Improve Mistral format checks. (#33253)

Signed-off-by: Julien Denize <julien.denize@mistral.ai>
Signed-off-by: juliendenize <julien.denize@mistral.ai>
Co-authored-by: Michael Goin <mgoin64@gmail.com>

tests/config/test_model_arch_config.py

@@ -83,7 +83,10 @@ def _assert_model_arch_config(
     assert model_arch_config.is_deepseek_mla == expected["is_deepseek_mla"]
 
     torch_dtype = ModelArchConfigConvertorBase.get_torch_dtype(
-        model_config.hf_config, model_config.model, revision=model_config.revision
+        model_config.hf_config,
+        model_config.model,
+        revision=model_config.revision,
+        config_format="hf",
     )
     assert str(torch_dtype) == expected["dtype"]
 

tests/conftest.py

@@ -365,6 +365,7 @@ class HfRunner:
             self.config,
             dtype=dtype,
             is_pooling_model=is_sentence_transformer or is_cross_encoder,
+            config_format="hf",
         )
 
         model_kwargs = model_kwargs if model_kwargs is not None else {}

tests/transformers_utils/test_repo_utils.py

@@ -8,7 +8,11 @@ from unittest.mock import MagicMock, call, patch
 
 import pytest
 
-from vllm.transformers_utils.repo_utils import list_filtered_repo_files
+from vllm.transformers_utils.repo_utils import (
+    any_pattern_in_repo_files,
+    is_mistral_model_repo,
+    list_filtered_repo_files,
+)
 
 
 @pytest.mark.parametrize(
@@ -60,3 +64,95 @@ def test_list_filtered_repo_files(
             repo_type="model",
             token="token",
         )
+
+
+@pytest.mark.parametrize(
+    ("allow_patterns", "expected_bool"),
+    [
+        (["*.json", "correct*.txt"], True),
+        (
+            ["*.jpeg"],
+            True,
+        ),
+        (
+            ["not_found.jpeg"],
+            False,
+        ),
+    ],
+)
+def test_one_filtered_repo_files(allow_patterns: list[str], expected_bool: bool):
+    with tempfile.TemporaryDirectory() as tmp_dir:
+        # Prep folder and files
+        path_tmp_dir = Path(tmp_dir)
+        subfolder = path_tmp_dir / "subfolder"
+        subfolder.mkdir()
+        (path_tmp_dir / "uncorrect.jpeg").touch()
+        (subfolder / "correct.txt").touch()
+
+        def _glob_path() -> list[str]:
+            return [
+                str(file.relative_to(path_tmp_dir))
+                for file in path_tmp_dir.glob("**/*")
+                if file.is_file()
+            ]
+
+        # Patch list_repo_files called by fn
+        with patch(
+            "vllm.transformers_utils.repo_utils.list_repo_files",
+            MagicMock(return_value=_glob_path()),
+        ) as mock_list_repo_files:
+            assert (
+                any_pattern_in_repo_files(
+                    tmp_dir, allow_patterns, "revision", "model", "token"
+                )
+            ) is expected_bool
+        assert mock_list_repo_files.call_count == 1
+        assert mock_list_repo_files.call_args_list[0] == call(
+            repo_id=tmp_dir,
+            revision="revision",
+            repo_type="model",
+            token="token",
+        )
+
+
+@pytest.mark.parametrize(
+    ("files", "expected_bool"),
+    [
+        (["consolidated.safetensors", "incorrect.txt"], True),
+        (["consolidated-1.safetensors", "incorrect.txt"], True),
+        (
+            ["consolidated-1.json"],
+            False,
+        ),
+    ],
+)
+def test_is_mistral_model_repo(files: list[str], expected_bool: bool):
+    with tempfile.TemporaryDirectory() as tmp_dir:
+        # Prep folder and files
+        path_tmp_dir = Path(tmp_dir)
+        for file in files:
+            (path_tmp_dir / file).touch()
+
+        def _glob_path() -> list[str]:
+            return [
+                str(file.relative_to(path_tmp_dir))
+                for file in path_tmp_dir.glob("**/*")
+                if file.is_file()
+            ]
+
+        # Patch list_repo_files called by fn
+        with patch(
+            "vllm.transformers_utils.repo_utils.list_repo_files",
+            MagicMock(return_value=_glob_path()),
+        ) as mock_list_repo_files:
+            assert (
+                is_mistral_model_repo(tmp_dir, "revision", "model", "token")
+                is expected_bool
+            )
+        assert mock_list_repo_files.call_count == 1
+        assert mock_list_repo_files.call_args_list[0] == call(
+            repo_id=tmp_dir,
+            revision="revision",
+            repo_type="model",
+            token="token",
+        )

