liger-kernel-nightly 0.6.2.dev20251005233549__tar.gz → 0.6.4.dev20260121000828__tar.gz

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/amd-ci.yml

@@ -26,10 +26,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 
@@ -50,10 +50,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/benchmark.yml

@@ -35,22 +35,26 @@ jobs:
       OUTPUT_DIR: benchmarks
       OUTPUT_FILENAME: benchmark.csv
       GENERATED_CSV: benchmark/data/all_benchmark_data.csv
+      # Sanitize user-controlled inputs by declaring them as environment variables
+      # This prevents command injection attacks by filtering dangerous characters
+      INPUT_COMMIT_HASH: ${{ github.event.inputs.commit_hash }}
+      INPUT_OVERWRITE: ${{ github.event.inputs.overwrite }}
 
 
     steps:
       # Step: Decide the commit hash to use
       # Step: Checkout full history so we can check out any commit
       - name: Checkout full repo history
-        uses: actions/checkout@v3
+        uses: actions/checkout@v6
         with:
           fetch-depth: 0  # Important: so we can checkout arbitrary commit
           
       - name: Determine commit hash to checkout
         id: choose_commit
         run: |
-          if [ "${{ github.event_name}}" == "workflow_dispatch" ] && [ "${{ github.event.inputs.commit_hash }}" != "main" ]; then
-            echo "Using manual input commit: ${{ github.event.inputs.commit_hash }}"
-            echo "hash=${{ github.event.inputs.commit_hash }}" >> $GITHUB_OUTPUT
+          if [ "${{ github.event_name}}" == "workflow_dispatch" ] && [ "$INPUT_COMMIT_HASH" != "main" ]; then
+            echo "Using manual input commit: $INPUT_COMMIT_HASH"
+            echo "hash=$INPUT_COMMIT_HASH" >> $GITHUB_OUTPUT
           else
             echo "Using latest commit from main"
             echo "hash=$(git rev-parse --short HEAD)" >> $GITHUB_OUTPUT
@@ -60,10 +64,10 @@ jobs:
       - name: Replace benchmark folder from main (manual only, commit ≠ main)
         if: ${{ github.event_name == 'workflow_dispatch' && github.event.inputs.commit_hash != 'main' }}
         run: |
-          echo "Detected manual trigger with commit_hash = ${{ github.event.inputs.commit_hash }}"
+          echo "Detected manual trigger with commit_hash = $INPUT_COMMIT_HASH"
           
           # Save current branch (detached HEAD at old commit)
-          ORIG_COMMIT=${{ github.event.inputs.commit_hash }}
+          ORIG_COMMIT="$INPUT_COMMIT_HASH"
 
           # Fetch and checkout main
           git fetch origin main
@@ -72,7 +76,7 @@ jobs:
           # Save benchmark folder from main
           cp -r benchmark /tmp/benchmark_main
           # Checkout back to target commit
-          git checkout $ORIG_COMMIT
+          git checkout "$ORIG_COMMIT"
           # Replace old benchmark with one from main
           rm -rf benchmark
           cp -r /tmp/benchmark_main benchmark
@@ -85,7 +89,7 @@ jobs:
           
           if curl --output /dev/null --silent --head --fail "$BENCHMARK_URL"; then
             echo "Benchmark already exists for commit $COMMIT_HASH"
-            if [ "${{ github.event.inputs.overwrite }}" != "true" ]; then
+            if [ "$INPUT_OVERWRITE" != "true" ]; then
               echo "Overwrite is false - exiting"
               exit 1
             else
@@ -96,7 +100,7 @@ jobs:
           fi
 
       - name: Set up Python
-        uses: actions/setup-python@v3
+        uses: actions/setup-python@v6
         with:
           python-version: '3.10'
 
@@ -117,7 +121,7 @@ jobs:
         
