liger-kernel-nightly 0.5.4.dev20250305025024__tar.gz → 0.5.4.dev20250305231637__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.


This version of liger-kernel-nightly might be problematic. Click here for more details.

Files changed (229) hide show
  1. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/PKG-INFO +1 -1
  2. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/pyproject.toml +1 -1
  3. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel_nightly.egg-info/PKG-INFO +1 -1
  4. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_kto_loss.py +16 -16
  5. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/ISSUE_TEMPLATE/bug_report.yaml +0 -0
  6. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/ISSUE_TEMPLATE/feature_request.yaml +0 -0
  7. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/pull_request_template.md +0 -0
  8. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/amd-ci.yml +0 -0
  9. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/docs.yml +0 -0
  10. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/intel-ci.yml +0 -0
  11. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/nvi-ci.yml +0 -0
  12. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/publish-nightly.yml +0 -0
  13. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.github/workflows/publish-release.yml +0 -0
  14. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/.gitignore +0 -0
  15. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/LICENSE +0 -0
  16. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/Makefile +0 -0
  17. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/NOTICE +0 -0
  18. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/README.md +0 -0
  19. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/README.md +0 -0
  20. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/__init__.py +0 -0
  21. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/benchmarks_visualizer.py +0 -0
  22. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/data/all_benchmark_data.csv +0 -0
  23. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/__init__.py +0 -0
  24. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_cpo_loss.py +0 -0
  25. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_cross_entropy.py +0 -0
  26. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_distill_jsd_loss.py +0 -0
  27. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_dpo_loss.py +0 -0
  28. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_embedding.py +0 -0
  29. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_fused_linear_cross_entropy.py +0 -0
  30. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_fused_linear_jsd.py +0 -0
  31. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_geglu.py +0 -0
  32. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_group_norm.py +0 -0
  33. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_jsd.py +0 -0
  34. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_kl_div.py +0 -0
  35. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_kto_loss.py +0 -0
  36. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_layer_norm.py +0 -0
  37. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_orpo_loss.py +0 -0
  38. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_qwen2vl_mrope.py +0 -0
  39. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_rms_norm.py +0 -0
  40. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_rope.py +0 -0
  41. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_simpo_loss.py +0 -0
  42. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_swiglu.py +0 -0
  43. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/benchmark_tvd.py +0 -0
  44. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/benchmark/scripts/utils.py +0 -0
  45. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/dev/fmt-requirements.txt +0 -0
  46. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/dev/modal/tests.py +0 -0
  47. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/dev/modal/tests_bwd.py +0 -0
  48. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/Examples.md +0 -0
  49. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/Getting-Started.md +0 -0
  50. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/High-Level-APIs.md +0 -0
  51. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/Low-Level-APIs.md +0 -0
  52. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/acknowledgement.md +0 -0
  53. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/contributing.md +0 -0
  54. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/banner.GIF +0 -0
  55. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/compose.gif +0 -0
  56. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/e2e-memory.png +0 -0
  57. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/e2e-tps.png +0 -0
  58. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/logo-banner.png +0 -0
  59. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/patch.gif +0 -0
  60. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/images/post-training.png +0 -0
  61. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/index.md +0 -0
  62. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/docs/license.md +0 -0
  63. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/alignment/accelerate_config.yaml +0 -0
  64. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/alignment/run_orpo.py +0 -0
  65. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/README.md +0 -0
  66. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/callback.py +0 -0
  67. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/config/fsdp_config.json +0 -0
  68. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/gemma_7b_mem.png +0 -0
  69. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/gemma_7b_tp.png +0 -0
  70. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/llama_mem_alloc.png +0 -0
  71. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/llama_tps.png +0 -0
  72. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/qwen_mem_alloc.png +0 -0
  73. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/img/qwen_tps.png +0 -0
  74. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/launch_on_modal.py +0 -0
  75. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/requirements.txt +0 -0
  76. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/run_benchmarks.sh +0 -0
  77. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/run_gemma.sh +0 -0
  78. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/run_llama.sh +0 -0
  79. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/run_qwen.sh +0 -0
