This is a multimodal large language model fine-tuned from Qwen2.5-VL using our innovative Curr-ReFT methodology. The model has undergone a two-stage training process: first through Curriculum Reinforcement Learning, which gradually increases task complexity, followed by Rejected Sample based Self-improvement to maintain foundational capabilities. The model significantly enhances vision-language understanding and reasoning capabilities, making it exceptionally well-suited for complex tasks such as visual reasoning, detailed image understanding, and multimodal problem-solving. With its robust ability to perform sophisticated multimodal reasoning, Curr-ReFT emerges as a powerful AI assistant capable of addressing a wide range of challenges across diverse domains with improved accuracy and contextual awareness.

Training Configuration

Framework: The training process uses the open-source R1-V library, with Qwen2.5-VL-Instruct as the base model. This model comes in three variants: 3B, 7B.

The training configuration for grpo is as follows:

python

max_pixels 401408
per_device_train_batch_size: 1
gradient_accumulation_steps: 1
learning_rate: 1.0e-5

num_train_epochs: 1.0
lr_scheduler_type: cosine
bf16: true
flash_attn: fa2

Usage

You can load the model using the Hugging Face transformers library:

python

from transformers import AutoProcessor, Qwen2_5_VLForConditionalGeneration
import torch
from qwen_vl_utils import process_vision_info

MODEL_ID = "Curr-ReFT-3B"
processor = AutoProcessor.from_pretrained(MODEL_ID, trust_remote_code=True)
model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    MODEL_ID,
    trust_remote_code=True,
    torch_dtype=torch.bfloat16
).to("cuda").eval()

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "<your image path>"},
            {"type": "text", "text": "Hint: Please answer the question and provide the final answer at the end. Question: Which number do you have to write in the last daisy?"},
        ],
    }
]

# 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",
)
inputs = inputs.to(model.device)

generated_ids = model.generate(**inputs, max_new_tokens=4096)
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)

Institution

ZTE-AIM
University of Science and Technology of China

ZTE-AIM/3B-Curr-ReFT

Curr-ReFT-data

Curr-ReFT-model

Model Overview

Training Configuration

Usage

Institution

Model Contact

ZTE-AIM/
3B-Curr-ReFT