        # Step 5: Checkout gh-pages branch in a subfolderAdd commentMore actions
       - name: Checkout gh-pages
-        uses: actions/checkout@v3
+        uses: actions/checkout@v6
         with:
           ref: gh-pages
           path: gh-pages

liger_kernel_nightly-0.6.4.dev20260121000828/.github/workflows/docs.yml

@@ -0,0 +1,64 @@
+name: Publish documentation
+on:
+  push:
+    branches:
+      - main
+    paths:
+      - 'docs/**'
+      - 'mkdocs.yml'
+
+permissions:
+  contents: write
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v6
+      - name: Configure Git Credentials
+        run: |
+          git config user.name github-actions[bot]
+          git config user.email 41898282+github-actions[bot]@users.noreply.github.com
+      - uses: actions/setup-python@v6
+        with:
+          python-version: 3.x
+      - run: echo "cache_id=$(date --utc '+%V')" >> $GITHUB_ENV 
+      - uses: actions/cache@v5
+        with:
+          key: mkdocs-material-${{ env.cache_id }}
+          path: .cache
+          restore-keys: |
+            mkdocs-material-
+      - run: pip install mkdocs-material mkdocstrings[python]
+      # ====== Backup the benchmarks from gh-pages ======
+      # This is necessary because the benchmarks are not included in the documentation build process.
+      # So we need to backup the benchmarks from gh-pages and restore them after the documentation is built.
+      - name: Backup benchmarks from gh-pages
+        run: |
+          git fetch origin gh-pages
+          # create worktree bound to local gh-pages, tracking origin/gh-pages
+          git branch -f gh-pages origin/gh-pages || true
+          mkdir -p ghp && git worktree add ghp gh-pages || true
+          if [ -d ghp/benchmarks ]; then
+            tar -C ghp -czf /tmp/benchmarks.tgz benchmarks
+          fi
+          # IMPORTANT: remove worktree so gh-pages isn't checked out anywhere
+          git worktree remove ghp --force || true
+          echo "Backed up benchmarks from gh-pages"
+      # ====== Deploy the documentation ======
+      - name: Deploy documentation
+        run: mkdocs gh-deploy --force
+      # ====== Restore the benchmarks onto gh-pages ======
+      # This is necessary because the benchmarks are not included in the documentation build process.
+      # So we need to restore the benchmarks onto gh-pages after the documentation is built.
+      - name: Restore benchmarks onto gh-pages
+        run: |
+          # Refresh remote tracking and recreate a clean worktree
+          git fetch origin gh-pages
+          git worktree add -B gh-pages ghp origin/gh-pages
+          if [ -f /tmp/benchmarks.tgz ]; then
+            tar -C ghp -xzf /tmp/benchmarks.tgz
+            git -C ghp add -A
+            git -C ghp commit -m "Restore benchmarks after gh-deploy" || echo "No changes"
+            git -C ghp push origin gh-pages
+          fi
+          git worktree remove ghp --force || true

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/intel-ci.yml

@@ -26,10 +26,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 
@@ -58,7 +58,7 @@ jobs:
         apt-get clean && rm -rf /var/lib/apt/lists/*
 
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
       
     - name: Setup Dependencies
       shell: bash

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/nvi-ci.yml

@@ -25,10 +25,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 
@@ -49,10 +49,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 
@@ -75,10 +75,10 @@ jobs:
 
     steps:
     - name: Checkout code
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/publish-nightly.yml

@@ -13,10 +13,10 @@ jobs:
 
     steps:
     - name: Checkout repository
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.8'
 

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/.github/workflows/publish-release.yml

@@ -10,10 +10,10 @@ jobs:
 
     steps:
     - name: Checkout repository
-      uses: actions/checkout@v3
+      uses: actions/checkout@v6
 
     - name: Set up Python
-      uses: actions/setup-python@v3
+      uses: actions/setup-python@v6
       with:
         python-version: '3.10'
 

liger_kernel_nightly-0.6.4.dev20260121000828/.pre-commit-config.yaml

@@ -0,0 +1,10 @@
+repos:
+- repo: https://github.com/astral-sh/ruff-pre-commit
+  # Ruff version.
+  rev: v0.14.11
+  hooks:
+    # Run the linter.
+    - id: ruff-check
+      args: [ --fix ]
+    # Run the formatter.
+    - id: ruff-format

