Setup the dataset structure

Signed-off-by: Ubospica <ubospica@gmail.com>

README.md

@@ -1,3 +1,42 @@
 ---
 license: apache-2.0
 ---
+
+# MLSys 2026 FlashInfer-Bench Challenge Dataset
+
+This repository contains the FlashInfer-Bench dataset for the MLSys 2026 Kenrel Generation Challenge.
+
+This dataset targets to be used in the [FlashInfer-Bench](https://github.com/flashinfer-ai/flashinfer-bench) benchmark system.
+
+It follows the [FlashInfer Trace Schema](https://bench.flashinfer.ai/docs/flashinfer-trace). To use the dataset in the competition, please refer to our [starter kit](https://github.com/flashinfer-ai/flashinfer-bench-starter-kit).
+
+## Download
+
+Use this command to download the dataset:
+```bash
+git clone https://huggingface.co/datasets/flashinfer-ai/mlsys26-contest
+```
+
+## Tasks
+
+This dataset contains the definitions and workloads for these kernels:
+* Fused Mixture of Experts (MoE)
+* Gated Delta Network (GDN)
+* DeepSeek Sparse Attention (DSA)
+
+## Dataset Structure
+
+It is organized as follows:
+```
+mlsys26-contest/
+├── definitions/
+└── workloads/
+```
+
+These components are provided in the dataset:
+* **Definition**: describes the input, output, and computation logic of a kernel task.
+* **Workload**: describes the inputs for a definition during real inference. This will be used to benchmark the **Solution** you provided.
+
+During benchmarking, these components should be provided or generated:
+* **Solution**: provided by participants, your implementation of the kernel task.
+* **Trace**: generated by FlashInfer-Bench, the performance and correctness results of your solution on the workloads.

definitions/dsa_paged/dsa_sparse_decode_h16_ckv512_kpe64_topk256_ps1.json

@@ -0,0 +1,121 @@
+{
+  "name": "dsa_sparse_decode_h16_ckv512_kpe64_topk256_ps1",
+  "description": "Batched Native Sparse Attention (DSA) decode with sparse TopK KV cache selection. Captured from DeepSeek-V3 with tensor parallel size 8. Uses sparse indexing to select only top-K KV cache entries for attention computation.",
+  "op_type": "dsa_paged",
+  "tags": [
+    "stage:decode",
+    "status:verified",
+    "model:deepseek-v3",
+    "model:deepseek-r1",
+    "sparse:topk"
+  ],
+  "axes": {
+    "batch_size": {
+      "type": "var",
+      "description": "Batch size (number of sequences)."
+    },
+    "num_qo_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads after tensor parallel split (128/8=16)."
+    },
+    "head_dim_ckv": {
+      "type": "const",
+      "value": 512,
+      "description": "Compressed KV head dimension."
+    },
+    "head_dim_kpe": {
+      "type": "const",
+      "value": 64,
+      "description": "Key positional encoding dimension."
+    },
+    "page_size": {
+      "type": "const",
+      "value": 1,
+      "description": "Page size for KV cache (token-level)."
+    },
+    "topk": {
+      "type": "const",
+      "value": 256,
+      "description": "Number of top-K KV cache entries selected for sparse attention."
+    },
+    "num_pages": {
+      "type": "var",
+      "description": "Total number of allocated pages in the KV cache."
+    }
+  },
+  "constraints": [
+    "sparse_indices.shape[-1] == topk",
+    "ckv_cache.shape[1] == page_size"
+  ],
+  "inputs": {
+    "q_nope": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query tensor without positional encoding component."
+    },
+    "q_pe": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query positional encoding component."
+    },
+    "ckv_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Compressed key-value cache with page_size=1."
+    },
+    "kpe_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Key positional encoding cache."
+    },
+    "sparse_indices": {
+      "shape": [
+        "batch_size",
+        "topk"
+      ],
+      "dtype": "int32",
+      "description": "Sparse indices selecting top-K KV cache entries for each batch element. Values of -1 indicate padding (invalid indices)."
+    },
+    "sm_scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Softmax scale. For MLA, uses pre-absorption head dimension: 1/sqrt(head_dim_qk + head_dim_kpe) = 1/sqrt(128 + 64) = 1/sqrt(192)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16"
+    },
+    "lse": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads"
+      ],
+      "dtype": "float32",
+      "description": "The 2-based log-sum-exp of attention logits."
+    }
+  },
+  "reference": "import math\nimport torch\n\n\n@torch.no_grad()\ndef run(q_nope, q_pe, ckv_cache, kpe_cache, sparse_indices, sm_scale):\n    batch_size, num_qo_heads, head_dim_ckv = q_nope.shape\n    head_dim_kpe = q_pe.shape[-1]\n    page_size = ckv_cache.shape[1]\n    topk = sparse_indices.shape[-1]\n\n    # Check constants\n    assert num_qo_heads == 16\n    assert head_dim_ckv == 512\n    assert head_dim_kpe == 64\n    assert page_size == 1\n    assert topk == 256\n\n    # Check constraints\n    assert sparse_indices.shape[-1] == topk\n    assert ckv_cache.shape[1] == page_size\n\n    device = q_nope.device\n\n    # Squeeze page dimension (page_size=1)\n    Kc_all = ckv_cache.squeeze(1).to(torch.float32)  # [num_pages, head_dim_ckv]\n    Kp_all = kpe_cache.squeeze(1).to(torch.float32)  # [num_pages, head_dim_kpe]\n\n    output = torch.zeros(\n        (batch_size, num_qo_heads, head_dim_ckv), dtype=torch.bfloat16, device=device\n    )\n    lse = torch.full((batch_size, num_qo_heads), -float(\"inf\"), dtype=torch.float32, device=device)\n\n    for b in range(batch_size):\n        indices = sparse_indices[b]  # [topk]\n\n        # Handle padding: -1 indicates invalid indices\n        valid_mask = indices != -1\n        valid_indices = indices[valid_mask]\n\n        if valid_indices.numel() == 0:\n            output[b].zero_()\n            continue\n\n        tok_idx = valid_indices.to(torch.long)\n\n        Kc = Kc_all[tok_idx]  # [num_valid, head_dim_ckv]\n        Kp = Kp_all[tok_idx]  # [num_valid, head_dim_kpe]\n        qn = q_nope[b].to(torch.float32)  # [num_qo_heads, head_dim_ckv]\n        qp = q_pe[b].to(torch.float32)  # [num_qo_heads, head_dim_kpe]\n\n        # Compute attention logits\n        logits = (qn @ Kc.T) + (qp @ Kp.T)  # [num_qo_heads, num_valid]\n        logits_scaled = logits * sm_scale\n\n        # Compute 2-base LSE\n        lse[b] = torch.logsumexp(logits_scaled, dim=-1) / math.log(2.0)\n\n        # Compute attention output\n        attn = torch.softmax(logits_scaled, dim=-1)  # [num_qo_heads, num_valid]\n        out = attn @ Kc  # [num_qo_heads, head_dim_ckv]\n        output[b] = out.to(torch.bfloat16)\n\n    return output, lse"
+}

