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
#561 - Update pyproject.toml
Pull Request -
State: closed - Opened by shivam15s 6 days ago
#560 - add fuse reglu op with test
Pull Request -
State: closed - Opened by YixinSong-e 7 days ago
#559 - add batch_norm op with test and benchmark
Pull Request -
State: open - Opened by yanghailong-git 9 days ago
- 1 comment
#558 - Support Granite 3.0 and 3.1 models
Pull Request -
State: open - Opened by JamesKunstle 12 days ago
- 1 comment
#557 - Support IBM Granite 3.(0, 1) models
Issue -
State: open - Opened by JamesKunstle 12 days ago
- 1 comment
#556 - add batch_norm op
Pull Request -
State: closed - Opened by yanghailong-git 14 days ago
#555 - RMSNorm & SwiGLU activation recomputation
Issue -
State: open - Opened by huyiwen 15 days ago
- 2 comments
#554 - Fix DPO unit test fail and refactor
Pull Request -
State: closed - Opened by Tcc0403 15 days ago
- 5 comments
#553 - Grpo loss
Pull Request -
State: open - Opened by kashif 15 days ago
#552 - Monkeypatch for Qwen2.5-VL
Pull Request -
State: open - Opened by BenasdTW 16 days ago
- 5 comments
#551 - [Verl] Add entropy loss to `cross_entropy_loss` and `fused_linear_cross_entropy_loss`
Pull Request -
State: open - Opened by hongpeng-guo 18 days ago
- 2 comments
#546 - [WIP] Fix convergence test for future transfomers version
Pull Request -
State: open - Opened by Tcc0403 18 days ago
#543 - The convergence test `test_mini_models_with_logits` is failing with the latest transformers
Issue -
State: open - Opened by Tcc0403 20 days ago
- 7 comments
Labels: bug
#542 - `revert_liger_kernel_to_xxx` can't revert LigerCrossEntropyLoss for transformers>=4.46.1
Issue -
State: open - Opened by Tcc0403 20 days ago
- 3 comments
Labels: bug
#541 - Format files
Pull Request -
State: closed - Opened by austin362667 21 days ago
#540 - Add Flex Attention Monkey Patch for LLAMA
Pull Request -
State: open - Opened by austin362667 22 days ago
#539 - Improve Hugging Face SFT Script
Pull Request -
State: open - Opened by ParagEkbote 23 days ago
- 5 comments
#538 - Issue while building from source on ROCM
Issue -
State: open - Opened by agunapal 25 days ago
- 3 comments
#537 - Support the new Solar architecture
Issue -
State: open - Opened by arnavgarg1 25 days ago
#536 - add GitHub CI for Intel GPU
Pull Request -
State: open - Opened by faaany 26 days ago
- 6 comments
#535 - [tests] use a valid hexadecimal string instead of a placeholder
Pull Request -
State: closed - Opened by faaany 26 days ago
- 1 comment
#534 - Add Mkdocs related dependencies to setup.py
Pull Request -
State: closed - Opened by hebiao064 26 days ago
#533 - Gradient checkpointing for `grad_weight` in LFCE
Issue -
State: open - Opened by cassanof 27 days ago
- 4 comments
#532 - Fix LlamaRotaryEmbedding Tests [#520]
Pull Request -
State: closed - Opened by manojks1999 28 days ago
- 1 comment
#531 - Remove extra print
Pull Request -
State: closed - Opened by apaz-cli 29 days ago
- 1 comment
#530 - Add argument `return_z_loss` to flce
Pull Request -
State: closed - Opened by Tcc0403 29 days ago
- 1 comment
#529 - Handle cache_position for transformers 4.47.0 and later (#528)
Pull Request -
State: closed - Opened by BenasdTW 29 days ago
- 3 comments
#528 - Qwen2-VL breaks with transformers version 4.47.0+: `TypeError: lce_forward() got an unexpected keyword argument 'cache_position'`
Issue -
State: closed - Opened by BenasdTW 30 days ago
- 1 comment
#527 - `return_z_loss` is not supported for `LigerFusedLinearCrossEntropyFunction` and `LigerFusedLinearCrossEntropyLoss`
Issue -
State: closed - Opened by apaz-cli 30 days ago
Labels: good first issue
#526 - Fix HF `transformers` Breaking Changes
Pull Request -
State: closed - Opened by austin362667 about 1 month ago
- 1 comment
#525 - `LlamaRotaryEmbedding` Input Argument Is Inconsistent with Hugging Face
Issue -
State: closed - Opened by austin362667 about 1 month ago
#524 - Add huggingface llava
Pull Request -
State: open - Opened by jp1924 about 1 month ago
- 20 comments
#523 - Fix Unit Test error brought by Transformer Breaking Changes
Pull Request -
State: closed - Opened by hebiao064 about 1 month ago
- 3 comments
#522 - [Tiny] Add QVQ to readme
Pull Request -
State: closed - Opened by tyler-romero about 1 month ago
#521 - [DPO] add reference log-prob outputs in DPO
Pull Request -
State: open - Opened by kashif about 1 month ago
#520 - NVIDIA CI failing due to transformers v4.48.0 refactor
Issue -
State: closed - Opened by Tcc0403 about 1 month ago
- 2 comments
Labels: bug, good first issue
#519 - Fix mean subtraction in layer norm kernels
Pull Request -
State: open - Opened by nhamanasu about 1 month ago
- 5 comments
#518 - For better numerical accuracy in LayerNorm
Issue -
State: open - Opened by nhamanasu about 1 month ago
- 2 comments
#517 - Memory Optimization with Liger Kernel Shows Limited Effect on larger Model (more than 7B)
Issue -
State: open - Opened by dyyoungg about 1 month ago
- 3 comments
#515 - IndexError: The shape of the mask [7387] at index 0 does not match the shape of the indexed tensor [1] at index 0
Issue -
State: closed - Opened by 14H034160212 about 1 month ago
- 2 comments
Labels: bug, good first issue
#514 - Any plans to add models from the llava series?