  80. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/run_qwen2_vl.sh +0 -0
  81. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/training.py +0 -0
  82. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/huggingface/training_multimodal.py +0 -0
  83. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/lightning/README.md +0 -0
  84. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/lightning/requirements.txt +0 -0
  85. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/lightning/training.py +0 -0
  86. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/README.md +0 -0
  87. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/callback.py +0 -0
  88. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Memory_Stage1_num_head_3.png +0 -0
  89. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Memory_Stage1_num_head_5.png +0 -0
  90. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Memory_Stage2_num_head_3.png +0 -0
  91. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Memory_Stage2_num_head_5.png +0 -0
  92. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Throughput_Stage1_num_head_3.png +0 -0
  93. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Throughput_Stage1_num_head_5.png +0 -0
  94. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Throughput_Stage2_num_head_3.png +0 -0
  95. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/docs/images/Throughput_Stage2_num_head_5.png +0 -0
  96. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/fsdp/acc-fsdp.conf +0 -0
  97. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/medusa_util.py +0 -0
  98. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/requirements.txt +0 -0
  99. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/scripts/llama3_8b_medusa.sh +0 -0
  100. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/examples/medusa/train.py +0 -0
  101. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/licenses/LICENSE-Apache-2.0 +0 -0
  102. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/licenses/LICENSE-MIT-AutoAWQ +0 -0
  103. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/licenses/LICENSE-MIT-Efficient-Cross-Entropy +0 -0
  104. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/licenses/LICENSE-MIT-llmc +0 -0
  105. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/licenses/LICENSE-MIT-triton +0 -0
  106. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/mkdocs.yml +0 -0
  107. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/setup.cfg +0 -0
  108. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/setup.py +0 -0
  109. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/__init__.py +0 -0
  110. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/README.md +0 -0
  111. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/__init__.py +0 -0
  112. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/cpo_loss.py +0 -0
  113. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/dpo_loss.py +0 -0
  114. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/functional.py +0 -0
  115. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/fused_linear_distillation.py +0 -0
  116. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/fused_linear_preference.py +0 -0
  117. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/fused_linear_rlhf.py +0 -0
  118. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/fused_linear_unpaired_preference.py +0 -0
  119. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/grpo_loss.py +0 -0
  120. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/jsd_loss.py +0 -0
  121. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/kto_loss.py +0 -0
  122. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/orpo_loss.py +0 -0
  123. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/chunked_loss/simpo_loss.py +0 -0
  124. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/env_report.py +0 -0
  125. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/__init__.py +0 -0
  126. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/cross_entropy.py +0 -0
  127. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/experimental/embedding.py +0 -0
  128. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/experimental/mm_int8int2.py +0 -0
  129. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/fused_linear_cross_entropy.py +0 -0
  130. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/fused_linear_jsd.py +0 -0
  131. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/geglu.py +0 -0
  132. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/group_norm.py +0 -0
  133. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/jsd.py +0 -0
  134. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/kl_div.py +0 -0
  135. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/layer_norm.py +0 -0
  136. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/qwen2vl_mrope.py +0 -0
  137. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/rms_norm.py +0 -0
  138. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/rope.py +0 -0
  139. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/swiglu.py +0 -0
  140. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/tvd.py +0 -0
  141. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/ops/utils.py +0 -0
  142. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/__init__.py +0 -0
  143. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/auto_model.py +0 -0
  144. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/cross_entropy.py +0 -0
  145. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/experimental/embedding.py +0 -0
  146. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/functional.py +0 -0
  147. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/fused_linear_cross_entropy.py +0 -0
  148. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/fused_linear_jsd.py +0 -0
  149. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/geglu.py +0 -0
  150. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/group_norm.py +0 -0
  151. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/jsd.py +0 -0
  152. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/kl_div.py +0 -0
  153. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/layer_norm.py +0 -0
  154. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/__init__.py +0 -0
  155. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/gemma.py +0 -0
  156. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/gemma2.py +0 -0
  157. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/llama.py +0 -0
  158. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/mistral.py +0 -0
