### 镜像是官方的： harbor.sourcefind.cn:5443/dcu/admin/base/pytorch:2.7.1-ubuntu22.04-dtk26.04-py3.10 官网上手动下载安装了 torchaudio ```requirements.txt Package Version ---------------------------------- ------------------------------------------------------- accelerate 1.13.0 aiofiles 25.1.0 aiohappyeyeballs 2.6.1 aiohttp 3.13.3 aiohttp-cors 0.8.1 aiosignal 1.4.0 amdsmi 24.5.3+02cbffb.dirty annotated-doc 0.0.4 annotated-types 0.7.0 anyio 4.12.1 apex 1.5.0+das.opt1.dtk2604.20260128.g89c00191 astor 0.8.1 async-timeout 5.0.1 attrs 25.4.0 audioread 3.1.0 backports.asyncio.runner 1.2.0 blake3 1.0.8 blinker 1.9.0 boto3 1.42.61 botocore 1.42.61 cachetools 7.0.2 cbor2 5.8.0 certifi 2026.2.25 cffi 2.0.0 charset-normalizer 3.4.4 click 8.2.1 cloudpickle 3.1.2 cmake 3.29.0 colorful 0.5.8 compressed-tensors 0.11.0 contourpy 1.3.2 cryptography 46.0.5 cuda-pathfinder 1.4.0 cupy 12.3.0 cupy-cuda12x 14.0.1 cycler 0.12.1 datasets 4.6.1 dbus-python 1.2.18 dcu-megatron 0.15.0+das.opt1.dtk2604.torch271.20260204.gae67d7e2 decorator 5.3.1 depyf 0.19.0 dill 0.4.0 diskcache 5.6.3 distlib 0.4.0 distro 1.7.0 dnspython 2.8.0 edit-distance 1.0.9 editdistance 0.8.1 eft 0.0.7 einops 0.8.2 email-validator 2.3.0 evaluate 0.4.6 exceptiongroup 1.3.1 fastapi 0.135.1 fastapi-cli 0.0.24 fastapi-cloud-cli 0.14.0 fastar 0.8.0 fastrlock 0.8.3 filelock 3.25.0 flash_attn 2.6.1+das.opt1.dtk2604.torch271.20260304.g6c6c3ed1 flash_mla 1.0.0+das.opt1.dtk2604.torch271.20260303.g4742da75 Flask 3.1.3 fonttools 4.61.1 frozendict 2.4.7 frozenlist 1.8.0 fsspec 2026.2.0 future 1.0.0 gguf 0.18.0 google-api-core 2.30.0 google-auth 2.48.0 googleapis-common-protos 1.72.0 greenlet 3.3.2 grpcio 1.78.0 h11 0.16.0 h2 4.3.0 hf-xet 1.3.2 hiredis 3.3.0 hpack 4.1.0 httpcore 1.0.9 httplib2 0.20.2 httptools 0.7.1 httpx 0.28.1 huggingface_hub 1.5.0 humanize 4.15.0 Hypercorn 0.18.0 hyperframe 6.1.0 hypothesis 6.151.9 idna 3.11 importlib_metadata 8.7.1 iniconfig 2.3.0 interegular 0.3.3 itsdangerous 2.2.0 jeepney 0.7.1 Jinja2 3.1.6 jiter 0.13.0 jmespath 1.1.0 joblib 1.5.3 jsonschema 4.26.0 jsonschema-specifications 2025.9.1 keyring 23.5.0 kiwisolver 1.4.9 lark 1.2.2 launchpadlib 1.10.16 lazr.restfulclient 0.14.4 lazr.uri 1.0.6 lazy-loader 0.5 libnacl 2.1.0 librosa 0.11.0 lightop 0.6.0+das.dtk2604.torch271.20260210.g1fdc3ed2 llguidance 0.7.30 llvmlite 0.44.0 lm-format-enforcer 0.11.3 lmslim 0.3.1+das.opt4.dtk2604.torch271.20260210.gc8327617 markdown-it-py 4.0.0 MarkupSafe 3.0.3 matplotlib 3.10.8 mdurl 0.1.2 megatron-core 0.15.0+das.opt1.dtk2604.torch271.20260204.g0d7e02bd mistral_common 1.9.1 more-itertools 8.10.0 mpmath 1.3.0 msgpack 1.1.2 msgspec 0.20.0 multidict 6.7.1 multiprocess 0.70.18 networkx 3.4.2 ninja 1.11.1 numa 1.4.6 numba 0.61.2 numpy 1.25.0 oauthlib 3.2.0 openai 2.24.0 openai-harmony 0.0.8 opencensus 0.11.4 opencensus-context 0.1.3 opencv-python-headless 4.13.0.92 opentelemetry-api 1.40.0 opentelemetry-exporter-prometheus 0.61b0 opentelemetry-proto 1.40.0 opentelemetry-sdk 1.40.0 opentelemetry-semantic-conventions 0.61b0 outlines_core 0.2.11 packaging 26.0 pandas 2.3.3 partial-json-parser 0.2.1.1.post7 peft 0.18.1 pillow 12.1.1 pip 26.0.1 platformdirs 4.9.2 pluggy 1.6.0 pooch 1.9.0 priority 2.0.0 prometheus_client 0.24.1 prometheus-fastapi-instrumentator 7.1.0 propcache 0.4.1 proto-plus 1.27.1 protobuf 7.34.0 psutil 7.2.2 py-cpuinfo 9.0.0 py-spy 0.4.1 pyarrow 23.0.1 pyasn1 0.6.2 pyasn1_modules 0.4.2 pybase64 1.4.3 pybind11 3.0.2 pycountry 26.2.16 pycparser 3.0 pydantic 2.12.5 pydantic_core 2.41.5 pydantic-extra-types 2.11.0 pydantic-settings 