音频文本到文本
音频-文本到文本模型同时接受音频和文本作为输入,并生成文本作为输出。它们将音频理解与语言生成相结合,支持诸如音频问答(例如,“这段音频中说了什么?”)、音频推理(例如,“说话者传达了什么情绪?”)以及口语对话理解等任务。与仅将语音转录为文本的传统自动语音识别(ASR)模型不同,音频-文本到文本模型能够对音频内容进行推理,遵循复杂的指令,并根据所听到的内容生成上下文相关的响应。
下面的示例展示了如何加载模型和处理器,传入带有文本提示的音频文件,并生成响应。在此场景中,我们要求模型对语音录音进行转录。
from transformers import AudioFlamingo3ForConditionalGeneration, AutoProcessor
model_id = "nvidia/audio-flamingo-3-hf"
processor = AutoProcessor.from_pretrained(model_id)
model = AudioFlamingo3ForConditionalGeneration.from_pretrained(model_id, device_map="auto")
conversation = [
{
"role": "user",
"content": [
{"type": "text", "text": "Transcribe the input speech."},
{"type": "audio", "path": "https://huggingface.co/datasets/nvidia/AudioSkills/resolve/main/assets/WhDJDIviAOg_120_10.mp3"},
],
}
]
inputs = processor.apply_chat_template(
conversation,
tokenize=True,
add_generation_prompt=True,
return_dict=True,
).to(model.device)
outputs = model.generate(**inputs, max_new_tokens=500)
decoded_outputs = processor.batch_decode(outputs[:, inputs.input_ids.shape[1]:], skip_special_tokens=True)
print(decoded_outputs)
## ["The transcription of the audio is 'summer follows spring the days grow longer and the nights are warm'."]本指南将向您展示如何:
- 使用 LoRA 在 AudioCaps 数据集上对 Audio Flamingo 3 进行微调,以实现音频字幕生成。
- 使用您微调后的模型进行推理。
要查看与此任务兼容的所有架构和检查点,我们建议查阅任务页面。
在开始之前,请确保已安装所有必要的库:
pip install transformers datasets peft accelerate
我们建议您登录 Hugging Face 账户,以便上传模型并与社区共享。当系统提示时,请输入您的令牌以完成登录:
>>> from huggingface_hub import notebook_login
>>> notebook_login()加载 AudioCaps 数据集
首先,以流式模式从 🤗 Datasets 库中加载 AudioCaps 数据集。该数据集包含带有描述性字幕的音频片段,非常适合用于音频字幕生成任务。
>>> from datasets import load_dataset, Audio
>>> dataset = load_dataset("OpenSound/AudioCaps", split="train", streaming=True)将音频列转换为 16kHz,这是 Audio Flamingo 的 Whisper 特征提取器所要求的:
>>> dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))使用 .take() 和 .skip() 将流式数据集拆分为训练集和测试集:
>>> train_dataset = dataset.take(1000)
>>> eval_dataset = dataset.skip(1000).take(100)查看一个示例:
>>> next(iter(train_dataset))
{'audiocap_id': 91139, 'youtube_id': 'r1nicOVtvkQ', 'start_time': 130, 'caption': 'A woman talks nearby as water pours', 'audio_length': 480000, 'audio': <datasets.features._torchcodec.AudioDecoder object at ...>}
该数据集包含:
audio:音频波形caption:音频的描述性文字说明
预处理
加载 Audio Flamingo 处理器以同时处理音频和文本输入:
>>> from transformers import AutoProcessor
>>> processor = AutoProcessor.from_pretrained("nvidia/audio-flamingo-3-hf")创建一个数据整理器,将音频-文本对处理为 Audio Flamingo 所期望的格式。该整理器使用聊天模板格式,并直接处理音频数组:
>>> class AudioFlamingo3DataCollator:
... """Data collator for Audio Flamingo 3 audio captioning training."""
...
... def __init__(self, processor):
... self.processor = processor
...
... def __call__(self, features):
... conversations = []
...
... for feature in features:
... # Build conversation format for Audio Flamingo
... # Audio is passed directly as an array, no base64 encoding needed
... sample = [
... {
... "role": "user",
... "content": [
... {"type": "text", "text": "Describe the audio."},
... {"type": "audio", "audio": feature["audio"].get_all_samples().data[0].numpy()},
... ],
... },
... {
... "role": "assistant",
... "content": [{"type": "text", "text": feature["caption"]}],
... }
... ]
... conversations.append(sample)
...
