README.md

---
license: mit
language:
- en
- zh
pipeline_tag: text-generation
---

# Innovator-VL-8B-Instruct

## Model Summary

**Innovator-VL-8B-Instruct** is a multimodal instruction-following large language model designed for scientific understanding and reasoning. The model integrates strong general-purpose vision-language capabilities with enhanced scientific multimodal alignment, while maintaining a fully transparent and reproducible training pipeline.

Unlike approaches that rely on large-scale domain-specific pretraining, Innovator-VL-8B-Instruct achieves competitive scientific performance using high-quality instruction tuning, without additional scientific text continued pretraining.

---

## Model Architecture

<img src="assets/innovator_vl_architecture.png" width="600"/>

- **Vision Encoder**: RICE-ViT (region-aware visual representation)  
- **Projector**: PatchMerger for visual token compression  
- **Language Model**: Qwen3-8B-Base  
- **Model Size**: 8B parameters  

The model supports native-resolution multi-image inputs and is suitable for complex scientific visual analysis.

---

## Training Overview

- **Multimodal Alignment**: LLaVA-1.5 (558K)
- **Mid-training**: LLaVA-OneVision-1.5 (85M)
- **Instruction Tuning**: High-quality multimodal and scientific instruction data (~46M)

No additional scientific text continued pretraining is applied.

---

## Intended Use

- Scientific image understanding and question answering
- Multimodal reasoning and analysis
- Interpretation of scientific figures, charts, and experimental results
- General-purpose vision-language instruction following

---

## Inference Example

Below is a minimal example to run multimodal inference (image + text) with `transformers`.

```python
import torch
from transformers import AutoProcessor, AutoModelForCausalLM
from qwen_vl_utils import process_vision_info

model_path = "InnovatorLab/Innovator-VL-8B-Instruct"

# Load the model on the available device(s)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    torch_dtype="auto",
    device_map="auto",
    trust_remote_code=True,
)

# Load processor
processor = AutoProcessor.from_pretrained(
    model_path,
    trust_remote_code=True,
)

messages = [
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen-VL/assets/demo.jpeg",
            },
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

# Preparation for inference
text = processor.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True,
)

image_inputs, video_inputs = process_vision_info(messages)

inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
)

# Move inputs to GPU (optional)
inputs = inputs.to("cuda")

# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=1024)

generated_ids_trimmed = [
    out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
]

output_text = processor.batch_decode(
    generated_ids_trimmed,
    skip_special_tokens=True,
    clean_up_tokenization_spaces=False,
)

print(output_text)
```

---

## Limitations

- The Instruct version does not explicitly optimize long-chain reasoning efficiency.
- For tasks requiring structured or token-efficient reasoning, a dedicated Thinking or RL-aligned model is recommended.

---

## Citation

```bibtex
@article{wen2026innovator,
  title={Innovator-VL: A Multimodal Large Language Model for Scientific Discovery},
  author={Wen, Zichen and Yang, Boxue and Chen, Shuang and Zhang, Yaojie and Han, Yuhang and Ke, Junlong and Wang, Cong and others},
  journal={arXiv preprint arXiv:2601.19325},
  year={2026}
}