TBAC-VLR1-7B

Overview

This is a multimodal language model fine-tuned by Tencent PCG Basic Algorithm Center. Based on Qwen2.5-VL-7B-Instruct, TBAC-VLR1-7B-SFT undergoes SFT training using 40k sft data filtered from OpenR1-Math-220k. TBAC-VLR1-3B then employs GRPO (Group Relative Policy Optimization) and adapts Clip-Higher from DAPO, achieving strong performance on several multimodal reasoning benchmarks among models of the same size.

Performance

Model	Average	MathVista	MathVision	MathVerse	DynaMath	LogicVista
Qwen2.5-VL-7B	40.5	68.0	25.7	45.5	21.8	41.2
VLAA-Thinker-Qwen2.5-7B	42.7	68.0	26.4	48.2	22.4	48.5
VL-Rethinker-7B	41.8	73.7	28.4	46.4	17.8	42.7
TBAC-VLR1-7B-RL	41.3	70.1	25.4	43.4	19.0	48.4
TBAC-VLR1-7B-SFT	41.8	65.1	28.5	49.1	20.6	45.5
TBAC-VLR1-7B	43.4	66.7	31.4	50.1	22.6	46.4

The results of our model are self-reported, obtained by running evaluations offline on each benchmark.

Usage

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

model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
    "TencentBAC/TBAC-VLR1-7B", torch_dtype="auto", device_map="auto"
)

processor = AutoProcessor.from_pretrained("TencentBAC/TBAC-VLR1-7B")

messages = [
    {
        "role": "system",
        "content": "You FIRST think about the reasoning process as an internal monologue and then provide the final answer. The reasoning process MUST BE enclosed within <think> </think> tags. The final answer MUST BE put in \\boxed{}."
    },
    {
        "role": "user",
        "content": [
            {
                "type": "image",
                "image": image_path,
            },
            {"type": "text", "text": query},
        ],
    }
]

# 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("cuda")

# Inference: Generation of the output
generated_ids = model.generate(**inputs, max_new_tokens=128, do_sample=False)
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)

Citation

If you find our model useful in your research, please consider giving ❤️ and citations. Thanks!

@misc{Ou2025TBACVLR1,
  title = {TBAC-VLR1-7B},
  author = {Ou, Linyu and Xu, Junzhe and Yin, Yuyang},
  year = {2025},
  url = {https://huggingface.co/TencentBAC/TBAC-VLR1-7B},
}

About