... # Apply chat template and format labels for training
... return self.processor.apply_chat_template(
... conversations,
... tokenize=True,
... add_generation_prompt=False,
... return_dict=True,
... output_labels=True, # Automatically creates labels for training
... )实例化数据整理器:
>>> data_collator = AudioFlamingo3DataCollator(processor)训练
加载 Audio Flamingo 模型。我们使用 bfloat16 精度和 device_map="auto" 以实现高效的内存使用:
>>> from transformers import AudioFlamingo3ForConditionalGeneration
>>> import torch
>>> model = AudioFlamingo3ForConditionalGeneration.from_pretrained(
... "nvidia/audio-flamingo-3-hf",
... torch_dtype=torch.bfloat16,
... device_map="auto",
... )配置 LoRA
LoRA(低秩适应)通过仅训练少量额外参数来实现高效微调。请配置 LoRA,使其针对语言模型的注意力层和前馈层:
>>> from peft import LoraConfig, get_peft_model
>>> lora_config = LoraConfig(
... r=16, # LoRA rank
... lora_alpha=32, # LoRA scaling factor
... target_modules=[
... # Language model attention
... "q_proj",
... "k_proj",
... "v_proj",
... "o_proj",
... # Feed-forward layers
... "gate_proj",
... "up_proj",
... "down_proj",
... ],
... lora_dropout=0.05,
... bias="none",
... task_type="CAUSAL_LM",
... )
>>> model = get_peft_model(model, lora_config)
>>> model.print_trainable_parameters()
trainable params: 44,302,336 || all params: 8,311,517,696 || trainable%: 0.5330
LoRA 通过仅更新少量适配器参数而非整个模型,显著降低了内存占用和训练时间。此配置针对语言模型的注意力和前馈层进行微调,同时保持音频编码器冻结,从而可以在单块 GPU 上完成微调。
设置训练
在 TrainingArguments 中定义训练超参数。请注意,由于我们使用的是流式数据集,因此使用 max_steps 而不是 epochs:
>>> from transformers import TrainingArguments, Trainer
>>> training_args = TrainingArguments(
... output_dir="audio-flamingo-3-hf-lora-finetuned",
... per_device_train_batch_size=4,
... per_device_eval_batch_size=4,
... gradient_accumulation_steps=4,
... learning_rate=1e-4,
... max_steps=500, # Use max_steps with streaming datasets
... bf16=True,
... logging_steps=10,
... eval_steps=100,
... save_steps=250,
... save_total_limit=2, # Keep only the latest 2 checkpoints
... save_only_model=True, # Skip saving optimizer state to save disk space
... eval_strategy="steps",
... save_strategy="steps",
... remove_unused_columns=False,
... dataloader_num_workers=0, # Must be 0 for streaming datasets
... gradient_checkpointing=True,
... report_to="none",
... )将训练参数与模型、数据集和数据整理器一起传递给 Trainer:
>>> trainer = Trainer(
... model=model,
... args=training_args,
... train_dataset=train_dataset,
... eval_dataset=eval_dataset,
... data_collator=data_collator,
... )
>>> trainer.train()保存 LoRA 适配器和处理器:
>>> trainer.save_model()
>>> processor.save_pretrained("audio-flamingo-3-hf-lora-finetuned")
['audio-flamingo-3-hf-lora-finetuned/processor_config.json']
训练完成后,将您的模型共享到 Hub:
>>> trainer.push_to_hub()推理
现在,您已对模型进行了微调,可以使用它进行音频字幕生成。
加载微调后的模型和处理器:
>>> from transformers import AudioFlamingo3ForConditionalGeneration, AutoProcessor
>>> from peft import PeftModel
>>> import torch
>>> base_model = AudioFlamingo3ForConditionalGeneration.from_pretrained(
... "nvidia/audio-flamingo-3-hf",
... torch_dtype=torch.bfloat16,
... device_map="auto",
... )
>>> model = PeftModel.from_pretrained(base_model, "audio-flamingo-3-hf-lora-finetuned")
>>> processor = AutoProcessor.from_pretrained("audio-flamingo-3-hf-lora-finetuned")加载用于推理的音频样本:
>>> from datasets import load_dataset, Audio
>>> dataset = load_dataset("OpenSound/AudioCaps", split="test", streaming=True)
>>> dataset = dataset.cast_column("audio", Audio(sampling_rate=16000))
>>> sample = next(iter(dataset))
>>> audio = sample["audio"]
>>> audio_array = audio.get_all_samples().data[0].numpy() if hasattr(audio, "get_all_samples") else audio["array"]使用对话格式准备输入:
>>> messages = [
... {
... "role": "user",
... "content": [
... {"type": "text", "text": "Describe the audio."},
... {"type": "audio", "audio": audio_array},
... ],
... }
... ]
>>> inputs = processor.apply_chat_template(
... messages,
... tokenize=True,
... add_generation_prompt=True,
... return_dict=True,
... ).to(device=model.device, dtype=model.dtype)生成响应:
>>> with torch.no_grad():
... output_ids = model.generate(**inputs, max_new_tokens=100)
>>> # Decode only the generated tokens
>>> input_len = inputs["input_ids"].shape[1]
>>> response = processor.tokenizer.decode(output_ids[0][input_len:], skip_special_tokens=True)
>>> print(response)
## A sewing machine is running while people are talkingPipeline
For quick inference, use the Pipeline API with an any-to-any model. The example below uses Voxtral, which accepts audio and text inputs and generates text. See the any-to-any task guide for more examples.
from transformers import pipeline
pipe = pipeline("any-to-any", model="mistralai/Voxtral-Mini-3B-2507")
messages = [
{
"role": "user",
"content": [
{
"type": "audio",
"url": "https://huggingface.co/datasets/raushan-testing-hf/audio-test/resolve/main/glass-breaking-151256.mp3",
},
{"type": "text", "text": "What do you hear in this audio?"},
],
},
]
outputs = pipe(text=messages, max_new_tokens=100, return_full_text=False)
print(outputs[0]["generated_text"])延伸阅读
- 音频文本到文本任务页面涵盖了模型类型、用例和数据集。
- 有关更多 LoRA 配置选项和其他适配器方法,请参阅 PEFT 文档。
- 有关模型特定细节和功能,请参阅 Audio Flamingo 3 模型卡片。