fix: update for ci build

drbh · drbh · commit fa36e16e01be · 2025-10-23T16:57:16.000Z
diff --git a/.github/workflows/deploy-to-huggingface.yml b/.github/workflows/deploy-to-huggingface.yml
@@ -36,31 +36,40 @@ jobs:
             git curl ca-certificates build-essential
           rm -rf /var/lib/apt/lists/*
 
+      - name: Install Node.js (for act)
+        if: ${{ env.ACT }}
+        run: |
+          curl -fsSL https://deb.nodesource.com/setup_20.x | bash -
+          apt-get install -y nodejs
+
       - name: Install uv
         uses: astral-sh/setup-uv@v3
         with:
           version: latest
 
       - name: Prepare Python 3.11
         run: |
+          echo "Installing Python 3.11"
           uv python install 3.11
           echo "UV_PYTHON=3.11" >> "$GITHUB_ENV"
 
       - name: Build site with uvnote (GPU available)
         run: |
           # If your build uses CUDA, it will see the GPUs here.
-          uvx https://github.com/drbh/uvnote.git build benches
+          # --strict will fail if any cells do not succeed.
+          uvx https://github.com/drbh/uvnote.git build benches --strict
 
       - name: Deploy to Hugging Face Space
         env:
           HF_TOKEN: ${{ secrets.HF_TOKEN }}
         run: |
           # Upload the built site/ directory to the Space, cleaning removed files
-          uvx --from huggingface-hub hf upload \
-            --token "$HF_TOKEN" \
-            --delete \
-            --exclude "*.json" \
-            --exclude "*.pt" \
-            --repo-type space \
-            drbh/uvnote-book-explore \
-            site/ /
+          echo "Deploying to Hugging Face Space drbh/uvnote-book-explore"
+          # uvx --from huggingface-hub hf upload \
+          #   --token "$HF_TOKEN" \
+          #   --delete \
+          #   --exclude "*.json" \
+          #   --exclude "*.pt" \
+          #   --repo-type space \
+          #   drbh/uvnote-book-explore \
+          #   site/ /
diff --git a/README.md b/README.md
@@ -0,0 +1,4 @@
+
+
+
+https://huggingface.co/spaces/drbh/uvnote-book-explore
diff --git a/benches/activation/impls/compiled_swiglu.md b/benches/activation/impls/compiled_swiglu.md
@@ -18,12 +18,12 @@ print(subprocess.run(["nvidia-smi"], capture_output=True, text=True).stdout)
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/activation/impls/hf_kernels_swiglu.md b/benches/activation/impls/hf_kernels_swiglu.md
@@ -18,13 +18,13 @@ print(subprocess.run(["nvidia-smi"], capture_output=True, text=True).stdout)
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 #     "kernels",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/activation/impls/torch_swiglu.md b/benches/activation/impls/torch_swiglu.md
@@ -18,12 +18,12 @@ print(subprocess.run(["nvidia-smi"], capture_output=True, text=True).stdout)
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/activation/results/combined_results.md b/benches/activation/results/combined_results.md
@@ -20,13 +20,13 @@ This document combines benchmark results from multiple SwiGLU activation impleme
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 #     "matplotlib",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 from kernels_benchmark_tools.core.visuals import generate_combined_results
 
diff --git a/benches/flash_attn/impls/compiled_variants.md b/benches/flash_attn/impls/compiled_variants.md
@@ -14,12 +14,12 @@ This file benchmarks Flash Attention with different torch.compile modes.
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
@@ -90,12 +90,12 @@ if __name__ == "__main__":
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/flash_attention.md b/benches/flash_attn/impls/flash_attention.md
@@ -19,12 +19,12 @@ print(subprocess.run(["nvidia-smi"], capture_output=True, text=True).stdout)
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/hf_kernels_flash_attn.md b/benches/flash_attn/impls/hf_kernels_flash_attn.md
@@ -18,7 +18,7 @@ on_huggingface: kernels-community/flash-attn2
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/hf_kernels_flash_attn3.md b/benches/flash_attn/impls/hf_kernels_flash_attn3.md
@@ -12,13 +12,13 @@ on_huggingface: kernels-community/flash-attn3
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 #     "kernels",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/mem_efficient_attention.md b/benches/flash_attn/impls/mem_efficient_attention.md
@@ -11,12 +11,12 @@ on_github: huggingface/kernels-uvnotes
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/sage_attention.md b/benches/flash_attn/impls/sage_attention.md
@@ -11,14 +11,14 @@ on_github: huggingface/kernels-uvnotes
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels",
 #     "kernels-benchmark-tools",
 #     "sageattention",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/impls/xformers.md b/benches/flash_attn/impls/xformers.md
@@ -11,13 +11,13 @@ on_github: huggingface/kernels-uvnotes
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 #     "xformers",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/benches/flash_attn/results/combined_results.md b/benches/flash_attn/results/combined_results.md
@@ -11,7 +11,7 @@ title: "Flash Attention Benchmark - Combined Results"
 # requires-python = ">=3.10"
 # dependencies = ["torch", "kernels-benchmark-tools", "matplotlib"]
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "../../../tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 from kernels_benchmark_tools.core.visuals import generate_combined_results
 
