Update README.md
Browse files# CapRL-InternVL3.5-8B
📖<a href="https://arxiv.org/abs/2509.22647">Paper</a> | 🏠<a href="https://github.com/InternLM/CapRL">Github</a> |🤗<a href="https://huggingface.co/internlm/CapRL-3B">CapRL-3B Model</a> |🤗<a href="https://huggingface.co/yuhangzang/CapRL-InternVL3.5-8B">CapRL-InternVL3.5-8B Model</a> |
🤗<a href="https://huggingface.co/datasets/internlm/CapRL-2M">CapRL-2M Dataset</a>
🤗<a href="https://huggingface.co/collections/long-xing1/caprl-68d64ac32ded31596c36e189">CapRL Collection</a> | 🤗<a href="https://huggingface.co/papers/2509.22647">Daily Paper</a>
Based on the same recipe as CapRL-3B, we used InternVL3.5-8B as the policy model and obtained CapRL-InternVL3.5-8B through CapRL. Its performance significantly surpasses that of Qwen2.5-VL-72B.
We are working on even stronger base models and upgrading our training recipe — stay tuned!
## Introduction
We are excited to introduce CapRL-3B, a lightweight 3B image captioner that achieves perception capabilities comparable to Qwen2.5-VL-72B.
This is the first study of applying Reinforcement Learning with Verifiable Rewards for the
open-ended and subjective image captioning task. Unlike traditional Supervised Fine-Tuning, which
can lead to models memorizing a limited set of annotated captions, our method allows the model to
explore and generate a broader range of creative and general descriptions.
CapRL is a new training paradigm featuring a decoupled two-stage pipeline. The initial
stage uses LVLMs to generate rich and accurate captions. Subsequently, the second stage evaluates
caption quality by using a vision-only LLM to perform the QA task. We also created a specific QA
curation pipeline to ensure the quality of the questions and answers used for the second stage.
By employing CapRL training framework, initializing with the Qwen2.5-VL-3B model, and using a carefully
filtered 75K QA dataset as the training set, we obtained a highly capable captioner, CapRL-3B.
<p align="center">
<img src="./assets/teaser.png" alt="Main Results on GPT2" width="750"/>
</p>
<p align="center">
<img src="./assets/performance_update.png" alt="Main Results on GPT2" width="750"/>
</p>
## Key Features
* **Remarkable visual understanding for Chart, Infographics and Document**: CapRL-3B achieves perception accuracy and visual information coverage comparable to Qwen2.5-VL-72B.
* **Well-organized output**: The outputs of CapRL-3B are relatively well-structured, making them clear and easy to understand.
* **Detailed description for natural images**: The outputs of CapRL-3B can perfectly cover all valid visual information while containing fewer hallucinations.
## Usage
If you want to use **CapRL-3B** for captioning, you can directly follow the exact same inference approach as in [Qwen2.5-VL-series](https://github.com/QwenLM/Qwen3-VL/tree/d2240f11656bfe404b9ba56db4e51cd09f522ff1).
We recommend using **vLLM** to speed up inference.
### Start an OpenAI API Service
Run the command below to start an OpenAI-compatible API service:
```bash
vllm serve "/PATH/CapRL-3B" \
--trust-remote-code \
--tensor-parallel-size=1 \
--pipeline-parallel-size=1 \
--gpu_memory_utilization=0.95 \
--served-model-name=caprl \
--port 8000 \
--host 0.0.0.0
```
Then you can use the chat API as below: (see [OpenAI API protocol document](https://platform.openai.com/docs/guides/vision/uploading-base-64-encoded-images) for more details):
```python
import base64
from openai import OpenAI
# Set OpenAI's API key and API base to use vLLM's API server.
openai_api_key = "EMPTY"
openai_api_base = "http://localhost:8000/v1"
client = OpenAI(
api_key=openai_api_key,
base_url=openai_api_base,
)
image_path = "/path/to/local/image.png"
with open(image_path, "rb") as f:
encoded_image = base64.b64encode(f.read())
encoded_image_text = encoded_image.decode("utf-8")
base64_qwen = f"data:image;base64,{encoded_image_text}"
chat_response = client.chat.completions.create(
model="caprl",
messages=[
{"role": "system", "content": "You are a helpful assistant."},
{
"role": "user",
"content": [
{
"type": "image_url",
"image_url": {
"url": base64_qwen
},
},
{"type": "text", "text": "What is the text in the illustrate?"},
],
},
],
temperature=1.0,
max_tokens=max_tokens,
top_p=1.0,
extra_body={
"repetition_penalty": 1.0,
},
)
print("Chat response:", chat_response)
```
## Cases
<p align="center">
<img src="./assets/comparison.png" alt="Main Results on GPT2" width="750"/>
</p>
<p align="center">
<img src="./assets/info_caprl.png" alt="Main Results on GPT2" width="750"/>
</p>
<p align="center">
<img src="./assets/info_caprl2.png" alt="Main Results on GPT2" width="750"/>
</p>
<p align="center">
<img src="./assets/natural_caprl.png" alt="Main Results on GPT2" width="750"/>
</p>