Add vLLM-Omni library info to Hugging Face Hub requests (#5381)

Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
Co-authored-by: Hongsheng Liu <liuhongsheng4@huawei.com>
Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
This commit is contained in:
authored and GitHub committed 2026-09-08 10:08:38 -06:00
1 parent 7a80b79698
commit fdaf4e31bd
57 files changed
+259 -220

No files matched your search

@@ -1,3 +1,6 @@
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
"""Seed-TTS WER aligned with Bytedance ``seed-tts-eval`` / ``run_wer.py``.
Matches the published protocol (see Hugging Face dataset card and
@@ -329,7 +332,8 @@ def _ensure_utmos_jit_model() -> Any | None:
return _utmos_jit_model
try:
import torch
from huggingface_hub import hf_hub_download
from vllm_omni.transformers_utils.repo_utils import hf_api
repo = os.environ.get("SEED_TTS_UTMOS_HF_REPO", "balacoon/utmos").strip() or "balacoon/utmos"
fname = os.environ.get("SEED_TTS_UTMOS_JIT_FILE", "utmos.jit").strip() or "utmos.jit"
@@ -338,7 +342,7 @@ def _ensure_utmos_jit_model() -> Any | None:
repo,
fname,
)
path = hf_hub_download(repo_id=repo, filename=fname, repo_type="model")
path = hf_api().hf_hub_download(repo_id=repo, filename=fname, repo_type="model")
# TODO The model weights in UTMOS must be loaded in cuda:0; otherwise, the model execution will fail.
want = "cuda:0"