Add pipeline tag and links to paper, project page and code

#1
by nielsr HF Staff - opened
Files changed (1) hide show
  1. README.md +24 -14
README.md CHANGED
@@ -1,19 +1,20 @@
1
  ---
2
- license: apache-2.0
3
  base_model:
4
- - Qwen/Qwen3.5-9B
5
- inference: false
 
6
  tags:
7
- - mmflash
8
- - dflash
9
- - speculative-decoding
10
- - speculative-decoding-draft
11
- - block-diffusion
12
- - draft-model
13
- - multimodal
14
- - sglang
15
- - qwen
16
- - qwen3.5
 
17
  ---
18
 
19
  <div align="center">
@@ -27,6 +28,8 @@ MMFlash draft model for speculative decoding of [`Qwen/Qwen3.5-9B`](https://hugg
27
 
28
  MMFlash builds on [DFlash](https://github.com/z-lab/dflash) block-diffusion drafting: the draft reads hidden states from five layers of the target and proposes a block of 16 tokens in one forward pass, and the target verifies the block in one forward pass. MMFlash keeps the DFlash architecture and is trained on multimodal data with a loss weighted by per-token visual-grounding (visual-KL) scores.
29
 
 
 
30
  <div align="center">
31
  <video autoplay muted loop playsinline controls width="100%">
32
  <source src="https://huggingface.co/FSCCS/Qwen3.5-9B-MMFlash/resolve/main/assets/demo.mp4" type="video/mp4">
@@ -107,4 +110,11 @@ MMFlash at 1, 8 and 32 requests in flight, on a single NVIDIA H200. Each cell is
107
 
108
  ## Citation
109
 
110
- Coming soon.
 
 
 
 
 
 
 
 
1
  ---
 
2
  base_model:
3
+ - Qwen/Qwen3.5-9B
4
+ license: apache-2.0
5
+ pipeline_tag: image-text-to-text
6
  tags:
7
+ - mmflash
8
+ - dflash
9
+ - speculative-decoding
10
+ - speculative-decoding-draft
11
+ - block-diffusion
12
+ - draft-model
13
+ - multimodal
14
+ - sglang
15
+ - qwen
16
+ - qwen3.5
17
+ inference: false
18
  ---
19
 
20
  <div align="center">
 
28
 
29
  MMFlash builds on [DFlash](https://github.com/z-lab/dflash) block-diffusion drafting: the draft reads hidden states from five layers of the target and proposes a block of 16 tokens in one forward pass, and the target verifies the block in one forward pass. MMFlash keeps the DFlash architecture and is trained on multimodal data with a loss weighted by per-token visual-grounding (visual-KL) scores.
30
 
31
+ This checkpoint is presented in the paper [MMFlash: Faster Speculative Decoding for Vision-Language Models](https://huggingface.co/papers/2611.18482). Check out the [project page](https://fscdc.github.io/MMFlash/) and the [GitHub repository](https://github.com/fscdc/MMFlash) for code, training scripts and evaluation details.
32
+
33
  <div align="center">
34
  <video autoplay muted loop playsinline controls width="100%">
35
  <source src="https://huggingface.co/FSCCS/Qwen3.5-9B-MMFlash/resolve/main/assets/demo.mp4" type="video/mp4">
 
110
 
111
  ## Citation
112
 
113
+ ```bibtex
114
+ @article{feng2026mmflash,
115
+ title = {MMFlash: Faster Speculative Decoding for Vision-Language Models},
116
+ author = {Feng, Sicheng and Chen, Zigeng and Wang, Song and Fang, Gongfan and Wang, Xinchao},
117
+ journal = {arXiv preprint arXiv:2611.18482},
118
+ year = {2026}
119
+ }
120
+ ```