Granite-Speech-4.1-2B Ascend NPU 部署指南

/data/ysws/agentsp/granite-speech-4.1-2b-ascend/
├── inference.py          # 精度测试脚本
├── log.txt               # 测试日志
├── README.md             # 本文档
├── test_audio_1s.pt      # 测试音频样本 (1秒)
└── test_audio_3s.pt      # 测试音频样本 (3秒)

docker exec -it test-modelagent bash

source /usr/local/Ascend/ascend-toolkit/set_env.sh

cd /data/ysws/agentsp/granite-speech-4.1-2b-ascend/
python3 inference.py

import torch
from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor

model_path = "/data/ysws/agentsp/granite-speech-4.1-2b"
device = torch.device("npu:0")

model = AutoModelForSpeechSeq2Seq.from_pretrained(
    model_path,
    dtype=torch.bfloat16,
    low_cpu_mem_usage=True
).to(device).eval()

processor = AutoProcessor.from_pretrained(model_path)

speech = torch.randn(16000 * 3).numpy()
inputs = processor(text="transcribe", audio=speech, return_tensors="pt")
input_ids = inputs["input_ids"].to(device)

with torch.no_grad():
    output = model.generate(input_ids=input_ids, max_new_tokens=256)
    transcription = processor.batch_decode(output, skip_special_tokens=True)[0]
    print(transcription)

inputs = processor(
    text="transcribe",  # 文本提示 (必需)
    audio=speech,      # 音频数据
    return_tensors="pt"
)