README.md

---
license: mit
pipeline_tag: video-text-to-text
library_name: transformers
---

# M4-Audio-LongVA-7B-Qwen2

Enhancing Omni Interactive Capabilities in MLLM

This repository contains the model described in [OmniMMI: A Comprehensive Multi-modal Interaction Benchmark in Streaming Video Contexts](https://huggingface.co/papers/2503.22952).
The code can be found at https://github.com/patrick-tssn/M4.

![images](./assets/framework.png)

M4-Audio-7B is an extension of [LongVA-7B](https://github.com/EvolvingLMMs-Lab/LongVA), further trained using the [M4-IT](https://huggingface.co/datasets/ColorfulAI/M4-IT) dataset, which comprises 9,963 visual-audio instruction tuning instances. This training was conducted without any special modifications to the existing training pipeline.


## Usage


*Please refer to [M4](https://github.com/patrick-tssn/M4) to install relvevant packages*

```python
import os
from PIL import Image
import numpy as np
import torchaudio
import torch
from decord import VideoReader, cpu
import whisper
# fix seed
torch.manual_seed(0)

from intersuit.model.builder import load_pretrained_model
from intersuit.mm_utils import tokenizer_image_speech_tokens, process_images
from intersuit.constants import IMAGE_TOKEN_INDEX, SPEECH_TOKEN_INDEX

import ChatTTS
chat = ChatTTS.Chat()
chat.load(source='local', compile=True)

import warnings
warnings.filterwarnings("ignore")

model_path = "checkpoints/M4-Audio-LongVA-7B-Qwen2"
video_path = "local_demo/assets/water.mp4"
audio_path = "local_demo/wav/infer.wav"
new_audio_path = "local_demo/wav/new_infer.wav"
max_frames_num = 16 # you can change this to several thousands so long you GPU memory can handle it :)
gen_kwargs = {"do_sample": True, "temperature": 0.5, "top_p": None, "num_beams": 1, "use_cache": True, "max_new_tokens": 1024}
tokenizer, model, image_processor, _ = load_pretrained_model(model_path, None, "llava_qwen", device_map="cuda:0", attn_implementation="eager")

# original query
query = "Give a detailed caption of the video as if I am blind."
query = None # comment this to use ChatTTS to convert the query to audio
prompt = "<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
<image><|im_end|>
<|im_start|>user
<speech>
<|im_end|>
<|im_start|>assistant
"
input_ids = tokenizer_image_speech_tokens(prompt, tokenizer, IMAGE_TOKEN_INDEX, SPEECH_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(model.device)
pad_token_ids = (tokenizer.pad_token_id if tokenizer.pad_token_id is not None else tokenizer.eos_token_id)
attention_masks = input_ids.ne(pad_token_ids).to(input_ids.device)
# audio input
if query is not None:
    audio_path = "./local_demo/wav/" + "infer.wav"
    if os.path.exists(audio_path): os.remove(audio_path) # refresh
    if not os.path.exists(audio_path):
        wav = chat.infer(query)
        try:
            torchaudio.save(audio_path, torch.from_numpy(wav).unsqueeze(0), 24000)
        except:
            torchaudio.save(audio_path, torch.from_numpy(wav), 24000)
speech = whisper.load_audio(audio_path)
speech = whisper.pad_or_trim(speech)
speech = whisper.log_mel_spectrogram(speech, n_mels=128).permute(1, 0).to(device=model.device, dtype=torch.float16)
speech_length = torch.LongTensor([speech.shape[0]]).to(model.device)

# new query
new_query = "How many people in the video?"
new_query = "Okay, I see."
new_query = "Sorry to interrupt."
new_query_pos = 10 # which token encounter the new query
new_query = None # comment this to use ChatTTS to convert the query to audio
new_prompt = "<|im_start|>system
You are a helpful assistant.<|im_end|>
<|im_start|>user
<speech>
<|im_end|>
<|im_start|>assistant
"
new_input_ids = tokenizer_image_speech_tokens(new_prompt, tokenizer, IMAGE_TOKEN_INDEX, SPEECH_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(model.device)
# audio input
if new_query is not None:
    new_audio_path = "./local_demo/wav/" + "new_infer.wav"
    if os.path.exists(new_audio_path): os.remove(new_audio_path) # refresh
    if not os.path.exists(new_audio_path):
        wav = chat.infer(new_query)
        try:
            torchaudio.save(new_audio_path, torch.from_numpy(wav).unsqueeze(0), 24000)
        except:
            torchaudio.save(new_audio_path, torch.from_numpy(wav), 24000)
new_speech = whisper.load_audio(new_audio_path)
new_speech = whisper.pad_or_trim(new_speech)
new_speech = whisper.log_mel_spectrogram(new_speech, n_mels=128).permute(1, 0).to(device=model.device, dtype=torch.float16)
new_speech_length = torch.LongTensor([new_speech.shape[0]]).to(model.device)

#video input
vr = VideoReader(video_path, ctx=cpu(0))
total_frame_num = len(vr)
uniform_sampled_frames = np.linspace(0, total_frame_num - 1, max_frames_num, dtype=int)
frame_idx = uniform_sampled_frames.tolist()
frames = vr.get_batch(frame_idx).asnumpy()
video_tensor = image_processor.preprocess(frames, return_tensors="pt")["pixel_values"].to(model.device, dtype=torch.bfloat16)


with torch.inference_mode():
    output_ids = model.generate_parallel(input_ids, 
                                attention_mask=attention_masks,
                                images=[video_tensor], 
                                modalities=["video"], 
                                speeches=speech.unsqueeze(0), 
                                speech_lengths=speech_length,
                                new_query=new_input_ids,
                                new_query_pos=new_query_pos,
                                new_speeches=new_speech.unsqueeze(0),
                                new_speech_lengths=new_speech_length,
                                query_str=query,
                                new_query_str=new_query,
                                tokenizer=tokenizer,
                                **gen_kwargs)
outputs = tokenizer.batch_decode(output_ids, skip_special_tokens=True)[0].strip()

```


For more information about the interaction inference pipeline, please visit the [M4 GitHub repository](https://github.com/patrick-tssn/M4).