README.md

---
pipeline_tag: any-to-any
datasets:
- openbmb/RLAIF-V-Dataset
library_name: transformers
language:
- multilingual
tags:
- minicpm-o
- omni
- vision
- ocr
- multi-image
- video
- custom_code
- audio
- speech
- voice cloning
- live Streaming
- realtime speech conversation
- asr
- tts
---

<h1>A GPT-4o Level MLLM for Vision, Speech and Multimodal Live Streaming on Your Phone</h1>

## MiniCPM-o 2.6 int4
This is the int4 quantized version of [**MiniCPM-o 2.6**](https://huggingface.co/openbmb/MiniCPM-o-2_6).   
Running with int4 version would use lower GPU memory (about 9GB).

### Prepare code and install AutoGPTQ

We are submitting PR to officially support minicpm-o 2.6 inference

```python
git clone https://github.com/OpenBMB/AutoGPTQ.git && cd AutoGPTQ
git checkout minicpmo

# install AutoGPTQ
pip install -vvv --no-build-isolation -e .
```

### Usage of **MiniCPM-o-2_6-int4**

Change the model initialization part to `AutoGPTQForCausalLM.from_quantized`

```python
import torch
from transformers import AutoModel, AutoTokenizer
from auto_gptq import AutoGPTQForCausalLM

model = AutoGPTQForCausalLM.from_quantized(
    'openbmb/MiniCPM-o-2_6-int4',
    torch_dtype=torch.bfloat16,
    device="cuda:0",
    trust_remote_code=True,
    disable_exllama=True,
    disable_exllamav2=True
)
tokenizer = AutoTokenizer.from_pretrained(
    'openbmb/MiniCPM-o-2_6-int4',
    trust_remote_code=True
)

model.init_tts()

```

Usage reference [MiniCPM-o-2_6#usage](https://huggingface.co/openbmb/MiniCPM-o-2_6#usage)