definitions/dsa_paged/dsa_sparse_decode_h16_ckv512_kpe64_topk256_ps64.json

@@ -0,0 +1,121 @@
+{
+  "name": "dsa_sparse_decode_h16_ckv512_kpe64_topk256_ps64",
+  "description": "Batched Native Sparse Attention (DSA) decode with sparse TopK KV cache selection. Captured from DeepSeek-V3 with tensor parallel size 8. Uses sparse indexing to select only top-K KV cache entries for attention computation. Page size 64 variant.",
+  "op_type": "dsa_paged",
+  "tags": [
+    "stage:decode",
+    "status:verified",
+    "model:deepseek-v3",
+    "model:deepseek-r1",
+    "sparse:topk"
+  ],
+  "axes": {
+    "batch_size": {
+      "type": "var",
+      "description": "Batch size (number of sequences)."
+    },
+    "num_qo_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads after tensor parallel split (128/8=16)."
+    },
+    "head_dim_ckv": {
+      "type": "const",
+      "value": 512,
+      "description": "Compressed KV head dimension."
+    },
+    "head_dim_kpe": {
+      "type": "const",
+      "value": 64,
+      "description": "Key positional encoding dimension."
+    },
+    "page_size": {
+      "type": "const",
+      "value": 64,
+      "description": "Page size for KV cache (64 tokens per page)."
+    },
+    "topk": {
+      "type": "const",
+      "value": 256,
+      "description": "Number of top-K KV cache entries selected for sparse attention."
+    },
+    "num_pages": {
+      "type": "var",
+      "description": "Total number of allocated pages in the KV cache."
+    }
+  },
+  "constraints": [
+    "sparse_indices.shape[-1] == topk",
+    "ckv_cache.shape[1] == page_size"
+  ],
+  "inputs": {
+    "q_nope": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query tensor without positional encoding component."
+    },
+    "q_pe": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query positional encoding component."
+    },
+    "ckv_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Compressed key-value cache with page_size=64."
+    },
+    "kpe_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Key positional encoding cache."
+    },
+    "sparse_indices": {
+      "shape": [
+        "batch_size",
+        "topk"
+      ],
+      "dtype": "int32",
+      "description": "Sparse indices selecting top-K KV cache entries for each batch element. Values of -1 indicate padding (invalid indices). For page_size=64, indices encode (page_idx * 64 + offset)."
+    },
+    "sm_scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Softmax scale. For MLA, uses pre-absorption head dimension: 1/sqrt(head_dim_qk + head_dim_kpe) = 1/sqrt(128 + 64) = 1/sqrt(192)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16"
+    },
+    "lse": {
+      "shape": [
+        "batch_size",
+        "num_qo_heads"
+      ],
+      "dtype": "float32",
+      "description": "The 2-based log-sum-exp of attention logits."
+    }
+  },
+  "reference": "import math\nimport torch\n\n\n@torch.no_grad()\ndef run(q_nope, q_pe, ckv_cache, kpe_cache, sparse_indices, sm_scale):\n    batch_size, num_qo_heads, head_dim_ckv = q_nope.shape\n    head_dim_kpe = q_pe.shape[-1]\n    num_pages, page_size, _ = ckv_cache.shape\n    topk = sparse_indices.shape[-1]\n\n    # Check constants\n    assert num_qo_heads == 16\n    assert head_dim_ckv == 512\n    assert head_dim_kpe == 64\n    assert page_size == 64\n    assert topk == 256\n\n    # Check constraints\n    assert sparse_indices.shape[-1] == topk\n    assert ckv_cache.shape[1] == page_size\n\n    device = q_nope.device\n\n    # Flatten paged KV cache to token-level: [num_pages, page_size, dim] -> [num_pages * page_size, dim]\n    Kc_all = ckv_cache.reshape(-1, head_dim_ckv).to(torch.float32)  # [total_kv_tokens, head_dim_ckv]\n    Kp_all = kpe_cache.reshape(-1, head_dim_kpe).to(torch.float32)  # [total_kv_tokens, head_dim_kpe]\n\n    output = torch.zeros(\n        (batch_size, num_qo_heads, head_dim_ckv), dtype=torch.bfloat16, device=device\n    )\n    lse = torch.full((batch_size, num_qo_heads), -float(\"inf\"), dtype=torch.float32, device=device)\n\n    for b in range(batch_size):\n        indices = sparse_indices[b]  # [topk]\n\n        # Handle padding: -1 indicates invalid indices\n        valid_mask = indices != -1\n        valid_indices = indices[valid_mask]\n\n        if valid_indices.numel() == 0:\n            output[b].zero_()\n            continue\n\n        # For page_size=64, indices encode (page_idx * 64 + offset)\n        tok_idx = valid_indices.to(torch.long)\n\n        Kc = Kc_all[tok_idx]  # [num_valid, head_dim_ckv]\n        Kp = Kp_all[tok_idx]  # [num_valid, head_dim_kpe]\n        qn = q_nope[b].to(torch.float32)  # [num_qo_heads, head_dim_ckv]\n        qp = q_pe[b].to(torch.float32)  # [num_qo_heads, head_dim_kpe]\n\n        # Compute attention logits\n        logits = (qn @ Kc.T) + (qp @ Kp.T)  # [num_qo_heads, num_valid]\n        logits_scaled = logits * sm_scale\n\n        # Compute 2-base LSE\n        lse[b] = torch.logsumexp(logits_scaled, dim=-1) / math.log(2.0)\n\n        # Compute attention output\n        attn = torch.softmax(logits_scaled, dim=-1)  # [num_qo_heads, num_valid]\n        out = attn @ Kc  # [num_qo_heads, head_dim_ckv]\n        output[b] = out.to(torch.bfloat16)\n\n    return {\"output\": output, \"lse\": lse}"
+}

definitions/dsa_paged/dsa_sparse_prefill_causal_h16_ckv512_kpe64_topk256_ps1.json

@@ -0,0 +1,123 @@
+{
+  "name": "dsa_sparse_prefill_causal_h16_ckv512_kpe64_topk256_ps1",
+  "description": "Batched Native Sparse Attention (DSA) prefill with causal masking and sparse TopK KV cache selection. Captured from DeepSeek-V3 with tensor parallel size 8. Uses sparse indexing to select only top-K KV cache entries for attention computation during prefill.",
+  "op_type": "dsa_paged",
+  "tags": [
+    "stage:prefill",
+    "status:verified",
+    "model:deepseek-v3",
+    "model:deepseek-r1",
+    "sparse:topk",
+    "mask:causal"
+  ],
+  "axes": {
+    "total_num_tokens": {
+      "type": "var",
+      "description": "Total number of tokens across all sequences in the batch."
+    },
+    "num_qo_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads after tensor parallel split (128/8=16)."
+    },
+    "head_dim_ckv": {
+      "type": "const",
+      "value": 512,
+      "description": "Compressed KV head dimension."
+    },
+    "head_dim_kpe": {
+      "type": "const",
+      "value": 64,
+      "description": "Key positional encoding dimension."
+    },
+    "page_size": {
+      "type": "const",
+      "value": 1,
+      "description": "Page size for KV cache (token-level)."
+    },
+    "topk": {
+      "type": "const",
+      "value": 256,
+      "description": "Number of top-K KV cache entries selected for sparse attention per token."
+    },
+    "num_pages": {
+      "type": "var",
+      "description": "Total number of allocated pages in the KV cache."
+    }
+  },
+  "constraints": [
+    "sparse_indices.shape[0] == total_num_tokens",
+    "sparse_indices.shape[-1] == topk",
+    "ckv_cache.shape[1] == page_size"
+  ],
+  "inputs": {
+    "q_nope": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query tensor without positional encoding component."
+    },
+    "q_pe": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query positional encoding component."
+    },
+    "ckv_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Compressed key-value cache with page_size=1."
+    },
+    "kpe_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Key positional encoding cache."
+    },
+    "sparse_indices": {
+      "shape": [
+        "total_num_tokens",
+        "topk"
+      ],
+      "dtype": "int32",
+      "description": "Sparse indices selecting top-K KV cache entries for each token. Values of -1 indicate padding (invalid indices)."
+    },
+    "sm_scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Softmax scale. For MLA, uses pre-absorption head dimension: 1/sqrt(head_dim_qk + head_dim_kpe) = 1/sqrt(128 + 64) = 1/sqrt(192)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16"
+    },
+    "lse": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads"
+      ],
+      "dtype": "float32",
+      "description": "The 2-based log-sum-exp of attention logits."
+    }
+  },
+  "reference": "import math\nimport torch\n\n\n@torch.no_grad()\ndef run(q_nope, q_pe, ckv_cache, kpe_cache, sparse_indices, sm_scale):\n    total_num_tokens, num_qo_heads, head_dim_ckv = q_nope.shape\n    head_dim_kpe = q_pe.shape[-1]\n    page_size = ckv_cache.shape[1]\n    topk = sparse_indices.shape[-1]\n\n    # Check constants\n    assert num_qo_heads == 16\n    assert head_dim_ckv == 512\n    assert head_dim_kpe == 64\n    assert page_size == 1\n    assert topk == 256\n\n    # Check constraints\n    assert sparse_indices.shape[0] == total_num_tokens\n    assert sparse_indices.shape[-1] == topk\n    assert ckv_cache.shape[1] == page_size\n\n    device = q_nope.device\n\n    # Squeeze page dimension (page_size=1)\n    Kc_all = ckv_cache.squeeze(1).to(torch.float32)  # [num_pages, head_dim_ckv]\n    Kp_all = kpe_cache.squeeze(1).to(torch.float32)  # [num_pages, head_dim_kpe]\n\n    output = torch.zeros(\n        (total_num_tokens, num_qo_heads, head_dim_ckv), dtype=torch.bfloat16, device=device\n    )\n    lse = torch.full((total_num_tokens, num_qo_heads), -float(\"inf\"), dtype=torch.float32, device=device)\n\n    for t in range(total_num_tokens):\n        indices = sparse_indices[t]  # [topk]\n\n        # Handle padding: -1 indicates invalid indices\n        valid_mask = indices != -1\n        valid_indices = indices[valid_mask]\n\n        if valid_indices.numel() == 0:\n            output[t].zero_()\n            continue\n\n        tok_idx = valid_indices.to(torch.long)\n\n        Kc = Kc_all[tok_idx]  # [num_valid, head_dim_ckv]\n        Kp = Kp_all[tok_idx]  # [num_valid, head_dim_kpe]\n        qn = q_nope[t].to(torch.float32)  # [num_qo_heads, head_dim_ckv]\n        qp = q_pe[t].to(torch.float32)  # [num_qo_heads, head_dim_kpe]\n\n        # Compute attention logits\n        logits = (qn @ Kc.T) + (qp @ Kp.T)  # [num_qo_heads, num_valid]\n        logits_scaled = logits * sm_scale\n\n        # Compute 2-base LSE\n        lse[t] = torch.logsumexp(logits_scaled, dim=-1) / math.log(2.0)\n\n        # Compute attention output\n        attn = torch.softmax(logits_scaled, dim=-1)  # [num_qo_heads, num_valid]\n        out = attn @ Kc  # [num_qo_heads, head_dim_ckv]\n        output[t] = out.to(torch.bfloat16)\n\n    return output, lse"
+}