  159. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/mixtral.py +0 -0
  160. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/mllama.py +0 -0
  161. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/olmo2.py +0 -0
  162. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/phi3.py +0 -0
  163. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/qwen2.py +0 -0
  164. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/qwen2_5_vl.py +0 -0
  165. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/model/qwen2_vl.py +0 -0
  166. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/monkey_patch.py +0 -0
  167. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/qwen2vl_mrope.py +0 -0
  168. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/rms_norm.py +0 -0
  169. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/rope.py +0 -0
  170. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/swiglu.py +0 -0
  171. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/trainer/__init__.py +0 -0
  172. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/trainer/orpo_trainer.py +0 -0
  173. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/trainer_integration.py +0 -0
  174. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/transformers/tvd.py +0 -0
  175. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/triton/__init__.py +0 -0
  176. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/triton/monkey_patch.py +0 -0
  177. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel/utils.py +0 -0
  178. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel_nightly.egg-info/SOURCES.txt +0 -0
  179. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel_nightly.egg-info/dependency_links.txt +0 -0
  180. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel_nightly.egg-info/requires.txt +0 -0
  181. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/src/liger_kernel_nightly.egg-info/top_level.txt +0 -0
  182. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/__init__.py +0 -0
  183. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/__init__.py +0 -0
  184. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_cpo_loss.py +0 -0
  185. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_dpo_loss.py +0 -0
  186. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_grpo_loss.py +0 -0
  187. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_jsd_loss.py +0 -0
  188. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_orpo_loss.py +0 -0
  189. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/chunked_loss/test_simpo_loss.py +0 -0
  190. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/conftest.py +0 -0
  191. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/__init__.py +0 -0
  192. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/bf16/__init__.py +0 -0
  193. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/bf16/test_mini_models.py +0 -0
  194. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/bf16/test_mini_models_multimodal.py +0 -0
  195. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/bf16/test_mini_models_with_logits.py +0 -0
  196. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/fp32/__init__.py +0 -0
  197. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/fp32/test_mini_models.py +0 -0
  198. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/fp32/test_mini_models_multimodal.py +0 -0
  199. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/convergence/fp32/test_mini_models_with_logits.py +0 -0
  200. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/fake_configs/Qwen/Qwen2-VL-7B-Instruct/tokenizer_config.json +0 -0
  201. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/fake_configs/Qwen/Qwen2.5-VL-7B-Instruct/tokenizer_config.json +0 -0
  202. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/fake_configs/meta-llama/Llama-3.2-11B-Vision-Instruct/tokenizer_config.json +0 -0
  203. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/scripts/generate_tokenized_dataset.py +0 -0
  204. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/tiny_shakespeare.txt +0 -0
  205. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/tiny_shakespeare_tokenized/data-00000-of-00001.arrow +0 -0
  206. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/tiny_shakespeare_tokenized/dataset_info.json +0 -0
  207. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/resources/tiny_shakespeare_tokenized/state.json +0 -0
  208. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_auto_model.py +0 -0
  209. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_cross_entropy.py +0 -0
  210. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_embedding.py +0 -0
  211. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_flex_attention.py +0 -0
  212. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_fused_linear_cross_entropy.py +0 -0
  213. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_fused_linear_jsd.py +0 -0
  214. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_geglu.py +0 -0
  215. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_group_norm.py +0 -0
  216. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_jsd.py +0 -0
  217. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_kl_div.py +0 -0
  218. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_layer_norm.py +0 -0
  219. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_mm_int8int2.py +0 -0
  220. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_monkey_patch.py +0 -0
  221. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_qwen2vl_mrope.py +0 -0
  222. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_rms_norm.py +0 -0
  223. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_rope.py +0 -0
  224. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_swiglu.py +0 -0
  225. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_trainer_integration.py +0 -0
  226. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_transformers.py +0 -0
  227. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/transformers/test_tvd.py +0 -0
  228. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/triton/test_triton_monkey_patch.py +0 -0
  229. {liger_kernel_nightly-0.5.4.dev20250305025024 → liger_kernel_nightly-0.5.4.dev20250305231637}/test/utils.py +0 -0
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: liger_kernel_nightly
3
- Version: 0.5.4.dev20250305025024
3
+ Version: 0.5.4.dev20250305231637
4
4
  Summary: Efficient Triton kernels for LLM Training
5
5
  License: BSD 2-CLAUSE LICENSE
6
6
  Copyright 2024 LinkedIn Corporation
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
4
4
 
