feat: add gemm_n16_k2048 workloads and baseline solution

definitions/gemm/gemm_n16_k2048.json

@@ -0,0 +1,49 @@
+{
+    "op_type": "gemm",
+    "tags": [
+        "model:qwen3.5-35b-a3b",
+        "status:reference",
+        "tp:2"
+    ],
+    "inputs": {
+        "A": {
+            "shape": [
+                "M",
+                "K"
+            ],
+            "dtype": "bfloat16"
+        },
+        "B": {
+            "shape": [
+                "N",
+                "K"
+            ],
+            "dtype": "bfloat16"
+        }
+    },
+    "outputs": {
+        "C": {
+            "shape": [
+                "M",
+                "N"
+            ],
+            "dtype": "bfloat16"
+        }
+    },
+    "reference": "import torch\n\ndef run(A, B):\n    C = torch.matmul(A, B.T)\n    return C",
+    "name": "gemm_n16_k2048",
+    "description": "General matrix multiply (GEMM) C = A @ B.T. Captured from Qwen3.5-35B-A3B at TP=2. N=16, K=2048.",
+    "axes": {
+        "M": {
+            "type": "var"
+        },
+        "N": {
+            "type": "const",
+            "value": 16
+        },
+        "K": {
+            "type": "const",
+            "value": 2048
+        }
+    }
+}

solutions/baseline/gemm/gemm_n16_k2048/torch_matmul_8b8ea6.json

@@ -0,0 +1,23 @@
+{
+  "name": "torch_matmul_8b8ea6",
+  "definition": "gemm_n16_k2048",
+  "author": "PyTorch",
+  "spec": {
+    "language": "python",
+    "target_hardware": [
+      "NVIDIA_H100",
+      "NVIDIA_A100",
+      "CPU"
+    ],
+    "entry_point": "main.py::run",
+    "dependencies": [],
+    "destination_passing_style": false
+  },
+  "sources": [
+    {
+      "path": "main.py",
+      "content": "import torch\n\ndef run(A, B):\n    C = torch.matmul(A, B.T)\n    return C"
+    }
+  ],
+  "description": "Baseline GEMM implemented with torch.matmul."
+}

tests/references/test_gemm_n16_k2048.py

@@ -0,0 +1,77 @@
+import torch
+
+
+def run(A, B):
+    M, K = A.shape
+    N, K2 = B.shape
+    assert K == K2
+    assert N == 16
+    assert K == 2048
+    C = torch.matmul(A, B.T)
+    return C
+
+
+def generate_random_inputs(M, N=16, K=2048, device="cuda"):
+    A = torch.randn(M, K, dtype=torch.bfloat16, device=device)
+    B = torch.randn(N, K, dtype=torch.bfloat16, device=device)
+    return {"A": A, "B": B}
+
+
+def test_correctness(M=32, atol=1e-2, rtol=1e-2):
+    print(f"\n{'='*60}")
+    print(f"Testing GEMM N=16, K=2048, M={M}")
+    print(f"{'='*60}")
+
+    device = "cuda" if torch.cuda.is_available() else "cpu"
+    if device == "cpu":
+        print("WARNING: CUDA not available, skipping test")
+        return True
+
+    inputs = generate_random_inputs(M, device=device)
+
+    # Reference
+    ref_C = run(inputs["A"], inputs["B"])
+
+    # Direct torch comparison (fp32)
+    A_f32 = inputs["A"].float()
+    B_f32 = inputs["B"].float()
+    expected = torch.matmul(A_f32, B_f32.T).to(torch.bfloat16)
+
+    abs_diff = torch.abs(ref_C.float() - expected.float())
+    max_abs_diff = abs_diff.max().item()
+    mean_abs_diff = abs_diff.mean().item()
+
+    print(f"Max absolute difference: {max_abs_diff:.6e}")
+    print(f"Mean absolute difference: {mean_abs_diff:.6e}")
+
+    close = torch.allclose(ref_C.float(), expected.float(), atol=atol, rtol=rtol)
+    if close:
+        print(f"\n✓ PASSED (atol={atol}, rtol={rtol})")
+    else:
+        print(f"\n✗ FAILED (atol={atol}, rtol={rtol})")
+    return close
+
+
+def main():
+    print("Testing GEMM N=16, K=2048 Reference Implementation")
+
+    test_configs = [1, 4, 16, 64, 256]
+    passed = 0
+    total = len(test_configs)
+
+    for M in test_configs:
+        try:
+            if test_correctness(M):
+                passed += 1
+        except Exception as e:
+            print(f"✗ Test failed with exception: {e}")
+            import traceback
+            traceback.print_exc()
+
+    print(f"\n{'='*60}")
+    print(f"Summary: {passed}/{total} tests passed")
+    print(f"{'='*60}")
+
+
+if __name__ == "__main__":
+    main()

workloads/gemm/gemm_n16_k2048.jsonl

@@ -0,0 +1,15 @@
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":8192},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"894d7a9c-ee2f-4cc0-95d5-006e9775d985"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":1111},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"e3e2855c-fde5-4aaf-8396-7bb15721cdab"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":100},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"d3585ec1-b594-4001-96a3-6ee2ba4c76a5"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":99},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"7d575e8d-5007-4976-a606-df07c12982c7"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":98},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"b70191f9-91a0-4805-a640-5cdb070f114a"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":97},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"9f6bf52f-6c02-4cdc-bd87-628cad5a5b27"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":96},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"2dd26cae-0e05-48f9-82a9-b497015b499b"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":51},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"0685c3f2-6451-447f-b120-4518dbcd6ccc"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":7962},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"9c8a7c74-32e2-4a46-a995-bb51dbd44f0f"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":113},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"577b98c5-8b78-4287-8029-53f4c9686893"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":6016},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"ad81ba73-66fd-4528-ad88-a6472be4bb74"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":95},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"e67265a4-7410-4096-b044-5ee9a1e6ef4b"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":219},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"fefb0904-b550-4cdb-b9a9-c3d26725c0c1"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":7794},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"6aa45276-3441-477d-a45c-840dd7e52ef9"},"solution":null,"evaluation":null}
+{"definition":"gemm_n16_k2048","workload":{"axes":{"M":5574},"inputs":{"A":{"type":"random"},"B":{"type":"random"}},"uuid":"7361cca7-4c72-45e6-bb46-702bf4695778"},"solution":null,"evaluation":null}