definitions/dsa_paged/dsa_sparse_prefill_causal_h16_ckv512_kpe64_topk256_ps64.json

@@ -0,0 +1,123 @@
+{
+  "name": "dsa_sparse_prefill_causal_h16_ckv512_kpe64_topk256_ps64",
+  "description": "Batched Native Sparse Attention (DSA) prefill with causal masking and sparse TopK KV cache selection. Captured from DeepSeek-V3 with tensor parallel size 8. Uses sparse indexing to select only top-K KV cache entries for attention computation during prefill. Page size 64 variant.",
+  "op_type": "dsa_paged",
+  "tags": [
+    "stage:prefill",
+    "status:verified",
+    "model:deepseek-v3",
+    "model:deepseek-r1",
+    "sparse:topk",
+    "mask:causal"
+  ],
+  "axes": {
+    "total_num_tokens": {
+      "type": "var",
+      "description": "Total number of tokens across all sequences in the batch."
+    },
+    "num_qo_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads after tensor parallel split (128/8=16)."
+    },
+    "head_dim_ckv": {
+      "type": "const",
+      "value": 512,
+      "description": "Compressed KV head dimension."
+    },
+    "head_dim_kpe": {
+      "type": "const",
+      "value": 64,
+      "description": "Key positional encoding dimension."
+    },
+    "page_size": {
+      "type": "const",
+      "value": 64,
+      "description": "Page size for KV cache (64 tokens per page)."
+    },
+    "topk": {
+      "type": "const",
+      "value": 256,
+      "description": "Number of top-K KV cache entries selected for sparse attention per token."
+    },
+    "num_pages": {
+      "type": "var",
+      "description": "Total number of allocated pages in the KV cache."
+    }
+  },
+  "constraints": [
+    "sparse_indices.shape[0] == total_num_tokens",
+    "sparse_indices.shape[-1] == topk",
+    "ckv_cache.shape[1] == page_size"
+  ],
+  "inputs": {
+    "q_nope": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query tensor without positional encoding component."
+    },
+    "q_pe": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Query positional encoding component."
+    },
+    "ckv_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16",
+      "description": "Compressed key-value cache with page_size=64."
+    },
+    "kpe_cache": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "head_dim_kpe"
+      ],
+      "dtype": "bfloat16",
+      "description": "Key positional encoding cache."
+    },
+    "sparse_indices": {
+      "shape": [
+        "total_num_tokens",
+        "topk"
+      ],
+      "dtype": "int32",
+      "description": "Sparse indices selecting top-K KV cache entries for each token. Values of -1 indicate padding (invalid indices). For page_size=64, indices encode (page_idx * 64 + offset)."
+    },
+    "sm_scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Softmax scale. For MLA, uses pre-absorption head dimension: 1/sqrt(head_dim_qk + head_dim_kpe) = 1/sqrt(128 + 64) = 1/sqrt(192)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads",
+        "head_dim_ckv"
+      ],
+      "dtype": "bfloat16"
+    },
+    "lse": {
+      "shape": [
+        "total_num_tokens",
+        "num_qo_heads"
+      ],
+      "dtype": "float32",
+      "description": "The 2-based log-sum-exp of attention logits."
+    }
+  },
+  "reference": "import math\nimport torch\n\n\n@torch.no_grad()\ndef run(q_nope, q_pe, ckv_cache, kpe_cache, sparse_indices, sm_scale):\n    total_num_tokens, num_qo_heads, head_dim_ckv = q_nope.shape\n    head_dim_kpe = q_pe.shape[-1]\n    num_pages, page_size, _ = ckv_cache.shape\n    topk = sparse_indices.shape[-1]\n\n    # Check constants\n    assert num_qo_heads == 16\n    assert head_dim_ckv == 512\n    assert head_dim_kpe == 64\n    assert page_size == 64\n    assert topk == 256\n\n    # Check constraints\n    assert sparse_indices.shape[0] == total_num_tokens\n    assert sparse_indices.shape[-1] == topk\n    assert ckv_cache.shape[1] == page_size\n\n    device = q_nope.device\n\n    # Flatten paged KV cache to token-level: [num_pages, page_size, dim] -> [num_pages * page_size, dim]\n    Kc_all = ckv_cache.reshape(-1, head_dim_ckv).to(torch.float32)  # [total_kv_tokens, head_dim_ckv]\n    Kp_all = kpe_cache.reshape(-1, head_dim_kpe).to(torch.float32)  # [total_kv_tokens, head_dim_kpe]\n\n    output = torch.zeros(\n        (total_num_tokens, num_qo_heads, head_dim_ckv), dtype=torch.bfloat16, device=device\n    )\n    lse = torch.full((total_num_tokens, num_qo_heads), -float(\"inf\"), dtype=torch.float32, device=device)\n\n    for t in range(total_num_tokens):\n        indices = sparse_indices[t]  # [topk]\n\n        # Handle padding: -1 indicates invalid indices\n        valid_mask = indices != -1\n        valid_indices = indices[valid_mask]\n\n        if valid_indices.numel() == 0:\n            output[t].zero_()\n            continue\n\n        # For page_size=64, indices encode (page_idx * 64 + offset)\n        tok_idx = valid_indices.to(torch.long)\n\n        Kc = Kc_all[tok_idx]  # [num_valid, head_dim_ckv]\n        Kp = Kp_all[tok_idx]  # [num_valid, head_dim_kpe]\n        qn = q_nope[t].to(torch.float32)  # [num_qo_heads, head_dim_ckv]\n        qp = q_pe[t].to(torch.float32)  # [num_qo_heads, head_dim_kpe]\n\n        # Compute attention logits\n        logits = (qn @ Kc.T) + (qp @ Kp.T)  # [num_qo_heads, num_valid]\n        logits_scaled = logits * sm_scale\n\n        # Compute 2-base LSE\n        lse[t] = torch.logsumexp(logits_scaled, dim=-1) / math.log(2.0)\n\n        # Compute attention output\n        attn = torch.softmax(logits_scaled, dim=-1)  # [num_qo_heads, num_valid]\n        out = attn @ Kc  # [num_qo_heads, head_dim_ckv]\n        output[t] = out.to(torch.bfloat16)\n\n    return {\"output\": output, \"lse\": lse}"
+}

