Ecosyste.ms: Issues

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

GitHub / linkedin/Liger-Kernel issues and pull requests

#329 - Cross entropy for packing

Issue - State: open - Opened by fzyzcjy 12 days ago

#328 - Logits seems to be missing

Issue - State: open - Opened by fzyzcjy 13 days ago - 4 comments

#327 - Creation of Python autodoc using Sphinx

Pull Request - State: open - Opened by ParagEkbote 13 days ago

#326 - [AMD] [ROCm] Pick `num_warps` based on platform

Pull Request - State: open - Opened by tjtanaa 14 days ago

#325 - Fix incorrect training of first and last Medusa heads

Pull Request - State: open - Opened by chiwanpark 15 days ago

#324 - Add TVD Loss Kernel

Pull Request - State: open - Opened by saurabhkoshatwar 15 days ago - 1 comment

#323 - Liger kernel does not speed up, and even slow down in many scenarios

Issue - State: closed - Opened by fzyzcjy 16 days ago - 3 comments

#322 - handle updated gradient accumulation fixes from transformers

Pull Request - State: open - Opened by winglian 17 days ago - 1 comment

#321 - added batch norm

Pull Request - State: open - Opened by vulkomilev 18 days ago

#320 - Support FusedLinearCrossEntropy for Gemma2

Pull Request - State: open - Opened by Tcc0403 19 days ago

#319 - Training LLaVA with the Liger kernel results in degraded performance.

Issue - State: open - Opened by y-rok 19 days ago - 1 comment

#318 - fix FLCE AMP issue

Pull Request - State: closed - Opened by yundai424 19 days ago - 5 comments

#317 - Update citation and add tech report

Pull Request - State: closed - Opened by ByronHsu 19 days ago

#316 - Introducing Liger Kernel Guru on Gurubase.io

Pull Request - State: closed - Opened by kursataktas 19 days ago - 2 comments

#315 - mllama patch modifies nn.LayerNorm globally

Issue - State: open - Opened by tyler-romero 21 days ago - 3 comments

#314 - why logits of model outputs is None

Issue - State: closed - Opened by Yangruipis 22 days ago - 2 comments

#312 - Memory usage is different between GPUs when using FSDP1 and meta device.

Issue - State: closed - Opened by cailun01 25 days ago - 2 comments

#311 - OOM issue occurs when training Llama-3-8B

Issue - State: closed - Opened by dongwook92 25 days ago - 2 comments

#310 - Add missing ignore_index tests

Pull Request - State: open - Opened by Tcc0403 25 days ago

#309 - Empty Medusa head tensors

Issue - State: open - Opened by vkc1vk 26 days ago - 1 comment

#308 - Move `logits.float()` call

Pull Request - State: closed - Opened by ringohoffman 27 days ago - 2 comments

#307 - Fix dtype mismatch in fused_linear_cross_entropy_forward

Pull Request - State: closed - Opened by kostum123 29 days ago - 1 comment

#306 - Add ignore_index and label to jsd and fl-jsd

Pull Request - State: closed - Opened by Tcc0403 29 days ago

#305 - RuntimeError due to dtype mismatch in fused_linear_cross_entropy_forward

Issue - State: closed - Opened by kostum123 29 days ago - 5 comments
Labels: bug

#304 - Added contributors and back to top

Pull Request - State: closed - Opened by barbarian360 29 days ago

#303 - how to use with peft lora?

