EVAN-3B

Built with Qwen. Released under the Qwen Research License, for non-commercial use only.

EVAN-3B is Qwen2.5-VL-3B-Instruct trained with EVAN (EVidence-Ablated Negatives), the method of our NeurIPS 2026 paper Grounding Multimodal Reasoning with Evidence-Ablated Negatives.

Paper: arXiv:2610.39237 | Code: github.com/DawnLIU35/EVAN | Models: EVAN-3B · EVAN-7B · EVAN-8B

Method

RLVR rewards typically check only the final answer, not whether it came from the image or from language priors and dataset shortcuts. EVAN masks the visual evidence a question needs, lets the current policy answer without it, and penalizes plausible wrong answers under the clean image. EVAN-3B is trained with DAPO on ViRL39K. The GitHub repository has the full method, data pipeline, and training scripts.

Results

Accuracy (%) from Table 1 of the paper, with sampling at temperature 1.0. DAPO uses the same data and training recipe without EVAN.

Model Geo3K MathVista MathVerse We-Math DynaMath MathVision MMK12 MMMU-Pro LogicVista Avg.
Qwen2.5-VL-3B-Instruct 19.4 46.3 33.5 29.9 40.4 18.2 36.3 19.8 32.1 30.7
DAPO 38.8 62.4 60.4 60.5 59.3 26.5 72.2 29.9 43.0 50.3
EVAN-3B 39.9 62.7 62.8 64.6 60.6 27.6 71.5 31.0 43.3 51.6

Usage

Requires transformers>=4.57, accelerate, and torchvision. Add the instruction below after the question, as in training. The default generation config holds the sampling settings we used for evaluation. With vLLM, also pass top_k=50 in SamplingParams: vLLM skips generation-config values that equal the transformers defaults.

from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "DawnLiu/EVAN-3B"
model = AutoModelForImageTextToText.from_pretrained(model_id, dtype="auto", device_map="auto")
processor = AutoProcessor.from_pretrained(model_id)

question = "Which object is the largest?"
prompt = (
    f"{question}\n\n"
    "You first think through the reasoning process as an internal monologue, enclosed within <think> </think> tags. Then, provide your final answer enclosed within \\boxed{}."
)
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "path/to/image.png"},
            {"type": "text", "text": prompt},
        ],
    }
]
inputs = processor.apply_chat_template(
    messages, tokenize=True, add_generation_prompt=True, return_dict=True, return_tensors="pt"
).to(model.device)
output = model.generate(**inputs)
print(processor.decode(output[0, inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Citation

@misc{liu2026grounding,
  title         = {Grounding Multimodal Reasoning with Evidence-Ablated Negatives},
  author        = {Chenxi Liu and Tianyi Xiong and Yanshuo Chen and Heng Huang},
  year          = {2026},
  eprint        = {2610.39237},
  archivePrefix = {arXiv},
  url           = {https://arxiv.org/abs/2610.39237},
}

License

EVAN-3B is released under the Qwen Research License of Qwen2.5-VL-3B-Instruct, which permits non-commercial use only. Relative to the base model, the weights were changed by RL training with EVAN, generation_config.json sets our evaluation sampling defaults, and the other configuration, tokenizer, and processor files were re-saved (with dtype set to bfloat16 in config.json).

Downloads last month
121
Safetensors
Model size
4B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for DawnLiu/EVAN-3B

Finetuned
(940)
this model
Quantizations
3 models

Dataset used to train DawnLiu/EVAN-3B

Paper for DawnLiu/EVAN-3B