Issues

• Does Medusa2 support llama2?

  FasterDecoding/Medusa#141 · Filter by author zszdsze opened on Jan 6, 2026

• The legacy Medusa Head structure is inconsistent with the new one.

  FasterDecoding/Medusa#140 · Filter by author Jianhua-Cui opened on Oct 11, 2025
  1

• loss value nan

  FasterDecoding/Medusa#139 · Filter by author wittycheng opened on Aug 22, 2025

• why什么使用连续提前开辟好的KVcache? 这样本身就引入一个比huggingface实现快的因素？

  FasterDecoding/Medusa#137 · Filter by author wenhaoli-xmu opened on Jun 19, 2025

• why my train medusa head result is {'medusa0_top1': nan, 'medusa0_loss': nan, 'medusa1_top1': nan, 'medusa1_loss': nan, 'medusa2_top1': nan, 'medusa2_loss': nan, 'epoch': 0}

  FasterDecoding/Medusa#134 · Filter by author Mewo518 opened on Mar 22, 2025
  1

• Support for other types of LLM

  FasterDecoding/Medusa#132 · Filter by author Shubin-vadim opened on Mar 4, 2025

• Is Medusa(-2) compatible with vision language models (VLMs) ?

  FasterDecoding/Medusa#129 · Filter by author MoritzLaurer opened on Dec 3, 2024

• Question about the Tree Attention Mechanism

  FasterDecoding/Medusa#127 · Filter by author chansonzhang opened on Nov 18, 2024

• About Code compatability

  FasterDecoding/Medusa#126 · Filter by author kimjoohyungsd opened on Nov 1, 2024

• Ask for data recipe to reproduce Medusa-2

  FasterDecoding/Medusa#125 · Filter by author Achazwl opened on Oct 24, 2024

• [report bug] Encountered when inferencing with Mistral models

  FasterDecoding/Medusa#124 · Filter by author shrango opened on Oct 10, 2024

• How to run inference of a baseline model without medusa support?

  FasterDecoding/Medusa#123 · Filter by author kailashg26 opened on Sep 16, 2024

• About the Tree Sparsity

  FasterDecoding/Medusa#122 · Filter by author PineTreeWss opened on Sep 13, 2024

• [Retraining] Use Liger Kernel to avoid multi-head logits materialization and scale the context length by N times

  FasterDecoding/Medusa#119 · Filter by author ByronHsu opened on Aug 26, 2024
  1

• Training code is not working

  FasterDecoding/Medusa#118 · Filter by author ksajan opened on Aug 25, 2024
  2

• Instruct data format

  FasterDecoding/Medusa#117 · Filter by author orhan6116 opened on Aug 13, 2024

• jinja2.exceptions.UndefinedError: dict object has no element 0

  FasterDecoding/Medusa#115 · Filter by author LLLL114 opened on Jul 30, 2024
  2

• updated medusa models in huggingface?

  FasterDecoding/Medusa#114 · Filter by author hustxiayang opened on Jul 25, 2024

• [ISSUE] The Pull Request at https://github.com/FasterDecoding/Medusa/pull/97 from Narsil/medusa2 needs to be rolled back.

  FasterDecoding/Medusa#112 · Filter by author super-ahn opened on Jul 11, 2024

• Does Medusa support beam search decoding strategy?

  FasterDecoding/Medusa#108 · Filter by author xs229 opened on Jun 14, 2024

• The implementation of stage 2 with axolotl

  FasterDecoding/Medusa#107 · Filter by author boxiaowave opened on May 24, 2024

• PPL compute

  FasterDecoding/Medusa#106 · Filter by author yuyangxie96 opened on May 23, 2024

• Containerization with Dockerfile to setup medusa

  FasterDecoding/Medusa#104 · Filter by author nguptaHQ opened on May 16, 2024
  1

• Conversation roles must alternate user/assistant/user/assistant/

  FasterDecoding/Medusa#102 · Filter by author nguptaHQ opened on May 13, 2024

• [bug] fix preprocess function

  FasterDecoding/Medusa#101 · Filter by author xiezipeng-ML opened on Apr 28, 2024