Skip to content

[Bug]: EAGLE speculative decoding crashes for multimodal wrapper architectures — load_eagle_model resolves lm_head from target_model instead of target_language_model #14256

[Bug]: EAGLE speculative decoding crashes for multimodal wrapper architectures — load_eagle_model resolves lm_head from target_model instead of target_language_model

[Bug]: EAGLE speculative decoding crashes for multimodal wrapper architectures — load_eagle_model resolves lm_head from target_model instead of target_language_model #14256

Triggered via issue July 15, 2026 06:24
Status Success
Total duration 7s
Artifacts

issue_autolabel.yml

on: issues
add-labels
3s
add-labels
Fit to window
Zoom out
Zoom in

Annotations

10 notices
add-labels
Processing label: cpu
add-labels
No matching terms found for label "rocm".
add-labels
Reason: No matching terms found
add-labels
Final decision: DO NOT ADD LABEL
add-labels
Searching in body (12772 characters)
add-labels
Searching in title: "[Bug]: EAGLE speculative decoding crashes for multimodal wrapper architectures — load_eagle_model resolves lm_head from target_model instead of target_language_model"
add-labels
Searching with 5 keywords, 12 substrings, and 1 regex patterns
add-labels
Issue Body length: 12772 characters
add-labels
Issue Title: "[Bug]: EAGLE speculative decoding crashes for multimodal wrapper architectures — load_eagle_model resolves lm_head from target_model instead of target_language_model"
add-labels
Processing label: rocm