[AMD] cookbook: serve Qwen3.5 MXFP4 on MI355X with an fp8_e4m3 KV cache (#35445)

This commit is contained in:
jacky.cheng
2026-08-19 18:49:32 +08:00
committed by GitHub
parent 3e5ce26c2d
commit 574274660f
@@ -480,6 +480,7 @@ export const Qwen35Deployment = () => {
// (this recipe uses quick all-reduce instead of AITER allreduce fusion).
// Add the FP4-specific flags here.
cmd += ' \\\n --disable-radix-cache';
cmd += ' \\\n --kv-cache-dtype fp8_e4m3';
// Cap concurrency under MTP to avoid OOM at tp=2.
if (speculative === 'enabled') {
cmd += ' \\\n --max-running-requests 128';