Issue - State: closed - Opened by YooSungHyun 30 days ago - 1 comment

#302 - Monkey patch layer norm in mllama

Pull Request - State: closed - Opened by shivam15s 30 days ago

#301 - Add end-to-end example of mllama

Issue - State: open - Opened by shivam15s about 1 month ago

#300 - Add FusedLinearJSD

Pull Request - State: closed - Opened by Tcc0403 about 1 month ago

#299 - Add Fused Linear JSD

Pull Request - State: closed - Opened by Tcc0403 about 1 month ago

#298 - Add fused_linear_jsd

Pull Request - State: closed - Opened by Tcc0403 about 1 month ago

#297 - Add TensorFlow Check and Improve Package Listing in Environment Report

Pull Request - State: closed - Opened by Vanshika110 about 1 month ago

#296 - fix typos and improve grammar in README.md

Pull Request - State: closed - Opened by amantyagiprojects about 1 month ago

#295 - Update rope.py

Pull Request - State: closed - Opened by Yash-2707 about 1 month ago - 1 comment

#294 - Apache and MIT license reference

Pull Request - State: closed - Opened by momochen about 1 month ago

#293 - chore: update cross_entropy.py

Pull Request - State: closed - Opened by eltociear about 1 month ago

#292 - [DO NOT REVIEW] Add sleep in tests

Pull Request - State: closed - Opened by helloworld1 about 1 month ago

#291 - WIP Set gpu ci env to have HF hub read token

Pull Request - State: closed - Opened by shivam15s about 1 month ago

#290 - Add beta support for jsd

Pull Request - State: closed - Opened by Tcc0403 about 1 month ago - 4 comments

#289 - Cancel in-progress but out-of-date GPU actions

Pull Request - State: closed - Opened by tyler-romero about 1 month ago - 1 comment

#288 - Disable gemma2 and qwen2_vl tests

Pull Request - State: closed - Opened by shimizust about 1 month ago

#287 - Acknowledgement in NOTICE file

Pull Request - State: closed - Opened by momochen about 1 month ago

#286 - Release version 0.3.1

Pull Request - State: closed - Opened by shimizust about 1 month ago - 1 comment

#285 - 2024 Q4 Roadmap

Issue - State: open - Opened by ByronHsu about 1 month ago

#285 - 2024 Q4 Roadmap

Issue - State: open - Opened by ByronHsu about 1 month ago

#284 - poke tests

Pull Request - State: closed - Opened by tyler-romero about 1 month ago

#283 - Add missing Qwen2-VL monkey patch test

Pull Request - State: closed - Opened by tyler-romero about 1 month ago - 1 comment

#282 - Monkeypatch for Llama 3.2-Vision

Pull Request - State: closed - Opened by tyler-romero about 1 month ago - 13 comments

#281 - Add TVD (Total variation distance) Kernel

Issue - State: open - Opened by qingquansong about 1 month ago - 1 comment
Labels: feature

#281 - Add TVD (Total variation distance) Kernel

Issue - State: open - Opened by qingquansong about 1 month ago - 3 comments
Labels: feature

#280 - Post-init model patching fix

Pull Request - State: closed - Opened by shimizust about 1 month ago - 1 comment

#280 - Post-init model patching fix

Pull Request - State: closed - Opened by shimizust about 1 month ago - 1 comment

#279 - Weights are not copied when model is already instantiated

Issue - State: closed - Opened by ByronHsu about 1 month ago - 1 comment

#278 - Add general JSD (w/ beta) support

Issue - State: closed - Opened by qingquansong about 1 month ago - 4 comments
Labels: feature

#277 - Adding ignore index support for divergence losses

Issue - State: closed - Opened by qingquansong about 1 month ago
Labels: feature

#277 - Adding ignore index support for divergence losses

Issue - State: open - Opened by qingquansong about 1 month ago
Labels: feature

#276 - fix qwen2-vl: create correct rope position_ids when position_ids is None

Pull Request - State: closed - Opened by Sanster about 1 month ago

#275 - [Kernel] Flash attention 2

Pull Request - State: open - Opened by remi-or about 1 month ago - 2 comments

#274 - FIX: tl.program_id() does indeed not have a cast method in triton2.3.1

Pull Request - State: closed - Opened by wizyoung about 1 month ago - 1 comment

#273 - Ensure In-place correctness checks work properly

Pull Request - State: open - Opened by Tcc0403 about 2 months ago - 4 comments

#272 - In-place operations in triton kernel might result in incorrect gradient calculations

Issue - State: open - Opened by Tcc0403 about 2 months ago - 3 comments
Labels: bug

#271 - Encountered errors when reproducing lightning training example

Issue - State: open - Opened by ReginaZh about 2 months ago - 3 comments

#270 - Relaxed transformers dependency

Pull Request - State: closed - Opened by shimizust about 2 months ago

#269 - Fix sharing a ResBlock layer for each head in Medusa example

Pull Request - State: closed - Opened by chiwanpark about 2 months ago - 2 comments