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: liger_kernel_nightly
-Version: 0.6.2.dev20251005233549
+Version: 0.6.4.dev20260121000828
 Summary: Efficient Triton kernels for LLM Training
 License: BSD 2-CLAUSE LICENSE
         Copyright 2024 LinkedIn Corporation 
@@ -45,6 +45,7 @@ Requires-Dist: datasets>=2.19.2; extra == "dev"
 Requires-Dist: seaborn; extra == "dev"
 Requires-Dist: mkdocs-material; extra == "dev"
 Requires-Dist: torchvision>=0.20; extra == "dev"
+Requires-Dist: prek>=0.2.28; extra == "dev"
 
 <a name="readme-top"></a>
 
@@ -79,8 +80,8 @@ Requires-Dist: torchvision>=0.20; extra == "dev"
             </a>
         </td>
         <td style="padding: 10px;">
-            <a href="https://discord.gg/gpumode">
-                <img src="https://dcbadge.limes.pink/api/server/gpumode?style=flat" alt="Join Our Discord">
+            <a href="https://discord.gg/X4MaxPgA">
+                <img src="https://dcbadge.limes.pink/api/server/https://discord.gg/X4MaxPgA?style=flat" alt="Join Our Discord">
             </a>
         </td>
     </tr>
@@ -95,6 +96,7 @@ Requires-Dist: torchvision>=0.20; extra == "dev"
 <details>
   <summary>Latest News 🔥</summary>
 
+  - [2025/12/19] We announced a liger kernel discord channel at https://discord.gg/X4MaxPgA; We will be hosting Liger Kernel x Triton China Meetup in mid of January 2026
   - [2025/03/06] We release a joint blog post on TorchTune × Liger - [Peak Performance, Minimized Memory: Optimizing torchtune’s performance with torch.compile & Liger Kernel](https://pytorch.org/blog/peak-performance-minimized-memory/)
   - [2024/12/11] We release [v0.5.0](https://github.com/linkedin/Liger-Kernel/releases/tag/v0.5.0): 80% more memory efficient post training losses (DPO, ORPO, CPO, etc)!
   - [2024/12/5] We release LinkedIn Engineering Blog - [Liger-Kernel: Empowering an open source ecosystem of Triton Kernels for Efficient LLM Training](https://www.linkedin.com/blog/engineering/open-source/liger-kernel-open-source-ecosystem-for-efficient-llm-training)
@@ -113,6 +115,8 @@ We've also added optimized Post-Training kernels that deliver **up to 80% memory
 
 You can view the documentation site for additional installation, usage examples, and API references:https://linkedin.github.io/Liger-Kernel/
 
+You can view the Liger Kernel Technical Report: https://openreview.net/forum?id=36SjAIT42G
+
 ## Supercharge Your Model with Liger Kernel
 
 ![Banner](https://raw.githubusercontent.com/linkedin/Liger-Kernel/main/docs/images/banner.GIF)
@@ -310,8 +314,12 @@ loss.backward()
 | Phi3 & Phi3.5       | `liger_kernel.transformers.apply_liger_kernel_to_phi3`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy         |
 | Granite 3.0 & 3.1   | `liger_kernel.transformers.apply_liger_kernel_to_granite`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss |
 | OLMo2   | `liger_kernel.transformers.apply_liger_kernel_to_olmo2`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| Olmo3   | `liger_kernel.transformers.apply_liger_kernel_to_olmo3`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
 | GLM-4   | `liger_kernel.transformers.apply_liger_kernel_to_glm4`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| GPT-OSS   | `liger_kernel.transformers.apply_liger_kernel_to_gpt_oss`     | RoPE, RMSNorm, CrossEntropyLoss, FusedLinearCrossEntropy |
 | InternVL3   | `liger_kernel.transformers.apply_liger_kernel_to_internvl`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| HunyuanV1   | `liger_kernel.transformers.apply_liger_kernel_to_hunyuan_v1_dense`    |  RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy       |
+| HunyuanV1 MoE | `liger_kernel.transformers.apply_liger_kernel_to_hunyuan_v1_moe` | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy       |
 
