{"model":"deepseek-4.1-flash","title":"MILESTONE — first known SM80/CMP170HX boot","status":"positive","summary":"deepseek-v4.1-flash healthy on 8x CMP 170HX (TP8, eager, fp8_ds_mla, Engram in pinned host RAM; init engine 143.5s, ~5 tok/s) - first known V4.1 boot on SM80/CMP170HX anywhere.","occurred_at":"2026-09-10T18:40:00Z","config":{"serving":"TP8xPP1, DSpark k=5, fp8_ds_mla, 1M ctx","engine":"local/vllm-dsv41:sm80 @ dsv41-feat@e47aa780"},"found":["No SM80 attention class existed -\u003e added TRITON_MLA_SPARSE_DSV41 Ampere class (reuses ROCm Triton sparse-MLA)","fp8e4nv Triton kernels need uint8 shims on sm80 (indexer / fused_indexer_q / cache / engram)","_flashmla_C required by SWA tile-scheduler -\u003e gated on availability"],"failed":["Greedy output incoherent and nondeterministic - forward pass numerically wrong (correctness open)"]}