Add Qwen235 Frontier MoE portability smoke

runs/frontier-fidelity-envelope-v1/smoke_qwen235_v020_frontier_moe.py

@@ -0,0 +1,84 @@
+#!/usr/bin/env python3
+"""One-cell gate for the two Qwen235 vLLM 0.20 MoE runtime backends."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import sys
+from pathlib import Path
+
+import torch
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--frontier-source", type=Path, required=True)
+    parser.add_argument("--output", type=Path, required=True)
+    return parser.parse_args()
+
+
+def routing(tokens: int, experts: int = 128, topk: int = 8):
+    ids = torch.arange(tokens * topk, device="cuda", dtype=torch.int64)
+    ids = (ids % experts).view(tokens, topk)
+    weights = torch.full((tokens, topk), 1.0 / topk, device="cuda")
+    return weights, ids
+
+
+def main() -> None:
+    args = parse_args()
+    sys.path.insert(0, str(args.frontier_source.resolve()))
+    from frontier.profiling.moe.moe_vllm_kernel import profile_fused_moe_kernel
+
+    weights, ids = routing(8)
+    cells = []
+    cells.append(
+        {
+            "name": "tp4_ep1_triton",
+            "stats": profile_fused_moe_kernel(
+                num_tokens=8,
+                num_experts=128,
+                hidden_dim=4096,
+                expert_hidden_dim=1536,
+                top_k=8,
+                topk_weights=weights,
+                topk_ids=ids,
+                tensor_parallel_size=4,
+                use_fp8=True,
+                block_shape=[128, 128],
+                warmup_steps=1,
+                active_steps=2,
+            ),
+        }
+    )
+    expert_map = torch.full((128,), -1, device="cuda", dtype=torch.int32)
+    expert_map[:16] = torch.arange(16, device="cuda", dtype=torch.int32)
+    cells.append(
+        {
+            "name": "tp1_ep8_flashinfer_cutlass",
+            "stats": profile_fused_moe_kernel(
+                num_tokens=8,
+                num_experts=16,
+                hidden_dim=4096,
+                expert_hidden_dim=1536,
+                top_k=8,
+                topk_weights=weights,
+                topk_ids=ids,
+                tensor_parallel_size=1,
+                use_fp8=True,
+                block_shape=[128, 128],
+                warmup_steps=1,
+                active_steps=2,
+                global_num_experts=128,
+                expert_map=expert_map,
+            ),
+        }
+    )
+    payload = {"schema": "qwen235-v020-frontier-moe-smoke-v1", "cells": cells}
+    args.output.parent.mkdir(parents=True, exist_ok=True)
+    args.output.write_text(json.dumps(payload, indent=2, sort_keys=True) + "\n")
+    print(json.dumps(payload, sort_keys=True))
+
+
+if __name__ == "__main__":
+    main()