definitions/dsa_paged/dsa_topk_indexer_fp8_h64_d128_topk256_ps64.json

@@ -0,0 +1,113 @@
+{
+  "name": "dsa_topk_indexer_fp8_h64_d128_topk256_ps64",
+  "description": "Native Sparse Attention (DSA) TopK indexer with FP8 quantization for DeepSeek-V3. Computes sparse attention scores using ReLU activation and learned weights, then selects top-K KV cache indices. Formula: sum(relu(q @ K.T) * weights). Matches SGLang/deep_gemm implementation. Page size 64 variant.",
+  "op_type": "dsa_paged",
+  "tags": [
+    "stage:indexer",
+    "status:verified",
+    "model:deepseek-v3",
+    "sparse:topk",
+    "quant:fp8"
+  ],
+  "axes": {
+    "batch_size": {
+      "type": "var"
+    },
+    "num_index_heads": {
+      "type": "const",
+      "value": 64,
+      "description": "Number of indexer heads (64 required by deep_gemm)."
+    },
+    "index_head_dim": {
+      "type": "const",
+      "value": 128,
+      "description": "Indexer head dimension (matches deep_gemm requirement)."
+    },
+    "page_size": {
+      "type": "const",
+      "value": 64,
+      "description": "Page size for KV cache (64 tokens per page, required by deep_gemm)."
+    },
+    "topk": {
+      "type": "const",
+      "value": 256,
+      "description": "Number of top-K indices to select."
+    },
+    "max_num_pages": {
+      "type": "var",
+      "description": "Maximum number of pages per sequence."
+    },
+    "num_pages": {
+      "type": "var",
+      "description": "Total number of allocated pages in the KV cache."
+    },
+    "kv_cache_num_heads": {
+      "type": "const",
+      "value": 1,
+      "description": "Number of heads in KV cache (always 1 for deep_gemm MQA format)."
+    },
+    "head_dim_with_scale": {
+      "type": "const",
+      "value": 132,
+      "description": "Head dimension (128) + scale bytes (4) = 132 for deep_gemm FP8 format."
+    }
+  },
+  "constraints": [
+    "topk <= max_num_pages * page_size"
+  ],
+  "inputs": {
+    "q_index_fp8": {
+      "shape": [
+        "batch_size",
+        "num_index_heads",
+        "index_head_dim"
+      ],
+      "dtype": "float8_e4m3fn",
+      "description": "FP8 quantized query tensor for indexing."
+    },
+    "k_index_cache_fp8": {
+      "shape": [
+        "num_pages",
+        "page_size",
+        "kv_cache_num_heads",
+        "head_dim_with_scale"
+      ],
+      "dtype": "int8",
+      "description": "FP8 quantized key index cache with embedded scale factors (deep_gemm format). Memory layout: all FP8 values first (page_size * 128 bytes), then all scale factors (page_size * 4 bytes). Reshaped to [num_pages, page_size, 1, 132]. Uses int8 dtype but should be interpreted as uint8."
+    },
+    "weights": {
+      "shape": [
+        "batch_size",
+        "num_index_heads"
+      ],
+      "dtype": "float32",
+      "description": "Learned weights for combining heads. In SGLang: weights = weights_proj(x) * n_heads^-0.5 * q_scale * softmax_scale."
+    },
+    "seq_lens": {
+      "shape": [
+        "batch_size"
+      ],
+      "dtype": "int32",
+      "description": "Sequence lengths for each batch element."
+    },
+    "block_table": {
+      "shape": [
+        "batch_size",
+        "max_num_pages"
+      ],
+      "dtype": "int32",
+      "description": "Page-level block table mapping batch to page indices."
+    }
+  },
+  "outputs": {
+    "topk_indices": {
+      "shape": [
+        "batch_size",
+        "topk"
+      ],
+      "dtype": "int32",
+      "description": "Top-K token indices for each batch element. Values of -1 indicate padding."
+    }
+  },
+  "reference": "import torch\n\n\ndef dequant_fp8_kv_cache(k_index_cache_fp8):\n    \"\"\"Dequantize FP8 KV cache from deep_gemm format.\n    \n    Input: [num_pages, page_size, 1, 132] int8 (interpreted as uint8)\n           Memory layout (per page): [fp8_data (page_size * 128 bytes), scales (page_size * 4 bytes)]\n           After view to [num_pages, page_size, 1, 132]: NOT directly indexable as [fp8, scale] per token!\n    Output: [num_pages, page_size, 128] float32\n    \"\"\"\n    # View as uint8 for correct byte interpretation\n    k_index_cache_fp8 = k_index_cache_fp8.view(torch.uint8)\n    num_pages, page_size, num_heads, head_dim_sf = k_index_cache_fp8.shape\n    head_dim = head_dim_sf - 4  # 128\n    \n    # Go back to flat format to reverse the packing\n    kv_flat = k_index_cache_fp8.view(num_pages, page_size * head_dim_sf)\n    \n    # FP8 part: first page_size * head_dim bytes\n    fp8_bytes = kv_flat[:, :page_size * head_dim].contiguous()\n    fp8_tensor = fp8_bytes.view(num_pages, page_size, head_dim).view(torch.float8_e4m3fn)\n    fp8_float = fp8_tensor.to(torch.float32)\n    \n    # Scale part: last page_size * 4 bytes -> page_size float32 values\n    scale_bytes = kv_flat[:, page_size * head_dim:].contiguous()\n    scale = scale_bytes.view(num_pages, page_size, 4).view(torch.float32)  # [num_pages, page_size, 1]\n    \n    return fp8_float * scale\n\n\n@torch.no_grad()\ndef run(q_index_fp8, k_index_cache_fp8, weights, seq_lens, block_table):\n    batch_size, num_index_heads, index_head_dim = q_index_fp8.shape\n    num_pages, page_size, _, _ = k_index_cache_fp8.shape\n    topk = 256\n\n    # Check constants\n    assert num_index_heads == 64\n    assert index_head_dim == 128\n    assert page_size == 64\n\n    device = q_index_fp8.device\n\n    # Dequantize inputs\n    q = q_index_fp8.to(torch.float32)  # [batch, heads, head_dim]\n    K_all = dequant_fp8_kv_cache(k_index_cache_fp8)  # [num_pages, page_size, head_dim]\n\n    topk_indices = torch.full((batch_size, topk), -1, dtype=torch.int32, device=device)\n    max_num_pages = block_table.shape[1]\n\n    for b in range(batch_size):\n        seq_len = int(seq_lens[b].item())\n        \n        if seq_len == 0:\n            continue\n\n        # Get pages for this sequence\n        num_pages_for_seq = (seq_len + page_size - 1) // page_size\n        page_indices = block_table[b, :num_pages_for_seq].to(torch.long)\n        \n        # Gather K from pages\n        K_paged = K_all[page_indices]  # [num_pages_for_seq, page_size, head_dim]\n        K = K_paged.reshape(-1, index_head_dim)[:seq_len]  # [seq_len, head_dim]\n        \n        # Query for this batch element\n        q_b = q[b]  # [num_heads, head_dim]\n        \n        # Compute attention scores\n        scores = q_b @ K.T  # [num_heads, seq_len]\n        \n        # Apply ReLU (deep_gemm uses ReLU activation)\n        scores_relu = torch.relu(scores)  # [num_heads, seq_len]\n        \n        # Apply learned weights and sum across heads\n        w = weights[b]  # [num_heads]\n        weighted_scores = scores_relu * w[:, None]  # [num_heads, seq_len]\n        final_scores = weighted_scores.sum(dim=0)  # [seq_len]\n        \n        # Select top-K\n        actual_topk = min(topk, seq_len)\n        _, topk_idx = torch.topk(final_scores, actual_topk)\n        \n        # Convert to global token indices\n        # Token index = page_idx * page_size + offset_in_page\n        page_idx_per_token = topk_idx // page_size\n        offset_per_token = topk_idx % page_size\n        global_page_idx = page_indices[page_idx_per_token]\n        topk_tokens = global_page_idx * page_size + offset_per_token\n        \n        topk_indices[b, :actual_topk] = topk_tokens.to(torch.int32)\n\n    return {\"topk_indices\": topk_indices}"
+}

