Skip to content

fix bug in qwen3_vl eagle3 speculative decoding#176

Merged
irisliu10 merged 1 commit into
Tencent:mainfrom
irisliu10:qwen3_vl_speculative_decoding
Dec 15, 2025
Merged

fix bug in qwen3_vl eagle3 speculative decoding#176
irisliu10 merged 1 commit into
Tencent:mainfrom
irisliu10:qwen3_vl_speculative_decoding

Conversation

@irisliu10
Copy link
Copy Markdown
Collaborator

Bug fix:

  1. target model 采集hidden_states 信息 model.forward() 加入 pixel_values 数据
  2. draft mode inputs_embeds 通过 input_ids 计算得到
  3. 移除 attention_mask 移位操作
  4. cache_hidden 参数对齐
  5. 训练速度慢问题:load_model device 修改-> decide_device_for_distributed

@irisliu10 irisliu10 merged commit 819a0c1 into Tencent:main Dec 15, 2025
5 checks passed
dawnranger pushed a commit to dawnranger/AngelSlim that referenced this pull request Mar 11, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants