GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
BAAI
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
BAAI
Model Tree
Qwen/Qwen3-VL-4B-Instruct
Input Modalities
Output Modalities
Supported Functionality
This repository contains the reasoning model only. The retrieval-augmented scene-reconstruction extension is released separately.
| Property | Value |
|---|---|
| Model name | Recon2Reason Reasoning 4B |
| Model type | Vision-language conditional generation model |
| Architecture | Qwen3VLForConditionalGeneration |
| Parameters | 4,437,815,808 |
| Weight dtype | BF16 |
| Weight format | Safetensors, 2 shards |
| Primary domain | Indoor spatial reasoning |
| Base model | Qwen3-VL-4B-Instruct |
| Tested Transformers version | 4.57.1 |
| License | Apache-2.0 |
bash
pip install "transformers==4.57.1" "torch>=2.6" accelerate safetensors pillowpython
import torchfrom transformers import AutoProcessor, Qwen3VLForConditionalGeneration model_id = "BAAI/Recon2Reason-Reasoning-4B" processor = AutoProcessor.from_pretrained(model_id)model = Qwen3VLForConditionalGeneration.from_pretrained( model_id, dtype="auto", device_map="auto",).eval() messages = [ { "role": "user", "content": [ {"type": "image", "image": "path/to/scene.jpg"}, { "type": "text", "text": "Which object is closest to the chair? ", }, ], }] inputs = processor.apply_chat_template( messages, tokenize=True, add_generation_prompt=True, return_dict=True, return_tensors="pt",).to(model.device) with torch.inference_mode(): output_ids = model.generate( **inputs, max_new_tokens=512, do_sample=False, ) generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]answer = processor.batch_decode( generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False,)[0]print(answer)The model uses the standard Qwen3-VL chat format. For benchmark reproduction, use greedy decoding (do_sample=False) unless a benchmark specifies otherwise.
The checkpoint was evaluated in BF16 with PyTorch 2.8.0, Transformers 4.57.1, SDPA attention, and greedy decoding on NVIDIA RTX PRO 6000 Blackwell GPUs.
The model is released under the Apache License 2.0, subject to final confirmation that all training data and upstream artifacts permit this distribution. See LICENSE.
This work builds on Qwen3-VL. We thank the creators of the evaluation datasets and the open-source Transformers ecosystem.
pip install "transformers==4.57.1" "torch>=2.6" accelerate safetensors pillowimport torchfrom transformers import AutoProcessor, Qwen3VLForConditionalGeneration model_id = "BAAI/Recon2Reason-Reasoning-4B" processor = AutoProcessor.from_pretrained(model_id)model = Qwen3VLForConditionalGeneration.from_pretrained( model_id, dtype="auto", device_map="auto",).eval() messages = [ { "role": "user", "content": [ {"type": "image", "image": "path/to/scene.jpg"}, { "type": "text", "text": "Which object is closest to the chair? ", }, ], }] inputs = processor.apply_chat_template( messages, tokenize=True, add_generation_prompt=True, return_dict=True, return_tensors="pt",).to(model.device) with torch.inference_mode(): output_ids = model.generate( **inputs, max_new_tokens=512, do_sample=False, ) generated_ids = output_ids[:, inputs["input_ids"].shape[1]:]answer = processor.batch_decode( generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False,)[0]print(answer)