definitions/gdn/gdn_decode_qk16_v32_d128_k_last.json

@@ -0,0 +1,107 @@
+{
+  "name": "gdn_decode_qk16_v32_d128_k_last",
+  "description": "Gated Delta Net decode with GVA configuration and k-last state layout. Single-token generation with recurrent state update. Captured from Qwen3 Next linear attention layers.",
+  "op_type": "gdn",
+  "tags": [
+    "stage:decode",
+    "status:verified",
+    "model:qwen3-next",
+    "layout:k-last"
+  ],
+  "axes": {
+    "batch_size": {
+      "type": "var",
+      "description": "Number of sequences being decoded concurrently."
+    },
+    "seq_len": {
+      "type": "const",
+      "value": 1,
+      "description": "Sequence length (always 1 for single-token decode)."
+    },
+    "num_q_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads (same as key heads in GVA mode)."
+    },
+    "num_k_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of key heads."
+    },
+    "num_v_heads": {
+      "type": "const",
+      "value": 32,
+      "description": "Number of value heads (GVA: more value heads than query heads)."
+    },
+    "head_size": {
+      "type": "const",
+      "value": 128
+    }
+  },
+  "constraints": [
+    "num_v_heads >= num_q_heads",
+    "num_v_heads % num_q_heads == 0",
+    "num_k_heads == num_q_heads"
+  ],
+  "inputs": {
+    "q": {
+      "shape": ["batch_size", "seq_len", "num_q_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Query tensor for single token decode."
+    },
+    "k": {
+      "shape": ["batch_size", "seq_len", "num_k_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Key tensor for single token decode."
+    },
+    "v": {
+      "shape": ["batch_size", "seq_len", "num_v_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Value tensor for single token decode."
+    },
+    "state": {
+      "shape": ["batch_size", "num_v_heads", "head_size", "head_size"],
+      "dtype": "float32",
+      "description": "Recurrent state in k-last layout [B, H, V, K].",
+      "optional": true
+    },
+    "A_log": {
+      "shape": ["num_v_heads"],
+      "dtype": "float32",
+      "description": "Log decay parameter (learnable). Used to compute g = exp(-exp(A_log) * softplus(a + dt_bias))."
+    },
+    "a": {
+      "shape": ["batch_size", "seq_len", "num_v_heads"],
+      "dtype": "bfloat16",
+      "description": "Input-dependent decay from projection."
+    },
+    "dt_bias": {
+      "shape": ["num_v_heads"],
+      "dtype": "float32",
+      "description": "Decay bias (learnable). Added to 'a' before softplus."
+    },
+    "b": {
+      "shape": ["batch_size", "seq_len", "num_v_heads"],
+      "dtype": "bfloat16",
+      "description": "Update gate input from projection. beta = sigmoid(b)."
+    },
+    "scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Scale factor. Default is 1/sqrt(head_size)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": ["batch_size", "seq_len", "num_v_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Attention output. Shape follows num_v_heads in GVA mode."
+    },
+    "new_state": {
+      "shape": ["batch_size", "num_v_heads", "head_size", "head_size"],
+      "dtype": "float32",
+      "description": "Updated recurrent state in k-last layout [B, H, V, K]."
+    }
+  },
+  "reference": "import math\nimport torch\nimport torch.nn.functional as F\n\n\ndef matmul(a: torch.Tensor, b: torch.Tensor):\n    \"\"\"Float32 matmul for numerical stability.\"\"\"\n    return a.float() @ b.float()\n\n\n@torch.no_grad()\ndef run(q, k, v, state, A_log, a, dt_bias, b, scale):\n    \"\"\"\n    Gated Delta Net decode reference implementation (k-last layout).\n    \n    State layout: [B, H, V, K] (k-last, K dimension at the end)\n    \n    Gate computation:\n    g = exp(-exp(A_log) * softplus(a + dt_bias))\n    beta = sigmoid(b)\n    \n    Delta rule update:\n    state_new = g * state_old + k^T @ (beta * v + (1-beta) * k @ state_old) - k^T @ (k @ state_old)\n    output = scale * q @ state_new\n    \"\"\"\n    B, T, num_q_heads, K = q.shape\n    _, _, num_k_heads, _ = k.shape\n    _, _, num_v_heads, V = v.shape\n    num_heads = num_v_heads\n    device = q.device\n    \n    assert num_q_heads == 16\n    assert num_k_heads == 16\n    assert num_v_heads == 32\n    assert K == 128 and V == 128\n    assert T == 1\n    \n    if scale is None or scale == 0.0:\n        scale = 1.0 / math.sqrt(K)\n    \n    # Compute g and beta from raw parameters\n    x = a.float() + dt_bias.float()  # [B, 1, HV]\n    g = torch.exp(-torch.exp(A_log.float()) * F.softplus(x))  # [B, 1, HV]\n    beta = torch.sigmoid(b.float())  # [B, 1, HV]\n    \n    q_f32 = q.squeeze(1).float()\n    k_f32 = k.squeeze(1).float()\n    v_f32 = v.squeeze(1).float()\n    g_f32 = g.squeeze(1).float()\n    beta_f32 = beta.squeeze(1).float()\n    \n    if state is not None:\n        state_f32 = state.float()\n    else:\n        state_f32 = torch.zeros(B, num_heads, V, K, dtype=torch.float32, device=device)\n    \n    q_exp = q_f32.repeat_interleave(num_v_heads // num_q_heads, dim=1)\n    k_exp = k_f32.repeat_interleave(num_v_heads // num_k_heads, dim=1)\n    \n    new_state = torch.zeros_like(state_f32)\n    output = torch.zeros(B, num_heads, V, dtype=torch.float32, device=device)\n    \n    for b_idx in range(B):\n        for h_idx in range(num_heads):\n            q_h = q_exp[b_idx, h_idx]\n            k_h = k_exp[b_idx, h_idx]\n            v_h = v_f32[b_idx, h_idx]\n            h_state = state_f32[b_idx, h_idx].clone().transpose(-1, -2)  # [V,K] -> [K,V]\n            g_val = g_f32[b_idx, h_idx]\n            beta_val = beta_f32[b_idx, h_idx]\n            \n            old_state = g_val * h_state\n            old_v = k_h @ old_state\n            new_v = beta_val * v_h + (1 - beta_val) * old_v\n            state_remove = k_h.unsqueeze(1) @ old_v.unsqueeze(0)\n            state_update = k_h.unsqueeze(1) @ new_v.unsqueeze(0)\n            h_state = old_state - state_remove + state_update\n            \n            output[b_idx, h_idx] = scale * (q_h @ h_state)\n            new_state[b_idx, h_idx] = h_state.transpose(-1, -2)  # [K,V] -> [V,K]\n    \n    output = output.unsqueeze(1).to(torch.bfloat16)\n    return {\"output\": output, \"new_state\": new_state}"
+}