5
5
  [project]
6
6
  name = "liger_kernel_nightly"
7
- version = "0.5.4.dev20250305025024"
7
+ version = "0.5.4.dev20250305231637"
8
8
  description = "Efficient Triton kernels for LLM Training"
9
9
  urls = { "Homepage" = "https://github.com/linkedin/Liger-Kernel" }
10
10
  readme = { file = "README.md", content-type = "text/markdown" }
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.1
2
2
  Name: liger_kernel_nightly
3
- Version: 0.5.4.dev20250305025024
3
+ Version: 0.5.4.dev20250305231637
4
4
  Summary: Efficient Triton kernels for LLM Training
5
5
  License: BSD 2-CLAUSE LICENSE
6
6
  Copyright 2024 LinkedIn Corporation
@@ -250,8 +250,8 @@ def test_correctness(B, T, H, V, scalar, dtype, atol, rtol, bias, ref_bias, igno
250
250
  assert len(aggregated_aux_outputs1) == len(aggregated_aux_outputs2)
251
251
 
252
252
  # chosen_logps
253
- chosen_logps_mean1 = aggregated_aux_outputs1[0] / ((num_chosen_samples * T * V) + 1e-20)
254
- chosen_logps_mean2 = aggregated_aux_outputs2[0] / ((num_chosen_samples * T * V) + 1e-20)
253
+ chosen_logps_mean1 = aggregated_aux_outputs1[0] / ((num_chosen_samples) + 1e-20)
254
+ chosen_logps_mean2 = aggregated_aux_outputs2[0] / ((num_chosen_samples) + 1e-20)
255
255
  assert_verbose_allclose(chosen_logps_mean1, chosen_logps_mean2, atol=atol, rtol=rtol)
256
256
 
257
257
  # chosen_logits
@@ -260,13 +260,13 @@ def test_correctness(B, T, H, V, scalar, dtype, atol, rtol, bias, ref_bias, igno
260
260
  assert_verbose_allclose(chosen_logits_mean1, chosen_logits_mean2, atol=atol, rtol=rtol)
261
261
 
262
262
  # chosen_rewards
263
- chosen_rewards_mean1 = aggregated_aux_outputs1[4] / ((num_chosen_samples * T * V) + 1e-20)
264
- chosen_rewards_mean2 = aggregated_aux_outputs2[4] / ((num_chosen_samples * T * V) + 1e-20)
263
+ chosen_rewards_mean1 = aggregated_aux_outputs1[4] / ((num_chosen_samples) + 1e-20)
264
+ chosen_rewards_mean2 = aggregated_aux_outputs2[4] / ((num_chosen_samples) + 1e-20)
265
265
  assert_verbose_allclose(chosen_rewards_mean1, chosen_rewards_mean2, atol=atol, rtol=rtol)
266
266
 
267
267
  # rejected_logps
268
- rejected_logps_mean1 = aggregated_aux_outputs1[1] / ((num_rejected_samples * T * V) + 1e-20)
269
- rejected_logps_mean2 = aggregated_aux_outputs2[1] / ((num_rejected_samples * T * V) + 1e-20)
268
+ rejected_logps_mean1 = aggregated_aux_outputs1[1] / ((num_rejected_samples) + 1e-20)
269
+ rejected_logps_mean2 = aggregated_aux_outputs2[1] / ((num_rejected_samples) + 1e-20)
270
270
  assert_verbose_allclose(rejected_logps_mean1, rejected_logps_mean2, atol=atol, rtol=rtol)
271
271
 
272
272
  # rejected_logits
@@ -275,8 +275,8 @@ def test_correctness(B, T, H, V, scalar, dtype, atol, rtol, bias, ref_bias, igno
275
275
  assert_verbose_allclose(rejected_logits_mean1, rejected_logits_mean2, atol=atol, rtol=rtol)
276
276
 
