Data

Tools

Indexes

Applications

Experiments

Issues

An open API service for providing issue and pull request metadata for open source projects.

GitHub / FasterDecoding/Medusa issues and pull requests

#141 - Does Medusa2 support llama2?

Issue - State: open - Opened by zszdsze about 2 months ago

#140 - The legacy Medusa Head structure is inconsistent with the new one.

Issue - State: open - Opened by Jianhua-Cui 5 months ago - 1 comment

#139 - loss value nan

Issue - State: open - Opened by wittycheng 6 months ago

#138 - Support for Dynamic Cache?

Issue - State: open - Opened by Yi-Yu-Yvonne 7 months ago

#136 - New adaptive

Pull Request - State: closed - Opened by arab7716 10 months ago

#136 - New adaptive

Pull Request - State: closed - Opened by arab7716 10 months ago

#135 - small fixes for convenience

Pull Request - State: closed - Opened by JackCharlesZhang 11 months ago

#135 - small fixes for convenience

Pull Request - State: closed - Opened by JackCharlesZhang 11 months ago

#133 - Fix sharing of resblock layers (from Liger-Kernel#269)

Pull Request - State: open - Opened by loreloc 12 months ago

#132 - Support for other types of LLM

Issue - State: open - Opened by Shubin-vadim 12 months ago

#131 - Replaced broken TGI link

Pull Request - State: open - Opened by buvnswrn about 1 year ago

#131 - Replaced broken TGI link

Pull Request - State: open - Opened by buvnswrn about 1 year ago

#130 - Llama3

Pull Request - State: closed - Opened by alex4321 about 1 year ago

#130 - Llama3

Pull Request - State: closed - Opened by alex4321 about 1 year ago

#127 - Question about the Tree Attention Mechanism

Issue - State: open - Opened by chansonzhang over 1 year ago

#126 - About Code compatability

Issue - State: open - Opened by kimjoohyungsd over 1 year ago

#125 - Ask for data recipe to reproduce Medusa-2

Issue - State: open - Opened by Achazwl over 1 year ago

#122 - About the Tree Sparsity

Issue - State: open - Opened by PineTreeWss over 1 year ago

#121 - is_flash_attn_available has been renamed in transformers.utils

Pull Request - State: open - Opened by simrathanspal over 1 year ago

#120 - Update medusa_introduction.ipynb

Pull Request - State: closed - Opened by simrathanspal over 1 year ago

#118 - Training code is not working

Issue - State: open - Opened by ksajan over 1 year ago

#117 - Instruct data format

Issue - State: open - Opened by orhan6116 over 1 year ago

#116 - Are Medusa Heads computed in parallel or serially?

Issue - State: open - Opened by userljz over 1 year ago

#111 - do you support Amd gpu -- rocm ??

Issue - State: closed - Opened by amd-maheshs3 over 1 year ago

#110 - Errors occurred during the environment and training

Issue - State: closed - Opened by blacker521 over 1 year ago - 2 comments

#109 - train_legacy.py: try to fix indices bug in preprocess.

Pull Request - State: open - Opened by k-l-lambda over 1 year ago

#107 - The implementation of stage 2 with axolotl

Issue - State: open - Opened by boxiaowave almost 2 years ago

#106 - PPL compute

Issue - State: open - Opened by yuyangxie96 almost 2 years ago

#105 - Fix TGI's medusa link

Pull Request - State: open - Opened by fxmarty almost 2 years ago

#105 - Fix TGI's medusa link

Pull Request - State: open - Opened by fxmarty almost 2 years ago

#104 - Containerization with Dockerfile to setup medusa

Issue - State: open - Opened by gangooteli almost 2 years ago - 1 comment

#103 - Fix for removing LM_HEAD and upgrading Medusa v2

Pull Request - State: closed - Opened by tgaddair almost 2 years ago

#103 - Fix for removing LM_HEAD and upgrading Medusa v2

Pull Request - State: open - Opened by tgaddair almost 2 years ago

#101 - fix preprocess function

Issue - State: open - Opened by xiezipeng-ML almost 2 years ago

#100 - Using Medusa with Whisper

Issue - State: open - Opened by AvivSham almost 2 years ago

#99 - Token-wise the same generalization?

Issue - State: closed - Opened by Ageliss almost 2 years ago - 2 comments

#98 - ImportError: cannot import name 'is_flash_attn_available' from 'transformers.utils'

Issue - State: open - Opened by imneov almost 2 years ago - 3 comments

#97 - Creating medusa2.

Pull Request - State: closed - Opened by Narsil almost 2 years ago - 1 comment

#97 - Creating medusa2.

Pull Request - State: closed - Opened by Narsil almost 2 years ago - 1 comment

#96 - Is there a bug in gen_model_answer_baseline.py?

Issue - State: open - Opened by qspang almost 2 years ago

#95 - Medusa Training Loss

Issue - State: open - Opened by TomYang-TZ almost 2 years ago

#94 - train medusa stage-2

Issue - State: open - Opened by smartliuhw almost 2 years ago - 1 comment

#93 - mistral.json

Issue - State: open - Opened by Git-L1 almost 2 years ago

#90 - HYDRA support?

Issue - State: open - Opened by arunpatala almost 2 years ago

#89 - Misleading Name LLM Name MEDUSA

Issue - State: open - Opened by Pittconnect almost 2 years ago

#85 - Why medusa-2 train llama2 with no such great improvement?

