--- title: Sa2VA-26B canonical_url: "https://www.modelscope.cn/models/bytedance-community/Sa2VA-26B" md_url: "https://www.modelscope.cn/models/bytedance-community/Sa2VA-26B.md" repository: bytedance-community/Sa2VA-26B last_updated: 2026-06-05 license: apache-2.0 pipeline_tag: image-text-to-text tasks: - image-text-to-text model_type: - sa2va_chat architectures: - Sa2VAChatModel base_model: - OpenGVLab/InternVL2.5-26B base_model_relation: merge parameters: 25.8B tensor_type: - F32 - BF16 library_name: - pytorch - transformer - safetensors frameworks: - pytorch language: - multilingual downloads: 14 stars: 0 tags: - Sa2VA - custom_code --- # Sa2VA-26B > Sa2VA-26B - bytedance-community 在 ModelScope 开源的模型。Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos bytedance-community/Sa2VA-26B 是 ModelScope 魔搭社区上的 25.8B 参数image-text-to-text模型，采用 apache-2.0 许可，基于 OpenGVLab/InternVL2.5-26B 构建。 - **Repository**: bytedance-community/Sa2VA-26B - **License**: apache-2.0 - **Tasks**: image-text-to-text - **Parameters**: 25.8B - **Base model**: OpenGVLab/InternVL2.5-26B - **Tags**: Sa2VA, custom_code - **Downloads**: 14 - **Stars**: 0 - **Last updated**: 2026-06-05 Source: https://www.modelscope.cn/models/bytedance-community/Sa2VA-26B --- # Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos [\[📂 GitHub\]](https://github.com/magic-research/Sa2VA) [\[📜 Sa2VA paper\]](https://arxiv.org/abs/2501.04001) [\[🚀 Quick Start\]](#quick-start) ## Introduction Sa2VA is an MLLM capable of question answering, visual prompt understanding, and dense object segmentation at both image and video levels. It achieves comparable performance to SOTA MLLMs Qwen2-VL and InternVL2.5 on question-answering benchmarks. Additionally, Sa2VA possesses the visual prompt understanding and dense object segmentation capabilities that SOTA MLLMs Qwen2-VL and InternVL2.5 lack. Sa2VA achieves SOTA performance on both image and video grounding and segmentation benchmarks. ## Sa2VA Family We built the Sa2VA series based on Qwen2-VL and InternVL2/2.5. In the following table, we provide some Sa2VA models built on InternVL2.5. Other Sa2VA models will be open-sourced soon. | Model Name | Base MLLM | Language Part | HF Link | |:----------:|:------------------------------------------------------------------:|:---------------------------------------------------------------------------:|:-----------------------------------------------------:| | Sa2VA-1B | [InternVL2.5-1B](https://huggingface.co/OpenGVLab/InternVL2_5-1B) | [Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct) | [🤗 link](https://huggingface.co/ByteDance/Sa2VA-1B) | | Sa2VA-4B | [InternVL2.5-4B](https://huggingface.co/OpenGVLab/InternVL2_5-4B) | [Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct) | [🤗 link](https://huggingface.co/ByteDance/Sa2VA-4B) | | Sa2VA-8B | [InternVL2.5-8B](https://huggingface.co/OpenGVLab/InternVL2_5-8B) | [internlm2_5-7b-chat](https://huggingface.co/internlm/internlm2_5-7b-chat) | [🤗 link](https://huggingface.co/ByteDance/Sa2VA-8B) | | Sa2VA-26B | [InternVL2.5-26B](https://huggingface.co/OpenGVLab/InternVL2_5-26B) | [internlm2_5-20b-chat](https://huggingface.co/internlm/internlm2_5-20b-chat) | [🤗 link](https://huggingface.co/ByteDance/Sa2VA-26B) | ## Sa2VA Performance | Model Name | MME | MMBench | RefCOCO | RefCOCO+ | RefCOCOg | MeVIS (val_u) | DAVIS | |:----------:|:--------:|:----:|:-------:|:--------:|:--------:|:-------------:|:-----:| | Sa2VA-1B | 1504/434 | 71.9 | 79.6 | 73.6 | 77.7 | 53.4 | 69.5 | | Sa2VA-4B | 1691/610 | 81.8 | 82.4 | 77.6 | 79.7 | 55.9 | 73.7 | | Sa2VA-8B | 1690/610 | 84.4 | 82.6 | 78.0 | 80.3 | 58.9 | 75.9 | | Sa2VA-26B | 1698/653 | 85.8 | 82.9 | 79.3 | 81.2 | 61.8 | 78.6 | ## Quick Start We provide an example code to run `Sa2VA` using `transformers`. ```python import torch from transformers import AutoTokenizer, AutoModel from PIL import Image import numpy as np import os def get_rank_and_world_size(): rank = int(os.environ.get('RANK', 0)) world_size = int(os.environ.get('WORLD_SIZE', 1)) return rank, world_size def split_model(model_name): import math device_map = {} num_gpus = torch.cuda.device_count() rank, world_size = get_rank_and_world_size() num_gpus = num_gpus // world_size num_layers = {'Sa2VA-8B': 32, 'Sa2VA-26B': 48, 'Sa2VA-38B': 64, 'Sa2VA-78B': 80}[model_name] # Since the first GPU will be used for ViT, treat it as 0.8 GPU. num_layers_per_gpu = math.ceil(num_layers / (num_gpus - 0.2)) num_layers_per_gpu = [num_layers_per_gpu] * num_gpus num_layers_per_gpu[0] = math.ceil(num_layers_per_gpu[0] * 0.8) layer_cnt = 0 for i, num_layer in enumerate(num_layers_per_gpu): for j in range(num_layer): device_map[f'language_model.model.layers.