Add vLLM library info to Hugging Face Hub requests (#43857)

Signed-off-by: Wauplin <lucainp@gmail.com>
Signed-off-by: Lucain Pouget <lucain@huggingface.co>
Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
This commit is contained in:
Lucain
2026-05-29 14:04:58 +00:00
committed by GitHub
co-authored by Claude Opus 4.6
parent 3f6f508e14
commit 11dfa3169d
13 changed files with 78 additions and 43 deletions
@@ -10,7 +10,6 @@ from typing import Any
import numpy as np
import torch
from huggingface_hub import HfApi
from packaging import version
from torch import nn
from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
@@ -48,6 +47,7 @@ from vllm.model_executor.utils import (
set_weight_attrs,
)
from vllm.platforms import current_platform
from vllm.transformers_utils.repo_utils import hf_api
from vllm.utils.torch_utils import set_default_torch_dtype
logger = init_logger(__name__)
@@ -97,8 +97,7 @@ class BitsAndBytesModelLoader(BaseModelLoader):
if weight_files:
return model_name_or_path, weight_files, pattern
else:
hf_api = HfApi()
repo_files = hf_api.list_repo_files(repo_id=model_name_or_path)
repo_files = hf_api().list_repo_files(repo_id=model_name_or_path)
for pattern in allowed_patterns:
matching_files = fnmatch.filter(repo_files, pattern)
if matching_files:
@@ -8,7 +8,6 @@ import gguf
import regex as re
import torch
import torch.nn as nn
from huggingface_hub import hf_hub_download
from transformers import AutoModelForCausalLM, AutoModelForImageTextToText
from vllm.config import ModelConfig, VllmConfig
@@ -27,6 +26,7 @@ from vllm.model_executor.model_loader.weight_utils import (
gguf_quant_weights_iterator_multi,
)
from vllm.transformers_utils.gguf_utils import detect_gguf_multimodal
from vllm.transformers_utils.repo_utils import hf_api
from vllm.utils.torch_utils import set_default_torch_dtype
if TYPE_CHECKING:
@@ -57,7 +57,7 @@ class GGUFModelLoader(BaseModelLoader):
# repo id/filename.gguf
if "/" in model_name_or_path and model_name_or_path.endswith(".gguf"):
repo_id, filename = model_name_or_path.rsplit("/", 1)
return hf_hub_download(
return hf_api().hf_hub_download(
repo_id=repo_id,
filename=filename,
revision=model_config.revision,
@@ -16,7 +16,6 @@ from typing import TYPE_CHECKING, Any, ClassVar
import regex as re
import torch
from huggingface_hub import snapshot_download
from torch import nn
from torch.utils._python_dispatch import TorchDispatchMode
from transformers import PretrainedConfig
@@ -26,6 +25,7 @@ from vllm.config import ModelConfig, ParallelConfig, VllmConfig, set_current_vll
from vllm.logger import init_logger
from vllm.model_executor.layers.vocab_parallel_embedding import VocabParallelEmbedding
from vllm.platforms import current_platform
from vllm.transformers_utils.repo_utils import hf_api
from vllm.utils.argparse_utils import FlexibleArgumentParser
from vllm.utils.import_utils import PlaceholderModule
@@ -629,7 +629,7 @@ def serialize_extra_artifacts(
)
with tempfile.TemporaryDirectory() as tmpdir:
snapshot_download(
hf_api().snapshot_download(
served_model_name,
local_dir=tmpdir,
ignore_patterns=[
@@ -23,7 +23,6 @@ import huggingface_hub.constants
import numpy as np
import regex as re
import torch
from huggingface_hub import HfFileSystem, hf_hub_download, snapshot_download
from safetensors.torch import load, load_file, safe_open, save_file
from tqdm.auto import tqdm
from transformers.utils import SAFE_WEIGHTS_INDEX_NAME
@@ -46,6 +45,7 @@ from vllm.model_executor.model_loader.ep_weight_filter import (
)
from vllm.platforms import current_platform
from vllm.tracing import instrument
from vllm.transformers_utils.repo_utils import hf_api, hf_fs
from vllm.utils.import_utils import PlaceholderModule
try:
@@ -373,7 +373,7 @@ def get_quant_config(
if not is_local:
# Download the config files.
with get_lock(model_config.model, load_config.download_dir):
hf_folder = snapshot_download(
hf_folder = hf_api().snapshot_download(
model_config.model,
revision=model_config.revision,
allow_patterns="*.json",
@@ -431,7 +431,7 @@ def get_sparse_attention_config(
if not is_local:
# Download the config files.
with get_lock(model_name_or_path, load_config.download_dir):
hf_folder = snapshot_download(
hf_folder = hf_api().snapshot_download(
model_name_or_path,
revision=model_config.revision,
allow_patterns="*.json",
@@ -534,7 +534,7 @@ def download_weights_from_hf(
# Attempt to reduce allow_patterns to a single pattern
# so we only have to call snapshot_download once.
try:
fs = HfFileSystem()
fs = hf_fs()
file_list = fs.ls(
os.path.join(model_name_or_path, subfolder or ""),
detail=False,
@@ -546,7 +546,7 @@ def download_weights_from_hf(
# unnecessary files (e.g., from subdirectories like "original/").
index_file = f"{model_name_or_path}/{SAFE_WEIGHTS_INDEX_NAME}"
if "*.safetensors" in allow_patterns and index_file in file_list:
index_path = hf_hub_download(
index_path = hf_api().hf_hub_download(
repo_id=model_name_or_path,
filename=SAFE_WEIGHTS_INDEX_NAME,
cache_dir=cache_dir,
@@ -582,7 +582,7 @@ def download_weights_from_hf(
with get_lock(model_name_or_path, cache_dir):
start_time = time.perf_counter()
for allow_pattern in allow_patterns:
hf_folder = snapshot_download(
hf_folder = hf_api().snapshot_download(
model_name_or_path,
allow_patterns=allow_pattern,
ignore_patterns=ignore_patterns,
@@ -631,7 +631,7 @@ def download_safetensors_index_file_from_hf(
with get_lock(model_name_or_path, cache_dir):
try:
# Download the safetensors index file.
hf_hub_download(
hf_api().hf_hub_download(
repo_id=model_name_or_path,
filename=index_file,
cache_dir=cache_dir,