diff --git a/benches/layer_norm/impls/hf_kernels_layer_norm.md b/benches/layer_norm/impls/hf_kernels_layer_norm.md
@@ -12,13 +12,13 @@ Based on kernels-community `layer-norm` kernel.
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 from kernels import get_kernel
diff --git a/benches/layer_norm/impls/torch_layer_norm.md b/benches/layer_norm/impls/torch_layer_norm.md
@@ -17,12 +17,12 @@ print(subprocess.run(["nvidia-smi"], capture_output=True, text=True).stdout)
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import kernels_benchmark_tools as kbt
diff --git a/benches/layer_norm/results/combined_results.md b/benches/layer_norm/results/combined_results.md
@@ -11,7 +11,7 @@ title: "LayerNorm Benchmark - Combined Results"
 # requires-python = ">=3.10"
 # dependencies = ["torch", "kernels-benchmark-tools", "matplotlib"]
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "../../../tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 from kernels_benchmark_tools.core.visuals import generate_combined_results
 
diff --git a/docs/how_to_add_a_benchmark.md b/docs/how_to_add_a_benchmark.md
@@ -155,7 +155,7 @@ Create `benches/activation/impls/hf_kernels_swiglu.md`:
 # requires-python = ">=3.10"
 # dependencies = ["torch", "kernels-benchmark-tools", "kernels"]
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "../../tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import kernels_benchmark_tools as kbt
@@ -207,7 +207,7 @@ Create `benches/activation/results/combined_results.md`:
 # requires-python = ">=3.10"
 # dependencies = ["torch", "kernels-benchmark-tools", "matplotlib"]
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "../../../tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 from kernels_benchmark_tools.core.visuals import generate_combined_results
 
diff --git a/tools/examples/flash.py b/tools/examples/flash.py
@@ -2,12 +2,12 @@
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/tools/examples/moe.py b/tools/examples/moe.py
@@ -2,12 +2,12 @@
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import torch.nn.functional as F
diff --git a/tools/examples/rms.py b/tools/examples/rms.py
@@ -2,12 +2,12 @@
 # requires-python = ">=3.10"
 # dependencies = [
 #     "numpy",
-#     "torch",
+#     "torch==2.8.0",
 #     "kernels-benchmark-tools",
 # ]
 #
 # [tool.uv.sources]
-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }
+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }
 # ///
 import torch
 import sys
diff --git a/tools/kernels_benchmark_tools/layer_norm.py b/tools/kernels_benchmark_tools/layer_norm.py
@@ -27,9 +27,15 @@ def gen_inputs(wl: dict) -> Sequence[torch.Tensor]:
 def ref_layer_norm(inputs: Sequence[torch.Tensor], eps: float = 1e-5) -> torch.Tensor:
     """Reference LayerNorm in float32 with affine (weight, bias)."""
     x, weight, bias = inputs
-    x_f32 = x.to(torch.float32)
-    w_f32 = weight.to(torch.float32)
-    b_f32 = bias.to(torch.float32)
+    # x_f32 = x.to(torch.float32)
+    # w_f32 = weight.to(torch.float32)
+    # b_f32 = bias.to(torch.float32)
+
+    # make them bfloat16/float16 compatible
+    x_f32 = x.bfloat16()
+    w_f32 = weight.bfloat16()
+    b_f32 = bias.bfloat16()
+
     mean = x_f32.mean(dim=-1, keepdim=True)
     var = x_f32.var(dim=-1, keepdim=True, unbiased=False)
     y = (x_f32 - mean) * torch.rsqrt(var + eps)

-Original file line number
+Diff line change
@@ @@ -0,0 +1,4 @@ @@
++
++
++
 +https://huggingface.co/spaces/drbh/uvnote-book-explore
Original file line number	Diff line number	Diff line change
`@@ -18,7 +18,7 @@ on_huggingface: kernels-community/flash-attn2`
`18`	`18`	`# ]`
`19`	`19`	`#`
`20`	`20`	`# [tool.uv.sources]`
`21`		`-# kernels-benchmark-tools = { path = "/home/ubuntu/Projects/kernels-benchmarks-consolidated/tools", editable = true }`
	`21`	`+# kernels-benchmark-tools = { path = "../../../../../tools", editable = true }`
`22`	`22`	`# ///`
`23`	`23`	`import torch`
`24`	`24`	`import sys`