liger-kernel-nightly 0.5.3.dev20250221002845__py3-none-any.whl → 0.5.3.dev20250221003838__py3-none-any.whl
Sign up to get free protection for your applications and to get access to all the features.
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/METADATA +3 -2
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/RECORD +6 -6
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/LICENSE +0 -0
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/NOTICE +0 -0
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/WHEEL +0 -0
- {liger_kernel_nightly-0.5.3.dev20250221002845.dist-info → liger_kernel_nightly-0.5.3.dev20250221003838.dist-info}/top_level.txt +0 -0
@@ -1,6 +1,6 @@
|
|
1
1
|
Metadata-Version: 2.1
|
2
2
|
Name: liger_kernel_nightly
|
3
|
-
Version: 0.5.3.
|
3
|
+
Version: 0.5.3.dev20250221003838
|
4
4
|
Summary: Efficient Triton kernels for LLM Training
|
5
5
|
License: BSD 2-CLAUSE LICENSE
|
6
6
|
Copyright 2024 LinkedIn Corporation
|
@@ -126,7 +126,7 @@ Requires-Dist: mkdocs-material; extra == "dev"
|
|
126
126
|
|
127
127
|
**Liger Kernel** is a collection of Triton kernels designed specifically for LLM training. It can effectively increase multi-GPU **training throughput by 20%** and reduces **memory usage by 60%**. We have implemented **Hugging Face Compatible** `RMSNorm`, `RoPE`, `SwiGLU`, `CrossEntropy`, `FusedLinearCrossEntropy`, and more to come. The kernel works out of the box with [Flash Attention](https://github.com/Dao-AILab/flash-attention), [PyTorch FSDP](https://pytorch.org/tutorials/intermediate/FSDP_tutorial.html), and [Microsoft DeepSpeed](https://github.com/microsoft/DeepSpeed). We welcome contributions from the community to gather the best kernels for LLM training.
|
128
128
|
|
129
|
-
We've also added optimized Post-Training kernels that deliver **up to 80% memory savings** for alignment and distillation tasks. We support losses like DPO, CPO, ORPO, SimPO, JSD, and many more. Check out [how we optimize the memory](https://x.com/hsu_byron/status/1866577403918917655).
|
129
|
+
We've also added optimized Post-Training kernels that deliver **up to 80% memory savings** for alignment and distillation tasks. We support losses like DPO, CPO, ORPO, SimPO, KTO, JSD, and many more. Check out [how we optimize the memory](https://x.com/hsu_byron/status/1866577403918917655).
|
130
130
|
|
131
131
|
## Supercharge Your Model with Liger Kernel
|
132
132
|
|
@@ -341,6 +341,7 @@ loss.backward()
|
|
341
341
|
| Fused Linear DPO Loss | `liger_kernel.chunked_loss.LigerFusedLinearDPOLoss` |
|
342
342
|
| Fused Linear ORPO Loss | `liger_kernel.chunked_loss.LigerFusedLinearORPOLoss` |
|
343
343
|
| Fused Linear SimPO Loss | `liger_kernel.chunked_loss.LigerFusedLinearSimPOLoss` |
|
344
|
+
| Fused Linear KTO Loss | `liger_kernel.chunked_loss.LigerFusedLinearKTOLoss` |
|
344
345
|
|
345
346
|
### Distillation Kernels
|
346
347
|
|
@@ -65,9 +65,9 @@ liger_kernel/transformers/trainer/__init__.py,sha256=p7yQfklV8-467qSz_ZMimkbDF7H
|
|
65
65
|
liger_kernel/transformers/trainer/orpo_trainer.py,sha256=pdekW7l6Qg_aqa5SYKYlSWUF8m3lkOFvFLcIMEHrz9s,8338
|
66
66
|
liger_kernel/triton/__init__.py,sha256=qCiCamzCRv6lpV8IqpAc9YMdNKC7GKurClWceQPnlis,92
|
67
67
|
liger_kernel/triton/monkey_patch.py,sha256=Rd0hUHAzDkFfHvnX7-PBaNK5EKnZhtfM_h-fgQH9HPY,1568
|
68
|
-
liger_kernel_nightly-0.5.3.
|
69
|
-
liger_kernel_nightly-0.5.3.
|
70
|
-
liger_kernel_nightly-0.5.3.
|
71
|
-
liger_kernel_nightly-0.5.3.
|
72
|
-
liger_kernel_nightly-0.5.3.
|
73
|
-
liger_kernel_nightly-0.5.3.
|
68
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/LICENSE,sha256=OhzLDHJ0to4a8sodVLELZiCFylZ1NAAYLs-HrjPy0ag,1312
|
69
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/METADATA,sha256=zJp1YMQDbbOzeNRKFf7AN7hYqePdT49OEMQkN_buKl8,21963
|
70
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/NOTICE,sha256=njwnoPZLh9AN8SJQzxvCGLHi-8X__AvWRze6joNXIY8,2066
|
71
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/WHEEL,sha256=P9jw-gEje8ByB7_hXoICnHtVCrEwMQh-630tKvQWehc,91
|
72
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/top_level.txt,sha256=2eghu4hA3LnkM7ElW92tQ8zegWKgSbeo-k-aGe1YnvY,13
|
73
|
+
liger_kernel_nightly-0.5.3.dev20250221003838.dist-info/RECORD,,
|
File without changes
|
File without changes
|
File without changes
|