{layer_cnt}'] = rank + world_size * i layer_cnt += 1 device_map['vision_model'] = rank device_map['mlp1'] = rank device_map['language_model.model.tok_embeddings'] = rank device_map['language_model.model.embed_tokens'] = rank device_map['language_model.output'] = rank device_map['language_model.model.norm'] = rank device_map['language_model.lm_head'] = rank device_map[f'language_model.model.layers.{num_layers - 1}'] = rank device_map['grounding_encoder'] = rank device_map['text_hidden_fcs'] = rank return device_map # load the model and tokenizer path = "ByteDance/Sa2VA-26B" device_map = split_model("Sa2VA-26B") model = AutoModel.from_pretrained( path, torch_dtype=torch.bfloat16, low_cpu_mem_usage=True, use_flash_attn=True, trust_remote_code=True, device_map=device_map, ).eval() tokenizer = AutoTokenizer.from_pretrained(path, trust_remote_code=True, use_fast=False) # for image chat image_path = "/PATH/TO/IMAGE" text_prompts = "Please describe the image." image = Image.open(image_path).convert('RGB') input_dict = { 'image': image, 'text': text_prompts, 'past_text': '', 'mask_prompts': None, 'tokenizer': tokenizer, } return_dict = model.predict_forward(**input_dict) answer = return_dict["prediction"] # the text format answer # for image chat with segmentation output image_path = "/PATH/TO/IMAGE" text_prompts = "Could you please give me a brief description of the image? Please respond with interleaved segmentation masks for the corresponding parts of the answer." image = Image.open(image_path).convert('RGB') input_dict = { 'image': image, 'text': text_prompts, 'past_text': '', 'mask_prompts': None, 'tokenizer': tokenizer, } return_dict = model.predict_forward(**input_dict) answer = return_dict["prediction"] # the text format answer masks = return_dict['prediction_masks'] # segmentation masks, list(np.array(1, h, w), ...) # for chat with visual prompt (mask format) input mask_prompts = np.load('/PATH/TO/pred_masks.npy') # np.array(n_prompts, h, w) image_path = "/PATH/TO/IMAGE" text_prompts = "Can you provide me with a detailed description of the region in the picture marked by region1." image = Image.open(image_path).convert('RGB') input_dict = { 'image': image, 'text': text_prompts, 'past_text': '', 'mask_prompts': mask_prompts, 'tokenizer': tokenizer, } return_dict = model.predict_forward(**input_dict) answer = return_dict["prediction"] # the text format answer # for video chat video_folder = "/PATH/TO/VIDEO_FOLDER" images_paths = os.listdir(video_folder) images_paths = [os.path.join(video_folder, image_path) for image_name in images_paths] if len(images_paths) > 5: # uniformly sample 5 frames step = (len(images_paths) - 1) // (5 - 1) images_paths = [images_paths[0]] + images_paths[1:-1][::step][1:] + [images_paths[-1]] text_prompts = "Please describe the video." input_dict = { 'video': images_paths, 'text': text_prompts, 'past_text': '', 'mask_prompts': None, 'tokenizer': tokenizer, } return_dict = model.predict_forward(**input_dict) answer = return_dict["prediction"] # the text format answer # for video chat with segmentation mask output video_folder = "/PATH/TO/VIDEO_FOLDER" images_paths = os.listdir(video_folder) images_paths = [os.path.join(video_folder, image_path) for image_name in images_paths] text_prompts = "Please segment the person." input_dict = { 'video': images_paths, 'text': text_prompts, 'past_text': '', 'mask_prompts': None, 'tokenizer': tokenizer, } return_dict = model.predict_forward(**input_dict) answer = return_dict["prediction"] # the text format answer masks = return_dict['prediction_masks'] # segmentation masks, list(np.array(n_frames, h, w), ...) ``` ## Citation If you find this project useful in your research, please consider citing: ```BibTeX @article{sa2va, title={Sa2VA: Marrying SAM2 with LLaVA for Dense Grounded Understanding of Images and Videos}, author={Yuan, Haobo and Li, Xiangtai and Zhang, Tao and Huang, Zilong Huang and Xu, Shilin and Ji, Shunping and Tong, Yunhai and Qi, Lu and Feng, Jiashi and Yang, Ming-Hsuan}, journal={arXiv preprint}, year={2025} } ```