2.13.1 Pygments 2.19.2 PyGObject 3.42.1 PyHive 0.7.0 PyJWT 2.3.0 PyMySQL 1.1.2 pyparsing 3.3.2 pytest 9.0.2 pytest-asyncio 1.3.0 python-apt 2.4.0+ubuntu4 python-dateutil 2.9.0.post0 python-discovery 1.1.0 python-dotenv 1.2.2 python-json-logger 4.0.0 python-multipart 0.0.22 PyTrie 0.4.0 pytz 2026.1.post1 PyYAML 6.0.3 pyzmq 27.1.0 Quart 0.20.0 ray 2.48.0 redis 7.2.1 referencing 0.37.0 regex 2026.2.28 requests 2.32.5 rich 14.3.3 rich-toolkit 0.19.7 rignore 0.7.6 rpds-py 0.30.0 rsa 4.9.1 runai-model-streamer 0.11.0 runai-model-streamer-s3 0.11.0 s3transfer 0.16.0 safetensors 0.7.0 scikit-learn 1.7.2 scipy 1.15.3 SecretStorage 3.3.1 sentence-transformers 5.7.0 sentencepiece 0.2.1 sentry-sdk 2.54.0 setproctitle 1.3.7 setuptools 79.0.1 setuptools-scm 9.2.2 shellingham 1.5.4 six 1.16.0 smart_open 7.5.1 sniffio 1.3.1 sortedcontainers 2.4.0 soundfile 0.14.0 soxr 1.1.0 SQLAlchemy 2.0.48 starlette 0.52.1 sympy 1.14.0 taskgroup 0.2.2 tensorboardX 2.6.4 tensorizer 2.10.1 textdistance 4.6.3 threadpoolctl 3.6.0 tiktoken 0.12.0 timm 1.0.25 tokenizers 0.22.2 tomli 2.4.0 torch 2.7.1+das.opt1.dtk2604.20260206.g04100de4 torchaudio 2.7.1+das.opt1.dtk2604.torch271 torchdata 0.11.0 torchvision 0.22.0+das.opt1.dtk2604.20260210.g2a371b9c tqdm 4.67.3 transformer_engine 2.10.0+das.opt1.dtk2604.20260202.gd954fc54 transformers 5.3.0 triton 3.1.0+das.opt2.dtk2604.torch271.20260211.g6964b550 typer 0.24.1 typing_extensions 4.15.0 typing-inspection 0.4.2 tzdata 2025.3 urllib3 2.6.3 uvicorn 0.41.0 uvloop 0.22.1 virtualenv 21.1.0 vllm 0.11.0+das.opt1.rc4.dtk2604.torch271.20260305.g49a30c70 wadllib 1.3.6 watchfiles 1.1.1 websockets 16.0 Werkzeug 3.1.6 wheel 0.37.1 wrapt 2.1.1 wsproto 1.3.2 xgrammar 0.1.25 xxhash 3.6.0 yarl 1.23.0 zhconv 1.4.3 zipp 3.23.0 ``` ### 代码 ```python import os import torch from transformers import AutoModelForCausalLM, AutoProcessor, AutoTokenizer #os.environ["CUDA_VISIBLE_DEVICES"] = "4" os.environ["HF_HUB_OFFLINE"] = "1" os.environ["TRANSFORMERS_OFFLINE"] = "1" # 部 model_id = "mispeech/midashenglm-0.6b-fp32" # 或 7B model_path="/root/.cache/huggingface/hub/models--mispeech--midashenglm-0.6b-fp32/snapshots/3969e451d1c62618089c1135fe4c58ed33af2457" model = AutoModelForCausalLM.from_pretrained( model_path, device_map="cuda:0", # 加载到第一张可见 GPU（即物理 GPU 4） trust_remote_code=True, local_files_only=True ) tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True,local_files_only=True ) processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True,local_files_only=True ) user_prompt = "What is the speaker saying?" user_prompt = "Caption the audio" messages = [ { "role": "user", "content": [ {"type": "text", "text": user_prompt}, {"type": "audio", "path": "/root/xxx/code/audiollm/1.MP3"}, ], }, ] print("Processing inputs...") model_inputs = processor.apply_chat_template( messages, tokenize=True, add_generation_prompt=True, add_special_tokens=True, return_dict=True, ) # 关键：将所有输入张量移动到模型所在的设备 model_inputs = {k: v.to(model.device) for k, v in model_inputs.items()} print("Generating...") with torch.no_grad(): generation = model.generate( **model_inputs, max_new_tokens=64, # 建议设置，避免生成过长 do_sample=False, # 贪心解码，更快更稳定 ) output = tokenizer.batch_decode(generation, skip_special_tokens=True) print("Output:", output) ~ ``` ### 输出： ``` Output: ['! choosing festival.account/store.fr! TransmissionSerialize repeatedly courts Maid!Qt!!!!!!!!!!!!!!!!!!!!!!!!!!!UIImage naked contributionsMonster simulation governor gov! wage juvenile 数 trials DEV!sure Which assumed ine NAME philosophy>"\n Nations nylon'] ```