Issue -
State: open - Opened by jp1924 about 1 month ago
- 10 comments
#513 - Megatron Support
Issue -
State: open - Opened by huyiwen about 1 month ago
#512 - `LigerFusedLinearCrossEntropyLoss` Causes Training Loss to Diverge After Reaching ~8
Issue -
State: open - Opened by penghui-yang about 1 month ago
- 6 comments
#511 - Refactor CrossEntropy and FusedLinearCrossEntropy
Pull Request -
State: open - Opened by Tcc0403 about 1 month ago
#510 - Add `average_log_prob` args for cpo
Pull Request -
State: open - Opened by Mecoli1219 about 1 month ago
- 1 comment
#509 - Is Liger-Kernel significantly slower than torch based on benchmark?
Issue -
State: closed - Opened by wa008 about 2 months ago
- 1 comment
#508 - Set z_loss_1d=None when return_z_loss=False in cross_entropy_loss to avoid tl.store fail when triton_interpret=1(for tl.device_print etc.)
Pull Request -
State: closed - Opened by wa008 about 2 months ago
#507 - result of LigerCrossEntropyLoss is always 0
Issue -
State: closed - Opened by wa008 about 2 months ago
- 6 comments
#506 - [CI] Add ROCm 6.3 CI
Pull Request -
State: open - Opened by tjtanaa about 2 months ago
- 4 comments
#505 - Consider support liger kernel for internlm model
Issue -
State: open - Opened by 14H034160212 about 2 months ago
#504 - Add unit tests for shared prefix masked attention with `torch.FlexAttention`
Pull Request -
State: open - Opened by austin362667 about 2 months ago
- 2 comments
#503 - [ORPO] add nll_target for orpo nll loss
Pull Request -
State: open - Opened by kashif about 2 months ago
#502 - Fix Dtype Mismatch in torch.addmm within ops/fused_linear_cross_entropy.py in AMP training.
Pull Request -
State: closed - Opened by DandinPower about 2 months ago
- 5 comments
#501 - Dtype Mismatch in `torch.addmm` within `ops/fused_linear_cross_entropy.py` in AMP training
Issue -
State: closed - Opened by DandinPower about 2 months ago
#500 - Extending Liger-Kernel Optimizations to Encoder Models Like BER
Issue -
State: open - Opened by pengzhangzhi about 2 months ago
#499 - [Model] DeepseekV2 Support
Pull Request -
State: open - Opened by saurabhkoshatwar about 2 months ago
- 1 comment
#498 - [tests] skip failed tests for xpu
Pull Request -
State: closed - Opened by faaany about 2 months ago
- 2 comments
#497 - annotate tl constexpr values
Pull Request -
State: closed - Opened by winglian about 2 months ago
#496 - Fix/liger fused linear cross entropy function does not support reduction=none
Pull Request -
State: closed - Opened by ryankert01 about 2 months ago
- 1 comment
#495 - speed up kto loss and some refactor
Pull Request -
State: closed - Opened by shivam15s about 2 months ago
#494 - speed up kto loss and some refactor
Pull Request -
State: closed - Opened by shivam15s about 2 months ago
#493 - CPO & SimPO add label_smoothing
Pull Request -
State: closed - Opened by Mecoli1219 about 2 months ago
- 3 comments
#492 - Add `aux_outputs` for CPO and SimPO
Pull Request -
State: open - Opened by Mecoli1219 about 2 months ago
#491 - Refactor chunked preference functions and distillation base class
Pull Request -
State: open - Opened by shivam15s about 2 months ago
#490 - fix dpo tests: reduce tolerance and change default compute_nll_loss false
Pull Request -
State: closed - Opened by shivam15s about 2 months ago
#489 - Revert "fix chosen_nll_loss in chunked losses (#486)"
Pull Request -
State: closed - Opened by shivam15s about 2 months ago
#488 - LigerFusedLinearCrossEntropyFunction does not support reduction=None
Issue -
State: closed - Opened by Xiang-cd 2 months ago
- 1 comment
Labels: good first issue
#487 - error when run `sh run_qwen.sh`
Issue -
State: open - Opened by CharlesJhonson 2 months ago
- 3 comments
Labels: good first issue
#486 - fix chosen_nll_loss in chunked losses