#267 - rename cuda mode to gpu mode

Pull Request - State: closed - Opened by msaroufim about 2 months ago

#266 - Choice of num_warps

Issue - State: open - Opened by Edenzzzz about 2 months ago - 7 comments

#265 - chore: Add Qwen2.5 and Phi3.5 to Readme

Pull Request - State: closed - Opened by tyler-romero about 2 months ago

#264 - Add JSD kernel

Pull Request - State: closed - Opened by Tcc0403 about 2 months ago - 6 comments

#263 - Fix AutoLigerKernelForCausalLM to pass through original kwargs

Pull Request - State: closed - Opened by shimizust about 2 months ago - 1 comment

#262 - Fix/kldiv

Pull Request - State: closed - Opened by S1ro1 about 2 months ago - 13 comments

#261 - Fix assert_verbose_allclose bugs

Pull Request - State: closed - Opened by Tcc0403 about 2 months ago

#260 - Update contributing guide for adding a new model

Pull Request - State: closed - Opened by shivam15s about 2 months ago

#259 - test.utils.assert_verbose_allclose has multiple bugs

Issue - State: closed - Opened by Tcc0403 about 2 months ago - 4 comments

#258 - Optional dependency on transformers

Issue - State: closed - Opened by DuarteMRAlves about 2 months ago - 6 comments

#257 - Loss does not drop when using Liger Kernel at Qwen2.5

Issue - State: open - Opened by Se-Hun about 2 months ago - 9 comments

#256 - Fix a comment typo in flce

Pull Request - State: closed - Opened by Tcc0403 about 2 months ago

#255 - RMSNorm aggregation

Pull Request - State: closed - Opened by Tcc0403 about 2 months ago - 3 comments

#254 - Are you even allowed to do these ops inplace?

Issue - State: closed - Opened by mayank31398 about 2 months ago - 4 comments

#253 - [Model] Pixtral Support

Pull Request - State: open - Opened by AndreSlavescu about 2 months ago - 3 comments

#252 - reverse KL and JSD

Issue - State: closed - Opened by yundai424 about 2 months ago - 1 comment
Labels: feature

#251 - [Easy] Cast program_id to int64 in SwiGLU/GeGLU kernels

Pull Request - State: closed - Opened by hansonw about 2 months ago

#250 - AutoLigerKernelForCausalLM.from_pretrained discards hub_kwargs_names

Issue - State: closed - Opened by uris-opti about 2 months ago - 5 comments

#249 - ValueError when Loading Qwen2-VL Model with Liger Kernel

Issue - State: open - Opened by rahatarinasir about 2 months ago - 1 comment

#248 - Support for Cohere models

Issue - State: open - Opened by nyxkrage about 2 months ago - 1 comment

#247 - Remove debug print statement

Pull Request - State: closed - Opened by EdoardoLuciani about 2 months ago - 1 comment

#246 - Release Liger-Kernel version 0.3.0

Pull Request - State: closed - Opened by qingquansong about 2 months ago

#245 - Is not compatible with DoRA?

Issue - State: open - Opened by gotzmann about 2 months ago - 1 comment
Labels: bug

#244 - Add label smoothing to FLCE and unit tests

Pull Request - State: closed - Opened by Tcc0403 about 2 months ago - 1 comment

#243 - Lable smoothing is not applied and tested in flce

Issue - State: closed - Opened by Tcc0403 about 2 months ago

#241 - Reasons for upcasting the logits dtype outside the kernel

Issue - State: open - Opened by yzhangcs 2 months ago - 7 comments

#240 - SWIFT Trainer Integration

Pull Request - State: closed - Opened by tastelikefeet 2 months ago

#239 - Support Z Loss in CE

Pull Request - State: open - Opened by Tcc0403 2 months ago - 4 comments
Labels: reviewing

#238 - fused_linear_cross_entropy: Move float32 cast into kernel

Pull Request - State: open - Opened by hansonw 2 months ago - 6 comments