vllm/config/model.py

@@ -565,6 +565,7 @@ class ModelConfig:
             self.dtype,
             is_pooling_model=self.runner_type == "pooling",
             revision=self.revision,
+            config_format=self.config_format,
         )
 
         self.original_max_model_len = self.max_model_len
@@ -1844,9 +1845,10 @@ def _get_and_verify_dtype(
     *,
     is_pooling_model: bool,
     revision: str | None = None,
+    config_format: ConfigFormat = "hf",
 ) -> torch.dtype:
     config_dtype = ModelArchConfigConvertorBase.get_torch_dtype(
-        config, model_id, revision=revision
+        config, model_id, revision=revision, config_format=config_format
     )
     model_type = config.model_type
 

vllm/tokenizers/registry.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
-import importlib.util
 from dataclasses import dataclass, field
 from functools import lru_cache
 from pathlib import Path
@@ -18,7 +17,10 @@ from vllm.transformers_utils.gguf_utils import (
     is_remote_gguf,
     split_remote_gguf,
 )
-from vllm.transformers_utils.repo_utils import list_filtered_repo_files
+from vllm.transformers_utils.repo_utils import (
+    any_pattern_in_repo_files,
+    is_mistral_model_repo,
+)
 from vllm.utils.import_utils import resolve_obj_by_qualname
 
 from .protocol import TokenizerLike
@@ -142,25 +144,25 @@ def resolve_tokenizer_args(
         kwargs["use_fast"] = False
 
     # Try to use official Mistral tokenizer if possible
-    if tokenizer_mode == "auto" and importlib.util.find_spec("mistral_common"):
-        allow_patterns = ["tekken.json", "tokenizer.model.v*"]
-        files_list = list_filtered_repo_files(
+    if (
+        tokenizer_mode == "auto"
+        and is_mistral_model_repo(
+            model_name_or_path=str(tokenizer_name), revision=revision
+        )
+        and any_pattern_in_repo_files(
             model_name_or_path=str(tokenizer_name),
-            allow_patterns=allow_patterns,
+            allow_patterns=["tekken.json", "tokenizer.model.v*"],
             revision=revision,
         )
-        if len(files_list) > 0:
+    ):
         tokenizer_mode = "mistral"
 
     # Try to use Grok2 tiktoken tokenizer if possible
-    if tokenizer_mode == "auto":
-        allow_patterns = ["tokenizer.tok.json"]
-        files_list = list_filtered_repo_files(
+    if tokenizer_mode == "auto" and any_pattern_in_repo_files(
         model_name_or_path=str(tokenizer_name),
-            allow_patterns=allow_patterns,
+        allow_patterns=["tokenizer.tok.json"],
         revision=revision,
-        )
-        if len(files_list) > 0:
+    ):
         tokenizer_mode = "grok2"
 
     # Fallback to HF tokenizer

vllm/transformers_utils/config.py

@@ -23,6 +23,7 @@ from transformers.utils import CONFIG_NAME as HF_CONFIG_NAME
 
 from vllm import envs
 from vllm.logger import init_logger
+from vllm.transformers_utils.repo_utils import is_mistral_model_repo
 from vllm.transformers_utils.utils import parse_safetensors_file_metadata
 
 from .config_parser_base import ConfigParserBase
@@ -49,7 +50,6 @@ except ImportError:
         ALLOWED_LAYER_TYPES as ALLOWED_ATTENTION_LAYER_TYPES,
     )
 