definitions/gdn/gdn_prefill_qk16_v32_d128_k_last.json

@@ -0,0 +1,112 @@
+{
+  "name": "gdn_prefill_qk16_v32_d128_k_last",
+  "description": "Gated Delta Net prefill with GVA configuration and k-last state layout. The state is in k-last layout [N, H, V, K]. Captured from Qwen3 Next linear attention layers.",
+  "op_type": "gdn",
+  "tags": [
+    "stage:prefill",
+    "status:verified",
+    "model:qwen3-next",
+    "layout:k-last"
+  ],
+  "axes": {
+    "total_seq_len": {
+      "type": "var"
+    },
+    "num_seqs": {
+      "type": "var"
+    },
+    "num_q_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of query heads (same as key heads in GVA mode)."
+    },
+    "num_k_heads": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of key heads."
+    },
+    "num_v_heads": {
+      "type": "const",
+      "value": 32,
+      "description": "Number of value heads (GVA: more value heads than query heads)."
+    },
+    "head_size": {
+      "type": "const",
+      "value": 128
+    },
+    "len_cu_seqlens": {
+      "type": "var",
+      "description": "Length of cu_seqlens array (num_seqs + 1)."
+    }
+  },
+  "constraints": [
+    "len_cu_seqlens == num_seqs + 1",
+    "total_seq_len == cu_seqlens[-1].item()"
+  ],
+  "inputs": {
+    "q": {
+      "shape": ["total_seq_len", "num_q_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Query tensor."
+    },
+    "k": {
+      "shape": ["total_seq_len", "num_k_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Key tensor."
+    },
+    "v": {
+      "shape": ["total_seq_len", "num_v_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Value tensor."
+    },
+    "state": {
+      "shape": ["num_seqs", "num_v_heads", "head_size", "head_size"],
+      "dtype": "float32",
+      "description": "Recurrent state in k-last layout [N, H, V, K].",
+      "optional": true
+    },
+    "A_log": {
+      "shape": ["num_v_heads"],
+      "dtype": "float32",
+      "description": "Log decay parameter (learnable). Used to compute g = exp(-exp(A_log) * softplus(a + dt_bias))."
+    },
+    "a": {
+      "shape": ["total_seq_len", "num_v_heads"],
+      "dtype": "bfloat16",
+      "description": "Input-dependent decay from projection."
+    },
+    "dt_bias": {
+      "shape": ["num_v_heads"],
+      "dtype": "float32",
+      "description": "Decay bias (learnable). Added to 'a' before softplus."
+    },
+    "b": {
+      "shape": ["total_seq_len", "num_v_heads"],
+      "dtype": "bfloat16",
+      "description": "Update gate input from projection. beta = sigmoid(b)."
+    },
+    "cu_seqlens": {
+      "shape": ["len_cu_seqlens"],
+      "dtype": "int64",
+      "description": "Cumulative sequence lengths for variable-length batching."
+    },
+    "scale": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Scale factor. Default is 1/sqrt(head_size)."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": ["total_seq_len", "num_v_heads", "head_size"],
+      "dtype": "bfloat16",
+      "description": "Attention output. Shape follows num_v_heads in GVA mode."
+    },
+    "new_state": {
+      "shape": ["num_seqs", "num_v_heads", "head_size", "head_size"],
+      "dtype": "float32",
+      "description": "Updated recurrent state in k-last layout [N, H, V, K]."
+    }
+  },
+  "reference": "import math\nimport torch\nimport torch.nn.functional as F\n\n\ndef matmul(a: torch.Tensor, b: torch.Tensor):\n    \"\"\"Float32 matmul for numerical stability.\"\"\"\n    return a.float() @ b.float()\n\n\n@torch.no_grad()\ndef run(q, k, v, state, A_log, a, dt_bias, b, cu_seqlens, scale):\n    \"\"\"\n    Gated Delta Net prefill reference implementation (k-last layout).\n    \n    State layout: [H, V, K] (k-last, K dimension at the end)\n    \n    Gate computation:\n    g = exp(-exp(A_log) * softplus(a + dt_bias))\n    beta = sigmoid(b)\n    \n    Delta rule update:\n    state_new = g * state_old + k^T @ (beta * v + (1-beta) * k @ state_old) - k^T @ (k @ state_old)\n    output = scale * q @ state_new\n    \"\"\"\n    total_seq_len, num_q_heads, head_size = q.shape\n    num_v_heads = v.shape[1]\n    num_k_heads = k.shape[1]\n    num_sab_heads = max(num_q_heads, num_v_heads)\n    num_seqs = cu_seqlens.size(0) - 1\n    device = q.device\n\n    assert num_q_heads == 16\n    assert num_k_heads == 16\n    assert num_v_heads == 32\n    assert head_size == 128\n\n    if scale is None or scale == 0.0:\n        scale = 1.0 / math.sqrt(head_size)\n\n    # Compute g and beta from raw parameters\n    x = a.float() + dt_bias.float()  # [total_seq_len, HV]\n    g = torch.exp(-torch.exp(A_log.float()) * F.softplus(x))  # [total_seq_len, HV]\n    beta = torch.sigmoid(b.float())  # [total_seq_len, HV]\n\n    q_exp = q.repeat_interleave(num_v_heads // num_q_heads, dim=1)\n    k_exp = k.repeat_interleave(num_v_heads // num_k_heads, dim=1)\n\n    output = torch.zeros(\n        (total_seq_len, num_sab_heads, head_size), dtype=torch.bfloat16, device=device\n    )\n    new_state = torch.zeros(\n        (num_seqs, num_sab_heads, head_size, head_size), dtype=torch.float32, device=device\n    )\n\n    for seq_idx in range(num_seqs):\n        seq_start = int(cu_seqlens[seq_idx].item())\n        seq_end = int(cu_seqlens[seq_idx + 1].item())\n        seq_len = seq_end - seq_start\n\n        if seq_len <= 0:\n            continue\n\n        if state is not None:\n            state_HKV = state[seq_idx].clone().float().transpose(-1, -2)  # [H,V,K] -> [H,K,V]\n        else:\n            state_HKV = torch.zeros(\n                (num_sab_heads, head_size, head_size), dtype=torch.float32, device=device\n            )\n\n        for i in range(seq_len):\n            t = seq_start + i\n            q_H1K = q_exp[t].unsqueeze(1).float()\n            k_H1K = k_exp[t].unsqueeze(1).float()\n            v_H1V = v[t].unsqueeze(1).float()\n            g_H11 = g[t].unsqueeze(1).unsqueeze(2)\n            beta_H11 = beta[t].unsqueeze(1).unsqueeze(2)\n\n            old_state_HKV = g_H11 * state_HKV\n            old_v_H1V = matmul(k_H1K, old_state_HKV)\n            new_v_H1V = beta_H11 * v_H1V + (1 - beta_H11) * old_v_H1V\n            state_remove = torch.einsum('hkl,hlv->hkv', k_H1K.transpose(-1, -2), old_v_H1V)\n            state_update = torch.einsum('hkl,hlv->hkv', k_H1K.transpose(-1, -2), new_v_H1V)\n            state_HKV = old_state_HKV - state_remove + state_update\n\n            o_H1V = scale * matmul(q_H1K, state_HKV)\n            output[t] = o_H1V.squeeze(1).to(torch.bfloat16)\n\n        new_state[seq_idx] = state_HKV.transpose(-1, -2)  # [H,K,V] -> [H,V,K]\n\n    return {\"output\": output, \"new_state\": new_state}"
+}