Pull Request -
State: closed - Opened by kashif 2 months ago
#485 - Create Docs for Liger-Kernel
Pull Request -
State: closed - Opened by ParagEkbote 2 months ago
- 8 comments
#484 - Fix Preference Loss and Refactor for Readability
Pull Request -
State: closed - Opened by austin362667 2 months ago
#483 - Move the checkstyle to [Ruff](https://docs.astral.sh/ruff/)
Pull Request -
State: closed - Opened by shivam15s 2 months ago
- 1 comment
#482 - fix: correct typos in docstrings
Pull Request -
State: closed - Opened by shivam15s 2 months ago
#481 - preference loss sign is inverted and leads to negative loss
Pull Request -
State: closed - Opened by winglian 2 months ago
- 1 comment
#480 - Fine-tuned qwen2.5-7b reported a backward error
Issue -
State: closed - Opened by chenchen0611 2 months ago
- 1 comment
#479 - [Transformer] fix ORPO loss for MOE models
Pull Request -
State: closed - Opened by kashif 2 months ago
#478 - What's the different with FlagGems
Issue -
State: closed - Opened by CharlesJhonson 2 months ago
#477 - Fix Rope Compatibility with Cos/Sin Position Embedding for Batch Size > 1
Pull Request -
State: closed - Opened by wizyoung 2 months ago
- 1 comment
#476 - Potential Optimization for Preference Training with Prefix Sharing
Issue -
State: open - Opened by austin362667 2 months ago
#475 - Add KTO Loss
Pull Request -
State: closed - Opened by hebiao064 2 months ago
- 2 comments
#474 - error when run kernel test
Issue -
State: closed - Opened by CharlesJhonson 2 months ago
- 3 comments
#473 - align post training loss at the center
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#472 - Add more post training in readme
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#471 - [CI] runtime pip install using uv
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#470 - modify ref_input in chunked_loss base class and fix tests
Pull Request -
State: closed - Opened by shivam15s 2 months ago
#469 - Revert "Add ref_input parameter to support separate inputs for reference model"
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#468 - test: Add test for ref_input parameter in fused linear preference
Pull Request -
State: open - Opened by xingyaoww 2 months ago
#467 - Add ref_input parameter to support separate inputs for reference model
Pull Request -
State: closed - Opened by xingyaoww 2 months ago
- 3 comments
#466 - Revert Workaround of Disabling QWEN2_VL in Convergence Tests
Pull Request -
State: closed - Opened by austin362667 2 months ago
#465 - Add on-paper form of RoPE kernel
Pull Request -
State: open - Opened by Comet0322 2 months ago
- 1 comment
#464 - Fix Qwen2VL mrope for transformers 4.47.0
Pull Request -
State: closed - Opened by li-plus 2 months ago
- 5 comments
#463 - Disable Qwen2 VL test for with logits conv test
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#462 - Update pyproject.toml
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#461 - Qwen VL Convergence Test Fails for Transformers >= 4.47.0
Issue -
State: closed - Opened by ByronHsu 2 months ago
- 2 comments
#460 - Add dynamic dependency management for CUDA and ROCm
Pull Request -
State: closed - Opened by hebiao064 2 months ago
#459 - Fix liger orpo trainer import error
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#458 - [0.5.0] from trl.trainer import ORPOTrainer ModuleNotFoundError: No module named 'trl'
Issue -
State: closed - Opened by Fazziekey 2 months ago
- 2 comments
#457 - add sponsorship and collab
Pull Request -
State: closed - Opened by ByronHsu 2 months ago
#456 - Add HIP (ROCm) and Liger Kernel to env report
Pull Request -
State: closed - Opened by Comet0322 2 months ago
#455 - version bump to 0.5.0
Pull Request -
State: closed - Opened by shivam15s 2 months ago