-
 if envs.VLLM_USE_MODELSCOPE:
     from modelscope import AutoConfig
 else:
@@ -581,7 +581,11 @@ def get_config(
         try:
             # First check for Mistral to avoid defaulting to
             # Transformers implementation.
-            if file_or_path_exists(model, MISTRAL_CONFIG_NAME, revision=revision):
+            if is_mistral_model_repo(
+                model_name_or_path=str(model), revision=revision
+            ) and file_or_path_exists(
+                model=model, config_name=MISTRAL_CONFIG_NAME, revision=revision
+            ):
                 config_format = "mistral"
             elif (_is_gguf and not _is_remote_gguf) or file_or_path_exists(
                 model, HF_CONFIG_NAME, revision=revision

vllm/transformers_utils/model_arch_config_convertor.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
+from collections.abc import Iterator
+from contextlib import contextmanager
 from typing import final
 
 import torch
+from huggingface_hub import constants
 from safetensors.torch import _TYPES as _SAFETENSORS_TO_TORCH_DTYPE
 from transformers import PretrainedConfig
 
@@ -14,6 +17,7 @@ from vllm.config.model_arch import (
 from vllm.config.utils import getattr_iter
 from vllm.logger import init_logger
 from vllm.transformers_utils.config import (
+    ConfigFormat,
     try_get_safetensors_metadata,
 )
 from vllm.utils.torch_utils import common_broadcastable_dtype
@@ -21,6 +25,22 @@ from vllm.utils.torch_utils import common_broadcastable_dtype
 logger = init_logger(__name__)
 
 
+@contextmanager
+def _maybe_patch_hf_hub_constants(config_format: ConfigFormat) -> Iterator[None]:
+    if config_format == "mistral":
+        hf_safetensors_single_file = constants.SAFETENSORS_SINGLE_FILE
+        hf_safetensors_index_file = constants.SAFETENSORS_INDEX_FILE
+        constants.SAFETENSORS_SINGLE_FILE = "consolidated.safetensors"
+        constants.SAFETENSORS_INDEX_FILE = "consolidated.safetensors.index.json"
+        try:
+            yield
+        finally:
+            constants.SAFETENSORS_SINGLE_FILE = hf_safetensors_single_file
+            constants.SAFETENSORS_INDEX_FILE = hf_safetensors_index_file
+    else:
+        yield
+
+
 class ModelArchConfigConvertorBase:
     def __init__(self, hf_config: PretrainedConfig, hf_text_config: PretrainedConfig):
         self.hf_config = hf_config
@@ -123,7 +143,11 @@ class ModelArchConfigConvertorBase:
     @final
     @classmethod
     def get_torch_dtype(
-        cls, hf_config: PretrainedConfig, model_id: str, revision: str | None
+        cls,
+        hf_config: PretrainedConfig,
+        model_id: str,
+        revision: str | None,
+        config_format: ConfigFormat,
     ):
         # NOTE: getattr(config, "dtype", torch.float32) is not correct
         # because config.dtype can be None.
@@ -140,6 +164,7 @@ class ModelArchConfigConvertorBase:
 
         # Try to read the dtype of the weights if they are in safetensors format
         if config_dtype is None:
+            with _maybe_patch_hf_hub_constants(config_format):
                 repo_mt = try_get_safetensors_metadata(model_id, revision=revision)
 
             if repo_mt and (files_mt := repo_mt.files_metadata):

vllm/transformers_utils/repo_utils.py

@@ -127,6 +127,42 @@ def list_filtered_repo_files(
     return file_list
 
 
+def any_pattern_in_repo_files(
+    model_name_or_path: str,
+    allow_patterns: list[str],
+    revision: str | None = None,
+    repo_type: str | None = None,
+    token: str | bool | None = None,
+):
+    return (
+        len(
+            list_filtered_repo_files(
+                model_name_or_path=model_name_or_path,
+                allow_patterns=allow_patterns,
+                revision=revision,
+                repo_type=repo_type,
+                token=token,
+            )
+        )
+        > 0
+    )
+
+
+def is_mistral_model_repo(
+    model_name_or_path: str,
+    revision: str | None = None,
+    repo_type: str | None = None,
+    token: str | bool | None = None,
+) -> bool:
+    return any_pattern_in_repo_files(
+        model_name_or_path=model_name_or_path,
+        allow_patterns=["consolidated*.safetensors"],
+        revision=revision,
+        repo_type=repo_type,
+        token=token,
+    )
+
+
 def file_exists(
     repo_id: str,
     file_name: str,