 
 ## Low-level APIs
@@ -438,3 +446,5 @@ url={https://openreview.net/forum?id=36SjAIT42G}
         ↑ Back to Top ↑
     </a>
 </p>
+
+

{liger_kernel_nightly-0.6.2.dev20251005233549 → liger_kernel_nightly-0.6.4.dev20260121000828}/README.md

@@ -31,8 +31,8 @@
             </a>
         </td>
         <td style="padding: 10px;">
-            <a href="https://discord.gg/gpumode">
-                <img src="https://dcbadge.limes.pink/api/server/gpumode?style=flat" alt="Join Our Discord">
+            <a href="https://discord.gg/X4MaxPgA">
+                <img src="https://dcbadge.limes.pink/api/server/https://discord.gg/X4MaxPgA?style=flat" alt="Join Our Discord">
             </a>
         </td>
     </tr>
@@ -47,6 +47,7 @@
 <details>
   <summary>Latest News 🔥</summary>
 
+  - [2025/12/19] We announced a liger kernel discord channel at https://discord.gg/X4MaxPgA; We will be hosting Liger Kernel x Triton China Meetup in mid of January 2026
   - [2025/03/06] We release a joint blog post on TorchTune × Liger - [Peak Performance, Minimized Memory: Optimizing torchtune’s performance with torch.compile & Liger Kernel](https://pytorch.org/blog/peak-performance-minimized-memory/)
   - [2024/12/11] We release [v0.5.0](https://github.com/linkedin/Liger-Kernel/releases/tag/v0.5.0): 80% more memory efficient post training losses (DPO, ORPO, CPO, etc)!
   - [2024/12/5] We release LinkedIn Engineering Blog - [Liger-Kernel: Empowering an open source ecosystem of Triton Kernels for Efficient LLM Training](https://www.linkedin.com/blog/engineering/open-source/liger-kernel-open-source-ecosystem-for-efficient-llm-training)
@@ -65,6 +66,8 @@ We've also added optimized Post-Training kernels that deliver **up to 80% memory
 
 You can view the documentation site for additional installation, usage examples, and API references:https://linkedin.github.io/Liger-Kernel/
 
+You can view the Liger Kernel Technical Report: https://openreview.net/forum?id=36SjAIT42G
+
 ## Supercharge Your Model with Liger Kernel
 
 ![Banner](https://raw.githubusercontent.com/linkedin/Liger-Kernel/main/docs/images/banner.GIF)
@@ -262,8 +265,12 @@ loss.backward()
 | Phi3 & Phi3.5       | `liger_kernel.transformers.apply_liger_kernel_to_phi3`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy         |
 | Granite 3.0 & 3.1   | `liger_kernel.transformers.apply_liger_kernel_to_granite`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss |
 | OLMo2   | `liger_kernel.transformers.apply_liger_kernel_to_olmo2`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| Olmo3   | `liger_kernel.transformers.apply_liger_kernel_to_olmo3`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
 | GLM-4   | `liger_kernel.transformers.apply_liger_kernel_to_glm4`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| GPT-OSS   | `liger_kernel.transformers.apply_liger_kernel_to_gpt_oss`     | RoPE, RMSNorm, CrossEntropyLoss, FusedLinearCrossEntropy |
 | InternVL3   | `liger_kernel.transformers.apply_liger_kernel_to_internvl`     | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy |
+| HunyuanV1   | `liger_kernel.transformers.apply_liger_kernel_to_hunyuan_v1_dense`    |  RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy       |
+| HunyuanV1 MoE | `liger_kernel.transformers.apply_liger_kernel_to_hunyuan_v1_moe` | RoPE, RMSNorm, SwiGLU, CrossEntropyLoss, FusedLinearCrossEntropy       |
 
 
 ## Low-level APIs
@@ -390,3 +397,5 @@ url={https://openreview.net/forum?id=36SjAIT42G}
         ↑ Back to Top ↑
     </a>
 </p>
+
+