Issue - State: open - Opened by MeJerry215 about 2 years ago - 3 comments

#84 - release medusa-llm v0.2

Issue - State: closed - Opened by zhyncs about 2 years ago - 1 comment

#83 - Adding recipe for other models (non llama, non vicuna).

Pull Request - State: closed - Opened by Narsil about 2 years ago

#83 - Adding recipe for other models (non llama, non vicuna).

Pull Request - State: closed - Opened by Narsil about 2 years ago

#81 - Encounter an CUDA error when set Medusa head

Issue - State: open - Opened by 1649759610 about 2 years ago

#80 - Support batch size > 1

Pull Request - State: open - Opened by xwang365 about 2 years ago

#80 - Support batch size > 1

Pull Request - State: open - Opened by xwang365 about 2 years ago

#79 - Why the speed up of Medusa 1 on vicuna changed?

Issue - State: open - Opened by niyunsheng about 2 years ago

#78 - deepspeed support

Issue - State: open - Opened by jiangix-paper about 2 years ago

#77 - Is there no way to inference without training?

Issue - State: open - Opened by MoOo2mm about 2 years ago

#76 - medusa-2 HF repo has no 'medusa_num_heads' in config

Issue - State: closed - Opened by HaebinShin about 2 years ago - 1 comment

#74 - Question about Heads warmup

Issue - State: open - Opened by eloooooon about 2 years ago

#73 - Medusa 1 and 2 speed up

Issue - State: closed - Opened by LotuSrc about 2 years ago - 2 comments

#72 - update Community Adoption for RTP-LLM

Pull Request - State: closed - Opened by zhyncs about 2 years ago - 2 comments

#71 - V1.0 prerelease

Pull Request - State: closed - Opened by ctlllll about 2 years ago

#70 - Training Medusa heads

Issue - State: open - Opened by mmilunovic-mdcs about 2 years ago

#69 - OSError

Issue - State: open - Opened by qspang about 2 years ago

#68 - About changing LLM from LLAMA to LLAMA-2

Issue - State: closed - Opened by dydrkfl06 about 2 years ago - 2 comments

#67 - how did you construct the sparse tree architecture

Issue - State: closed - Opened by pengfeiwu1999 about 2 years ago - 2 comments

#66 - Clarifications on Models + Batch Size

Issue - State: closed - Opened by RonanKMcGovern about 2 years ago - 5 comments

#65 - Can I make an AWQ quantization?

Issue - State: closed - Opened by RonanKMcGovern about 2 years ago - 1 comment

#64 - Sparse candidate generation confusion

Issue - State: closed - Opened by zankner over 2 years ago - 6 comments

#63 - Some questions about sampling strategy

Issue - State: closed - Opened by qianxiao1111 over 2 years ago - 3 comments

#62 - Results for different configs

Issue - State: closed - Opened by zankner over 2 years ago - 8 comments

#61 - How to load finetune checkpoint files directly？

Issue - State: closed - Opened by qianxiao1111 over 2 years ago

#60 - AttributeError: 'LlamaForCausalLM' object has no attribute 'medusa_head'

Issue - State: closed - Opened by blwaji over 2 years ago - 2 comments

#57 - FasterTransformer support

Issue - State: open - Opened by niyunsheng over 2 years ago - 1 comment

#56 - Will using this method result in inconsistent output results?

Issue - State: closed - Opened by niyunsheng over 2 years ago - 8 comments

#55 - TypeError: __init__() got an unexpected keyword argument 'medusa_num_heads'

Issue - State: closed - Opened by HackGiter over 2 years ago - 6 comments

#54 - Mistral 7B model support

Pull Request - State: closed - Opened by JianbangZ over 2 years ago - 4 comments

#54 - Mistral 7B model support

Pull Request - State: closed - Opened by JianbangZ over 2 years ago - 4 comments

#53 - Llm judge update

Pull Request - State: closed - Opened by leeyeehoo over 2 years ago

#52 - [Feature Request] Qwen model support

Issue - State: open - Opened by JianbangZ over 2 years ago - 1 comment

#51 - errors occurred when running simple_gradio_interface.py

Issue - State: closed - Opened by MeWannaSleep over 2 years ago - 2 comments

#50 - Install the package with the console script ?

Issue - State: closed - Opened by devrimcavusoglu over 2 years ago - 1 comment

#49 - How to test latency between medusa & baseline

Issue - State: closed - Opened by YixinSong-e over 2 years ago - 3 comments

#48 - name not exist "from medusa.model.medusa_choices import medusa_choices"

Issue - State: closed - Opened by JianbangZ over 2 years ago - 4 comments

#46 - update roadmap

Pull Request - State: closed - Opened by leeyeehoo over 2 years ago

#42 - Sparse tree

Pull Request - State: closed - Opened by ctlllll over 2 years ago

#41 - vLLM support

Issue - State: open - Opened by MichaelJayW over 2 years ago - 12 comments

#40 - Pull main to sparse_tree

Pull Request - State: closed - Opened by leeyeehoo over 2 years ago

#39 - [New feature] More sampling schemes

Issue - State: closed - Opened by Jokoe66 over 2 years ago - 2 comments
Labels: enhancement

#38 - add development bounty

Pull Request - State: closed - Opened by ctlllll over 2 years ago

#37 - Benchmark results

Issue - State: closed - Opened by JianbangZ over 2 years ago - 3 comments