definitions/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048.json

@@ -0,0 +1,147 @@
+{
+  "name": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048",
+  "description": "FP8 block scale MoE operation. Routing and two grouped-GEMM included.",
+  "op_type": "moe",
+  "tags": [
+    "status:verified",
+    "model:deepseek-v3",
+    "model:deepseek-r1",
+    "quantization:float8_e4m3fn"
+  ],
+  "axes": {
+    "seq_len": {
+      "type": "var",
+      "description": "Sequence length (number of tokens)"
+    },
+    "num_experts": {
+      "type": "const",
+      "value": 256,
+      "description": "Total number of experts."
+    },
+    "num_local_experts": {
+      "type": "const",
+      "value": 32,
+      "description": "Number of local experts with EP size 8."
+    },
+    "hidden_size": {
+      "type": "const",
+      "value": 7168,
+      "description": "Hidden dimension size."
+    },
+    "intermediate_size": {
+      "type": "const",
+      "value": 2048,
+      "description": "MoE intermediate layer size."
+    },
+    "gemm1_out_size": {
+      "type": "const",
+      "value": 4096,
+      "description": "Output size of the first GEMM (W13). Should be 2 * intermediate_size"
+    },
+    "num_hidden_blocks": {
+      "type": "const",
+      "value": 56,
+      "description": "Number of quantized blocks along the hidden_size dimension (block_size=128)."
+    },
+    "num_intermediate_blocks": {
+      "type": "const",
+      "value": 16,
+      "description": "Number of quantized blocks along the intermediate_size dimension (block_size=128)."
+    },
+    "num_gemm1_out_blocks": {
+      "type": "const",
+      "value": 32,
+      "description": "Number of quantized blocks along the gemm1_out_size dimension (block_size=128)."
+    }
+  },
+  "inputs": {
+    "routing_logits": {
+      "shape": [
+        "seq_len",
+        "num_experts"
+      ],
+      "dtype": "float32",
+      "description": "Tensor of routing logits for expert selection"
+    },
+    "routing_bias": {
+      "shape": [
+        "num_experts"
+      ],
+      "dtype": "bfloat16",
+      "description": "Bias tensor for routing. Pass all zeros for no bias."
+    },
+    "hidden_states": {
+      "shape": [
+        "seq_len",
+        "hidden_size"
+      ],
+      "dtype": "float8_e4m3fn",
+      "description": "Input hidden states tensor (FP8 quantized)"
+    },
+    "hidden_states_scale": {
+      "shape": [
+        "num_hidden_blocks",
+        "seq_len"
+      ],
+      "dtype": "float32",
+      "description": "Block-wise scaling factors for hidden states."
+    },
+    "gemm1_weights": {
+      "shape": [
+        "num_local_experts",
+        "gemm1_out_size",
+        "hidden_size"
+      ],
+      "dtype": "float8_e4m3fn",
+      "description": "First GEMM weights for all local experts (gate and up projections)."
+    },
+    "gemm1_weights_scale": {
+      "shape": [
+        "num_local_experts",
+        "num_gemm1_out_blocks",
+        "num_hidden_blocks"
+      ],
+      "dtype": "float32",
+      "description": "Block-wise scaling factors for first GEMM weights."
+    },
+    "gemm2_weights": {
+      "shape": [
+        "num_local_experts",
+        "hidden_size",
+        "intermediate_size"
+      ],
+      "dtype": "float8_e4m3fn",
+      "description": "Second GEMM weights for all local experts (down projection)."
+    },
+    "gemm2_weights_scale": {
+      "shape": [
+        "num_local_experts",
+        "num_hidden_blocks",
+        "num_intermediate_blocks"
+      ],
+      "dtype": "float32",
+      "description": "Block-wise scaling factors for second GEMM weights."
+    },
+    "local_expert_offset": {
+      "shape": null,
+      "dtype": "int32",
+      "description": "Offset of local experts in global expert space."
+    },
+    "routed_scaling_factor": {
+      "shape": null,
+      "dtype": "float32",
+      "description": "Scaling factor for routing weights."
+    }
+  },
+  "outputs": {
+    "output": {
+      "shape": [
+        "seq_len",
+        "hidden_size"
+      ],
+      "dtype": "bfloat16",
+      "description": "Final MoE output tensor"
+    }
+  },
+  "reference": "import torch\r\n\r\n\r\n@torch.no_grad()\r\ndef run(\r\n    routing_logits: torch.Tensor,\r\n    routing_bias: torch.Tensor,\r\n    hidden_states: torch.Tensor,\r\n    hidden_states_scale: torch.Tensor,\r\n    gemm1_weights: torch.Tensor,\r\n    gemm1_weights_scale: torch.Tensor,\r\n    gemm2_weights: torch.Tensor,\r\n    gemm2_weights_scale: torch.Tensor,\r\n    local_expert_offset: int,\r\n    routed_scaling_factor: float,\r\n):\r\n    \"\"\"\r\n    • FP8 block-scale dequantization: float ≈ fp8 * scale\r\n    • DeepSeek-V3 no-aux routing:\r\n        s = sigmoid(logits)\r\n        s_with_bias = s + bias\r\n        group by n_group=8; per group take top-2 sum → pick topk_group=4 groups\r\n        on the kept groups, take global top_k=8 experts\r\n        combine with weights derived from s (without bias), normalized and\r\n        scaled by routed_scaling_factor\r\n    • Local computation:\r\n        only experts in [local_expert_offset, local_expert_offset + E_local) are\r\n        computed on this rank (GEMM1 → SwiGLU → GEMM2), then per-token weighted\r\n        accumulation.\r\n    \"\"\"\r\n\r\n    # Fixed DeepSeek-V3/R1 geometry\r\n    H = 7168\r\n    I = 2048\r\n    E_local = gemm1_weights.shape[0]\r\n    \r\n    BLOCK = 128\r\n    E_global = routing_logits.shape[1]\r\n    T = routing_logits.shape[0]\r\n\r\n    assert H == 7168, \"hidden_size must be 7168\" \r\n    assert I == 2048, \"intermediate_size must be 2048\"\r\n    assert E_global == 256, \"num_experts must be 256\"\r\n    assert E_local == 32, \"num_local_experts must be 32\"\r\n\r\n    # Routing constants\r\n    TOP_K = 8\r\n    N_GROUP = 8\r\n    TOPK_GROUP = 4\r\n\r\n    # Block counts\r\n    num_hidden_blocks = H // BLOCK          # 56\r\n    num_intermediate_blocks = I // BLOCK    # 16\r\n    num_gemm1_out_blocks = (2 * I) // BLOCK # 32\r\n\r\n    # Shape checks\r\n    assert hidden_states.shape == (T, H)\r\n    assert hidden_states_scale.shape == (num_hidden_blocks, T)\r\n    assert gemm1_weights.shape == (E_local, 2 * I, H)\r\n    assert gemm1_weights_scale.shape == (E_local, num_gemm1_out_blocks, num_hidden_blocks)\r\n    assert gemm2_weights.shape == (E_local, H, I)\r\n    assert gemm2_weights_scale.shape == (E_local, num_hidden_blocks, num_intermediate_blocks)\r\n    assert routing_bias.shape[-1] == E_global\r\n\r\n    device = hidden_states.device\r\n\r\n    # 1) FP8 block-scale dequantization\r\n    # hidden_states: [T, H], scale: [H/128, T] (transposed layout)\r\n    A_fp32 = hidden_states.to(torch.float32)\r\n    A_scale = hidden_states_scale.to(torch.float32)                # [H/128, T]\r\n    A_scale_TH = A_scale.permute(1, 0).contiguous()            # [T, H/128]\r\n    A_scale_expanded = (\r\n        A_scale_TH.unsqueeze(-1)\r\n        .repeat(1, 1, BLOCK)                                   # [T, H/128, 128]\r\n        .reshape(T, H)                                         # [T, H]\r\n        .contiguous()\r\n    )\r\n    A = A_fp32 * A_scale_expanded                              # [T, H] float32\r\n\r\n    # W13: [E_local, 2I, H], scale: [E_local, (2I)/128, H/128]\r\n    W13_fp32 = gemm1_weights.to(torch.float32)\r\n    S13 = gemm1_weights_scale.to(torch.float32)\r\n    S13_expanded = torch.repeat_interleave(S13, BLOCK, dim=1)  # [E, 2I, H/128]\r\n    S13_expanded = torch.repeat_interleave(S13_expanded, BLOCK, dim=2)  # [E, 2I, H]\r\n    W13 = W13_fp32 * S13_expanded                              # [E, 2I, H] float32\r\n\r\n    # W2: [E_local, H, I], scale: [E_local, H/128, I/128]\r\n    W2_fp32 = gemm2_weights.to(torch.float32)\r\n    S2 = gemm2_weights_scale.to(torch.float32)\r\n    S2_expanded = torch.repeat_interleave(S2, BLOCK, dim=1)    # [E, H, I/128]\r\n    S2_expanded = torch.repeat_interleave(S2_expanded, BLOCK, dim=2)    # [E, H, I]\r\n    W2 = W2_fp32 * S2_expanded                                 # [E, H, I] float32\r\n\r\n    # 2) No-aux routing\r\n    logits = routing_logits.to(torch.float32)                      # [T, E_global]\r\n    bias = routing_bias.to(torch.float32).reshape(-1)              # [E_global]\r\n\r\n    # Sigmoid\r\n    s = 1.0 / (1.0 + torch.exp(-logits))                       # [T, E]\r\n    s_with_bias = s + bias                                     # [T, E] (broadcast)\r\n\r\n    # Grouping\r\n    group_size = E_global // N_GROUP # 32\r\n    s_wb_grouped = s_with_bias.view(T, N_GROUP, group_size)    # [T, 8, 32]\r\n\r\n    # Group scores = sum of top-2 values within each group\r\n    top2_vals, _ = torch.topk(s_wb_grouped, k=2, dim=2, largest=True, sorted=False)  # [T, 8, 2]\r\n    group_scores = top2_vals.sum(dim=2)                        # [T, 8]\r\n\r\n    # Select topk_group groups → group mask\r\n    _, group_idx = torch.topk(group_scores, k=TOPK_GROUP, dim=1, largest=True, sorted=False)  # [T, 4]\r\n    group_mask = torch.zeros_like(group_scores)                # [T, 8]\r\n    group_mask.scatter_(1, group_idx, 1.0)\r\n    score_mask = group_mask.unsqueeze(2).expand(T, N_GROUP, group_size).reshape(T, E_global)  # [T, E]\r\n\r\n    # Global top-k (within kept groups), based on s_with_bias\r\n    neg_inf = torch.finfo(torch.float32).min\r\n    scores_pruned = s_with_bias.masked_fill(score_mask == 0, neg_inf)                  # [T, E]\r\n    _, topk_idx = torch.topk(scores_pruned, k=TOP_K, dim=1, largest=True, sorted=False)  # [T, 8]\r\n\r\n    # Combination weights: use s (without bias) for normalization\r\n    M = torch.zeros_like(s)                                    # [T, E]\r\n    M.scatter_(1, topk_idx, 1.0)                               # 0/1 mask\r\n    weights = s * M                                            # [T, E]\r\n    weights_sum = weights.sum(dim=1, keepdim=True) + 1e-20\r\n    weights = (weights / weights_sum) * routed_scaling_factor  # [T, E]\r\n\r\n    # 3) Local expert compute and accumulation\r\n    output = torch.zeros((T, H), dtype=torch.float32, device=device)\r\n\r\n    local_start = int(local_expert_offset)\r\n\r\n    # For each local expert: find selected tokens, run GEMM1→SwiGLU→GEMM2, accumulate by weights\r\n    for le in range(E_local):\r\n        ge = local_start + le\r\n        if ge < 0 or ge >= E_global:\r\n            continue\r\n\r\n        # Tokens that selected this global expert ge in their top-k\r\n        sel_mask_per_token = (topk_idx == ge).any(dim=1)       # [T] bool\r\n        if not sel_mask_per_token.any():\r\n            continue\r\n\r\n        token_idx = torch.nonzero(sel_mask_per_token, as_tuple=False).squeeze(1)  # [Tk]\r\n        Tk = token_idx.numel()\r\n\r\n        # Gather inputs and weights for this expert\r\n        A_e = A.index_select(0, token_idx)                     # [Tk, H]\r\n        W13_e = W13[le]                                        # [2I, H]\r\n        W2_e = W2[le]                                          # [H, I]\r\n\r\n        # GEMM1: [Tk, H] @ [H, 2I] = [Tk, 2I]\r\n        G1 = A_e.matmul(W13_e.t())                             # [Tk, 2I]\r\n\r\n        # SwiGLU: split and apply silu(x) = x / (1 + exp(-x))\r\n        X1 = G1[:, :I]                                         # [Tk, I]\r\n        X2 = G1[:, I:]                                         # [Tk, I]\r\n        silu_X2 = X2 / (1.0 + torch.exp(-X2))                  # [Tk, I]\r\n        C = silu_X2 * X1                                       # [Tk, I]\r\n\r\n        # GEMM2: [Tk, I] @ [I, H] = [Tk, H]\r\n        O = C.matmul(W2_e.t())                                 # [Tk, H]\r\n\r\n        # Accumulate with per-token routing weights for this expert\r\n        w_tok = weights.index_select(0, token_idx)[:, ge]      # [Tk]\r\n        output.index_add_(0, token_idx, O * w_tok.unsqueeze(1))  # [Tk,H] * [Tk,1]\r\n\r\n    return output.to(torch.bfloat16)"
+}

workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048.jsonl

@@ -0,0 +1,19 @@
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "b8f4f012-a32e-4356-b4e1-7665b3d598af", "axes": {"seq_len": 7}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_547d24f37f554e2fab107fb57a41e73e.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 192}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "e05c6c03-5603-4a1c-b34c-dcce0ecaeea4", "axes": {"seq_len": 1}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_25ff432053b5474d86dda63b7daf1734.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 32}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "6230e838-67ca-41dd-a9d6-6f36b7676c6b", "axes": {"seq_len": 32}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_963f2f970f2e4f1688595246c239372e.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 32}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "8f1ff9f1-6747-41d1-a1d8-2868cdacf893", "axes": {"seq_len": 80}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_96384077a7da46bcb65b7c694586960d.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 96}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "1a4c6ba1-3cd2-4d7d-b716-84f2d52b69fc", "axes": {"seq_len": 901}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_90235a13a92d4cb0b552a5d13b119b2e.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 96}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "a7c2bcfd-a2f4-479e-8d32-200115df89cf", "axes": {"seq_len": 16}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_b7a56d54b53740e1b95e08d5b31977f0.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 224}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "2e69caee-ae5c-473b-aa99-5dc6659829d4", "axes": {"seq_len": 15}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_62adbb89946c4512881ab23d3819f740.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 32}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "8cba5890-4288-448a-93b8-42c14c6b9420", "axes": {"seq_len": 14}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_886eb15cbc2f49ad8bd3d4244c38ed91.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 0}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "5e8dc11c-f2a9-42d5-8dce-9419cbf34d5d", "axes": {"seq_len": 14107}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_ad0adeae69b443b192dc6c7394302345.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 32}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "58a34f27-7995-4155-8b46-f60a7225e20e", "axes": {"seq_len": 11948}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_bcb1fd5c45f84a46b6e094fe7a988890.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 128}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "5eadab1e-a0e3-4966-b0fd-1115cd77497c", "axes": {"seq_len": 62}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_13f420b30bbe43548bf54556a0185219.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 96}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "eedc63b2-c03b-4cf2-8d16-1f46566e3af7", "axes": {"seq_len": 59}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5d8ecbc401024d0697fec81a0dab7163.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 160}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "e626d3e6-6c29-4fd8-bb7c-5b09eec61702", "axes": {"seq_len": 58}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_6c6354ff369c41f3b577e01b70a015a7.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 64}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "74d7ff04-0365-4cf8-a824-ce61b7131dea", "axes": {"seq_len": 57}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_a071bbc5a57648f6ac1346515429d92d.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 96}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "4822167c-dae5-4bb1-bb53-e4adb256245b", "axes": {"seq_len": 56}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_c0d3fb257396413c85038b179a27d3a1.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 64}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "81955b1e-086f-49c1-9f40-a18a5aaf509d", "axes": {"seq_len": 55}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_8d2c57225e994c41aca633b282f92c78.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 128}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "76010cb4-f73c-4145-8365-8642a2ce99de", "axes": {"seq_len": 54}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_d517fa44dc8245ef9053f097b6b0cb76.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 128}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "fc378037-e8fa-4305-b00f-4af47933fd53", "axes": {"seq_len": 53}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_e82e1e6107d24eb9ba26b15acd580832.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 32}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}
+{"definition": "moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048", "solution": null, "workload": {"uuid": "f7d6ac7c-24ec-47e4-aefc-4863a5e3e1d9", "axes": {"seq_len": 52}, "inputs": {"routing_logits": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "routing_logits"}, "hidden_states": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "hidden_states"}, "hidden_states_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "hidden_states_scale"}, "routing_bias": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "routing_bias"}, "gemm1_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "gemm1_weights"}, "gemm1_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "gemm1_weights_scale"}, "gemm2_weights": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "gemm2_weights"}, "gemm2_weights_scale": {"type": "safetensors", "path": "./blob/workloads/moe/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048/moe_fp8_block_scale_ds_routing_topk8_ng8_kg4_e32_h7168_i2048_5b074f60cf5d449db9375e6ff0fa89b6.safetensors", "tensor_key": "gemm2_weights_scale"}, "local_expert_offset": {"type": "scalar", "value": 160}, "routed_scaling_factor": {"type": "scalar", "value": 2.5}}}, "evaluation": null}