File size: 4,508 Bytes
55a5d66 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 | ---
license: apache-2.0
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- SAIL
---
# SAIL
[\[📂 GitHub\]](https://github.com/bytedance/SAIL)
[\[📜 paper\]](https://arxiv.org/abs/2504.10462)
[\[🚀 Quick Start\]](#quick-start)
## Introduction
SAIL is a **S**ingle tr**A**nsformer model for v**I**sion and **L**anguage. It is a unified multimodal large language model (MLLM) that seamlessly integrates raw pixel encoding and language decoding within a single architecture. **Without relying on pre-trained vision encoders**, SAIL achieves competitive performance across a wide range of vision-language tasks and demonstrates strong visual representation, rivaling state-of-the-art vision models in tasks like semantic segmentation.
## Model
| Model Name | HF Link |
|:----------:|:------------------------------------------------------------------:|
| SAIL-7B | [🤗 link](https://huggingface.co/ByteDance-Seed/SAIL-7B) |
## Quick Start
We provide an example code to run `SAIL`.
```python
from example import *
NON_VISION_TOKEN_ID = -1
PATH_TO_MODEL = "path to model"
PATH_TO_TOKENIZER = "path to tokenizer"
IMAGE_PATH = "path to image"
PROMPT = "content of prompt"
model, tokenizer = get_transformer_and_tokenizer(
PATH_TO_MODEL,
PATH_TO_TOKENIZER
)
model = model.cuda()
image_processor = lambda x: convert_image_base64_to_patches(load_image_to_base64(x), model.config.vision_patch_size, fix_res_size=None)
prompt_inp = tokenizer.bos_token + '[INST] {} [/INST]'.format(PROMPT)
image_path = IMAGE_PATH
image_patches = image_processor(image_path)
nh, nw = image_patches.shape[:2]
image_tokens, image_tokens_len = prepare_image_textual_seq_norowsep(nh, nw, tokenizer, add_cls=False)
input_tokens = image_tokens + prompt_inp
input_ids = tokenizer(input_tokens, add_special_tokens=False, return_tensors="pt").input_ids
vision_patch_indices = torch.full_like(input_ids, fill_value=NON_VISION_TOKEN_ID)
vision_patches = image_patches.view(nh * nw, -1)
assert (input_ids == tokenizer.vis_patch_tok_id).sum() == vision_patches.size(0)
assert (input_ids >= tokenizer.vis_beg_tok_id).sum() == image_tokens_len
vision_patch_indices[input_ids==tokenizer.vis_patch_tok_id] = torch.arange(vision_patches.size(0))
attention_mask = create_single_prefix_mask(image_tokens_len, input_ids.size(-1)).unsqueeze(0).unsqueeze(0)
position_ids = generate_mm_pos_ids_singleit(input_ids.squeeze(0).numpy().tolist(), tokenizer.vis_patch_tok_id, nh, nw).unsqueeze(1)
input_ids = input_ids.long().cuda()
vision_patch_indices = vision_patch_indices.long().cuda()
vision_patches = vision_patches.to(torch.bfloat16).cuda()
position_ids = position_ids.long().cuda()
attention_mask = attention_mask.cuda()
padding_attention_mask = torch.ones_like(input_ids).cuda()
inputs = dict(
input_ids = input_ids,
position_ids = position_ids,
attention_mask = padding_attention_mask,
vision_patches = vision_patches,
vision_patch_indices = vision_patch_indices,
use_cache=True
)
cached_inputs = dict(
input_ids = input_ids[:, :image_tokens_len],
position_ids = position_ids[:, :, :image_tokens_len],
attention_mask = attention_mask[:,:, :image_tokens_len, :image_tokens_len],
vision_patches = vision_patches,
vision_patch_indices = vision_patch_indices[:, :image_tokens_len],
use_cache=True
)
prefix_cache = DynamicCache()
with torch.no_grad():
prefix_cache = model.forward(**cached_inputs, past_key_values=prefix_cache).past_key_values
past_key_values = copy.deepcopy(prefix_cache)
generate_config = GenerationConfig(
max_new_tokens=1024,
return_dict_in_generate=True,
output_attentions=False
)
generated = model.generate(
**inputs,
past_key_values=past_key_values,
generation_config=generate_config
)
generated_ids = generated['sequences'][:, input_ids.size(1):]
response = tokenizer.batch_decode(
generated_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False
)[0]
print(f"\nModel Response: ===\n{response}\n===")
```
## Citation
If you find this project useful in your research, please consider citing:
```BibTeX
@article{lei2025sail,
title={The Scalability of Simplicity: Empirical Analysis of Vision-Language Learning with a Single Transformer},
author={Lei, Weixian and Wang, Jiacong and Wang, Haochen and Li, Xiangtai and Liew, Jun Hao and Feng, Jiashi and Huang, Zilong},
journal={arXiv preprint arXiv:2504.10462},
year={2025}
}
``` |