provided in this repo.
 
 The following checkpoints on HF Hub are HF OLMo checkpoints:
 - [OLMo-1.7-7B](https://huggingface.co/allenai/OLMo-1.7-7B)