Gahow Wang

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 14:18:37 +00:00

0e9b64893f Add full Frontier linear profile sweep

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 14:11:10 +00:00

48d2c2fb80 Add Frontier linear profiler smoke

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 14:00:02 +00:00

32ab584789 Initialize collective profiler with model config

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:57:51 +00:00

3121e35b0e Profile collective dispatch across tensor sizes

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:52:48 +00:00

4bf9bdf28f Add full Qwen30 MoE profiling matrix

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:48:58 +00:00

1fb777fc11 Set vLLM config for collective profiling

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:41:45 +00:00

4fed4329cf Profile MoE TP-local shards without collectives

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:40:14 +00:00

9bbfb87a85 Keep KV smoke artifacts separate

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:37:05 +00:00

9bc38d8851 Profile FA3 KV-cache updates separately

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:34:28 +00:00

a9aed87518 Add FlashInfer TRTLLM all-reduce profile smoke

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:27:42 +00:00

5a958077a7 Add exact vLLM 0.20 MoE profile smoke

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:24:17 +00:00

adaf68badc Add full per-TP FlashAttention profile sweep

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:19:28 +00:00

a70a1fac71 Retry profile smoke with cu129 vLLM wheel

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:17:33 +00:00

b3e437acb0 Serialize vLLM attention profile results

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:09:55 +00:00

2f04971c90 Freeze Qwen30 trace profile support

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:07:32 +00:00

fb12e3b502 Use local cache for vLLM profile smoke

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-16 13:00:43 +00:00

e9a4be6153 Add vLLM 0.20 per-TP FlashAttention profile smoke

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-15 22:21:50 +00:00

13f0713bf2 Document Frontier fixed-cohort fidelity results

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-15 21:21:00 +00:00

344af3a428 Prepare targeted disputed-boundary trial

def9e24c4d Allow targeted SLO boundary repeats

3db632552b Compare Frontier fidelity across real trials

Compare 3 commits »

gahow pushed to codex/fidelity-prefix-pilot-20260714 at gahow/aituner

2026-07-15 19:09:24 +00:00

501ceb8171 Skip null SLOs in boundary repeats