Update README: benchmark positioning and citations

README.md

@@ -6,20 +6,23 @@ base_model:
 - Qwen/Qwen2.5-VL-7B-Instruct
 ---
 
-# ViSpec: Accelerating Vision-Language Models with Vision-Aware Speculative Decoding
+# ViSpec-Qwen2.5-VL-7B-Instruct (Benchmark Release)
 
-*Jialiang Kang, Han Shu, Wenshuo Li, Yingjie Zhai, Xinghao Chen*
+This model repo is part of a **multimodal speculative decoding benchmark suite**.
 
-<a href="https://arxiv.org/abs/2509.15235"><img src="https://img.shields.io/static/v1?label=arXiv&message=Paper&color=red&logo=arxiv"></a>
-<a href="https://github.com/KangJialiang/ViSpec"><img src="https://img.shields.io/static/v1?label=GitHub&message=Code&color=blue&logo=github"></a>
+## Why this repo exists
 
-## Overview
+We maintain a unified benchmark codebase that includes multiple methods (Baseline, EAGLE, EAGLE2, Lookahead, MSD, ViSpec) so users can run training/evaluation more easily under one setup.
 
-Speculative decoding is a widely adopted technique for accelerating inference in large language models (LLMs), yet its application to vision-language models (VLMs) remains underexplored, with existing methods achieving only modest speedups ($<1.5\times$). This gap is increasingly significant as multimodal capabilities become central to large-scale models. We hypothesize that large VLMs can effectively filter redundant image information layer by layer without compromising textual comprehension, whereas smaller draft models struggle to do so. To address this, we introduce **Vision-Aware Speculative Decoding (ViSpec)**, a novel framework tailored for VLMs. ViSpec employs a lightweight vision adaptor module to compress image tokens into a compact representation, which is seamlessly integrated into the draft model's attention mechanism while preserving original image positional information. Additionally, we extract a global feature vector for each input image and augment all subsequent text tokens with this feature to enhance multimodal coherence. To overcome the scarcity of multimodal datasets with long assistant responses, we curate a specialized training dataset by repurposing existing datasets and generating extended outputs using the target VLM with modified prompts. Our training strategy mitigates the risk of the draft model exploiting direct access to the target model's hidden states, which could otherwise lead to shortcut learning when training solely on target model outputs. Extensive experiments validate ViSpec, achieving, to our knowledge, the first substantial speedup in VLM speculative decoding.
+- The methods are aggregated here for **user convenience** (shared dataset format, scripts, and metrics).
+- The original ideas and implementations belong to their respective authors.
+- This specific Hugging Face repo hosts the **ViSpec-Qwen2.5-VL-7B-Instruct checkpoint** used in our benchmark runs.
 
 ## Citation
 
-If you find our work useful, please consider citing:
+If you use this checkpoint and benchmark, please cite ViSpec and the original methods you compare against.
+
+### ViSpec
 
 ```bibtex
 @inproceedings{vispec,
@@ -28,4 +31,40 @@ If you find our work useful, please consider citing:
   booktitle={Annual Conference on Neural Information Processing Systems},
   year={2025}
 }
-```
+```
+
+### EAGLE / EAGLE2 / EAGLE3
+
+```bibtex
+@inproceedings{li2024eagle,
+  author = {Yuhui Li and Fangyun Wei and Chao Zhang and Hongyang Zhang},
+  title = {{EAGLE}: Speculative Sampling Requires Rethinking Feature Uncertainty},
+  booktitle = {International Conference on Machine Learning},
+  year = {2024}
+}
+
+@inproceedings{li2024eagle2,
+  author = {Yuhui Li and Fangyun Wei and Chao Zhang and Hongyang Zhang},
+  title = {{EAGLE-2}: Faster Inference of Language Models with Dynamic Draft Trees},
+  booktitle = {Empirical Methods in Natural Language Processing},
+  year = {2024}
+}
+
+@inproceedings{li2025eagle3,
+  author = {Yuhui Li and Fangyun Wei and Chao Zhang and Hongyang Zhang},
+  title = {{EAGLE-3}: Scaling up Inference Acceleration of Large Language Models via Training-Time Test},
+  booktitle = {Annual Conference on Neural Information Processing Systems},
+  year = {2025}
+}
+```
+
+### Other integrated baselines (links)
+
+- Lookahead Decoding: https://lmsys.org/blog/2023-11-21-lookahead-decoding/
+- MSD-LLaVA1.5-7B: https://huggingface.co/lucylyn/MSD-LLaVA1.5-7B
+- Medusa: https://github.com/FasterDecoding/Medusa
+
+## Notes
+
+- This model card focuses on benchmark usage and attribution.
+- For full benchmark code and scripts, please refer to the benchmark repository used in your experiment setup.