Issues

• EAGLE-2：Clarification on the "w/o value" Ablation in Table 3 and Draft Tree Connectivity

  SafeAILab/EAGLE#343 · Filter by author RanceChen06 opened on Oct 10, 2026

• loss mean instead of dividing by loss mask.sum()

  SafeAILab/EAGLE#340 · Filter by author Manjunath679 opened on Jun 9, 2026
  1

• [Discussion] Performance characteristics of EAGLE 3.1 (Post-norm) at shallow speculative depths ($k \le 3$)

  SafeAILab/EAGLE#339 · Filter by author changerjin opened on Jun 5, 2026
  3

• Does shifting by one position when constructing the input break the causal chain? For example: input_ids_target = input_ids[:, 1:]

  SafeAILab/EAGLE#335 · Filter by author fanbinqi opened on Apr 13, 2026
  1

• Garbled text generated when using qwen3 as inference in gen_ea_answer.py

  SafeAILab/EAGLE#334 · Filter by author Rachelcoll opened on Mar 30, 2026
  2

• 📋 Documentation Enhancement Suggestion

  SafeAILab/EAGLE#333 · Filter by author croviatrust opened on Mar 28, 2026
  2

• Support for multi-modal MTP？

  SafeAILab/EAGLE#331 · Filter by author 1benwu1 opened on Feb 28, 2026
  1

• Non‑greedy rejection sampling ignores draft (q); not lossless per Speculative Decoding

  SafeAILab/EAGLE#329 · Filter by author ethantsliu opened on Jan 31, 2026
  3

• Why is padding for target and input_ids performed before the for loop in traineagle3/cnets?

  SafeAILab/EAGLE#325 · Filter by author LuoKan00 opened on Jan 26, 2026
  1

• Does EAGLE3 support offloading?

  SafeAILab/EAGLE#322 · Filter by author zszdsze opened on Jan 12, 2026

• loss_mask not adjusted after target shift, causing loss on invalid padding

  SafeAILab/EAGLE#317 · Filter by author oluxembourg opened on Dec 22, 2025

• ImportError: transformers.utils.LossKwargs removed in transformers>=4.54 breaks import in modeling_qwen3_kv.py

  SafeAILab/EAGLE#306 · Filter by author timxu826 opened on Oct 11, 2025
  2

• Some thoughts on tree decoding

  SafeAILab/EAGLE#301 · Filter by author ggg-s opened on Sep 18, 2025

• Question regarding new_token in naivegenerate

  SafeAILab/EAGLE#296 · Filter by author aladinggit opened on Sep 11, 2025
  1

• Matrix shape not match with Qwen3-8B and Tengyunw/qwen3_8b_eagle3

  SafeAILab/EAGLE#294 · Filter by author wyattxuanyang opened on Sep 9, 2025
  2

• next token embedding was not used

  SafeAILab/EAGLE#293 · Filter by author jos-fisher-jejgkej opened on Sep 9, 2025

• Questions about eagenerate

  SafeAILab/EAGLE#288 · Filter by author ebubekir-pulat opened on Sep 2, 2025

• 没找到Eagle2对应DeepSeek-R1-Distill-LLaMA-8B的权重文件下载链接

  SafeAILab/EAGLE#287 · Filter by author minaduki-sora opened on Aug 29, 2025

• Training loss does not decrease

  SafeAILab/EAGLE#286 · Filter by author fan-niu opened on Aug 21, 2025
  3

• sglang 运行AngelSlim/Qwen3-4B_eagle3 问题

  SafeAILab/EAGLE#284 · Filter by author cicijohn1983 opened on Aug 19, 2025
  1

• 配环境问题

  SafeAILab/EAGLE#277 · Filter by author Ren-ww opened on Aug 12, 2025
  2

• Qwen2SdpaAttention中past_key_value = None，KV Cache 没有被正确更新和返回吗

  SafeAILab/EAGLE#270 · Filter by author slcccc opened on Jul 31, 2025
  1

• utils_alpha.py

  SafeAILab/EAGLE#253 · Filter by author Duze204 opened on Jul 6, 2025

• EAGLE-3 Training and Test Data

  SafeAILab/EAGLE#251 · Filter by author ebubekir-pulat opened on Jul 1, 2025
  21

• Batch size >1 execution fails

  SafeAILab/EAGLE#250 · Filter by author Lena-Jurkschat opened on Jul 1, 2025
  1