277
277
  # rejected_rewards
278
- rejected_rewards_mean1 = aggregated_aux_outputs1[5] / ((num_rejected_samples * T * V) + 1e-20)
279
- rejected_rewards_mean2 = aggregated_aux_outputs2[5] / ((num_rejected_samples * T * V) + 1e-20)
278
+ rejected_rewards_mean1 = aggregated_aux_outputs1[5] / ((num_rejected_samples) + 1e-20)
279
+ rejected_rewards_mean2 = aggregated_aux_outputs2[5] / ((num_rejected_samples) + 1e-20)
280
280
  assert_verbose_allclose(rejected_rewards_mean1, rejected_rewards_mean2, atol=atol, rtol=rtol)
281
281
 
282
282
  loss1.backward()
@@ -392,8 +392,8 @@ def test_correctness_functional(B, T, H, V, scalar, dtype, atol, rtol, bias, ref
392
392
  assert len(aggregated_aux_outputs1) == len(aggregated_aux_outputs2)
393
393
 
394
394
  # chosen_logps
395
- chosen_logps_mean1 = aggregated_aux_outputs1[0] / ((num_chosen_samples * T * V) + 1e-20)
396
- chosen_logps_mean2 = aggregated_aux_outputs2[0] / ((num_chosen_samples * T * V) + 1e-20)
395
+ chosen_logps_mean1 = aggregated_aux_outputs1[0] / ((num_chosen_samples) + 1e-20)
396
+ chosen_logps_mean2 = aggregated_aux_outputs2[0] / ((num_chosen_samples) + 1e-20)
397
397
  assert_verbose_allclose(chosen_logps_mean1, chosen_logps_mean2, atol=atol, rtol=rtol)
398
398
 
399
399
  # chosen_logits
@@ -402,13 +402,13 @@ def test_correctness_functional(B, T, H, V, scalar, dtype, atol, rtol, bias, ref
402
402
  assert_verbose_allclose(chosen_logits_mean1, chosen_logits_mean2, atol=atol, rtol=rtol)
403
403
 
404
404
  # chosen_rewards
405
- chosen_rewards_mean1 = aggregated_aux_outputs1[4] / ((num_chosen_samples * T * V) + 1e-20)
406
- chosen_rewards_mean2 = aggregated_aux_outputs2[4] / ((num_chosen_samples * T * V) + 1e-20)
405
+ chosen_rewards_mean1 = aggregated_aux_outputs1[4] / ((num_chosen_samples) + 1e-20)
406
+ chosen_rewards_mean2 = aggregated_aux_outputs2[4] / ((num_chosen_samples) + 1e-20)
407
407
  assert_verbose_allclose(chosen_rewards_mean1, chosen_rewards_mean2, atol=atol, rtol=rtol)
408
408
 
409
409
  # rejected_logps
410
- rejected_logps_mean1 = aggregated_aux_outputs1[1] / ((num_rejected_samples * T * V) + 1e-20)
411
- rejected_logps_mean2 = aggregated_aux_outputs2[1] / ((num_rejected_samples * T * V) + 1e-20)
410
+ rejected_logps_mean1 = aggregated_aux_outputs1[1] / ((num_rejected_samples) + 1e-20)
411
+ rejected_logps_mean2 = aggregated_aux_outputs2[1] / ((num_rejected_samples) + 1e-20)
412
412
  assert_verbose_allclose(rejected_logps_mean1, rejected_logps_mean2, atol=atol, rtol=rtol)
413
413
 
414
414
  # rejected_logits
@@ -417,8 +417,8 @@ def test_correctness_functional(B, T, H, V, scalar, dtype, atol, rtol, bias, ref
417
417
  assert_verbose_allclose(rejected_logits_mean1, rejected_logits_mean2, atol=atol, rtol=rtol)
418
418
 
419
419
  # rejected_rewards
420
- rejected_rewards_mean1 = aggregated_aux_outputs1[5] / ((num_rejected_samples * T * V) + 1e-20)
421
- rejected_rewards_mean2 = aggregated_aux_outputs2[5] / ((num_rejected_samples * T * V) + 1e-20)
420
+ rejected_rewards_mean1 = aggregated_aux_outputs1[5] / ((num_rejected_samples) + 1e-20)
421
+ rejected_rewards_mean2 = aggregated_aux_outputs2[5] / ((num_rejected_samples) + 1e-20)
422
422
  assert_verbose_allclose(rejected_rewards_mean1, rejected_rewards_mean2, atol=atol, rtol=rtol)
423
423
 
424
424
  loss1.backward()