HUGGING_FACE_HUB_TOKEN -> HF_TOKEN

README.md

@@ -264,7 +264,7 @@ Note: the `composer` command used above to train the model refers to the [Compos
 If you have a write-enabled [HuggingFace auth token](https://huggingface.co/docs/hub/security-tokens), you can optionally upload your model to the Hub! Just export your token like this:
 
 ```bash
-export HUGGING_FACE_HUB_TOKEN=your-auth-token
+export HF_TOKEN=your-auth-token
 ```
 
 and uncomment the line containing `--hf_repo_for_upload ...` in the above call to `inference/convert_composer_to_hf.py`.

llmfoundry/utils/model_download_utils.py

@@ -75,7 +75,7 @@ def download_from_hf_hub(
             available. Defaults to True.
         tokenizer_only (bool): If true, only download tokenizer files.
         token (str, optional): The HuggingFace API token. If not provided, the token will be read from the
-            `HUGGING_FACE_HUB_TOKEN` environment variable.
+            `HF_TOKEN` environment variable.
 
     Raises:
         RepositoryNotFoundError: If the model repo doesn't exist or the token is unauthorized.

mcli/mcli-llama2-finetune.yaml

@@ -36,7 +36,7 @@ parameters:
     init_device: mixed
     pretrained_model_name_or_path: meta-llama/Llama-2-7b-hf
     pretrained: true
-    # Note: you must have set the HUGGING_FACE_HUB_TOKEN environment variable and have access to the llama2 models
+    # Note: you must have set the HF_TOKEN environment variable and have access to the llama2 models
     use_auth_token: true
     use_flash_attention_2: true
 

scripts/inference/hf_chat.py

@@ -364,9 +364,7 @@ def main(args: Namespace) -> None:
     except Exception as e:
         raise RuntimeError(
             'If you are having auth problems, try logging in via `huggingface-cli login` '
-            +
-            'or by setting the environment variable `export HUGGING_FACE_HUB_TOKEN=... '
-            +
+            + 'or by setting the environment variable `export HF_TOKEN=... ' +
             'using your access token from https://huggingface.co/settings/tokens.',
         ) from e
 
@@ -389,9 +387,7 @@ def main(args: Namespace) -> None:
         raise RuntimeError(
             'Unable to load HF model. ' +
             'If you are having auth problems, try logging in via `huggingface-cli login` '
-            +
-            'or by setting the environment variable `export HUGGING_FACE_HUB_TOKEN=... '
-            +
+            + 'or by setting the environment variable `export HF_TOKEN=... ' +
             'using your access token from https://huggingface.co/settings/tokens.',
         ) from e
 

scripts/inference/hf_generate.py

@@ -200,7 +200,7 @@ def main(args: Namespace) -> None:
     except Exception as e:
         raise RuntimeError(
             'If you are having auth problems, try logging in via `huggingface-cli login` ' +\
-            'or by setting the environment variable `export HUGGING_FACE_HUB_TOKEN=... ' +\
+            'or by setting the environment variable `export HF_TOKEN=... ' +\
             'using your access token from https://huggingface.co/settings/tokens.',
         ) from e
 
@@ -236,9 +236,7 @@ def main(args: Namespace) -> None:
         raise RuntimeError(
             'Unable to load HF model. ' +
             'If you are having auth problems, try logging in via `huggingface-cli login` '
-            +
-            'or by setting the environment variable `export HUGGING_FACE_HUB_TOKEN=... '
-            +
+            + 'or by setting the environment variable `export HF_TOKEN=... ' +
             'using your access token from https://huggingface.co/settings/tokens.',
         ) from e
 

scripts/misc/download_model.py

@@ -27,7 +27,8 @@
     download_from_oras,
 )
 
-HF_TOKEN_ENV_VAR = 'HUGGING_FACE_HUB_TOKEN'
+DEPRECATED_HF_TOKEN_ENV_VAR = 'HUGGING_FACE_HUB_TOKEN'
+HF_TOKEN_ENV_VAR = 'HF_TOKEN'
 
 logging.basicConfig(
     format=f'%(asctime)s: %(levelname)s: %(name)s: %(message)s',
@@ -42,7 +43,10 @@ def add_hf_parser_arguments(parser: argparse.ArgumentParser) -> None:
     parser.add_argument(
         '--token',
         type=str,
-        default=os.getenv(HF_TOKEN_ENV_VAR),
+        default=os.getenv(
+            HF_TOKEN_ENV_VAR,
+            os.getenv(DEPRECATED_HF_TOKEN_ENV_VAR),
+        ),
     )
 
 

tests/a_scripts/inference/test_convert_composer_to_hf.py

@@ -533,7 +533,7 @@ def _get_model_and_tokenizer(
         tokenizer_name = 'EleutherAI/gpt-neo-125M'
     elif model == 'llama2':
         assert tie_word_embeddings is None
-        if 'HUGGING_FACE_HUB_TOKEN' not in os.environ:
+        if 'HF_TOKEN' not in os.environ:
             pytest.skip(
                 'The CI cluster does not have access to the Llama models, so skip this test.',
             )
@@ -985,7 +985,7 @@ def test_convert_and_generate(
         om_cfg['model']['config_overrides']['hidden_size'] = 36
     elif model == 'llama2':
         assert tie_word_embeddings is None
-        if 'HUGGING_FACE_HUB_TOKEN' not in os.environ:
+        if 'HF_TOKEN' not in os.environ:
             pytest.skip(
                 'The CI cluster does not have access to the Llama models, so skip this test.',
             )

tests/models/hf/test_hf_config.py

@@ -172,7 +172,7 @@ def test_hf_config_override(
 
 
 @pytest.mark.skipif(
-    'HUGGING_FACE_HUB_TOKEN' not in os.environ,
+    'HF_TOKEN' not in os.environ,
     reason='CI does not have access to llama2',
 )
 def test_rope_scaling_override():
@@ -205,7 +205,7 @@ def test_rope_scaling_override():
 
 
 @pytest.mark.skipif(
-    'HUGGING_FACE_HUB_TOKEN' not in os.environ,
+    'HF_TOKEN' not in os.environ,
     reason='CI does not have access to Dbrx',
 )
 def test_nested_override():