Skip to content

Commit 43d99c5

Browse files
seungrokjclaude
andcommitted
[AMD][AgentX] dsv4-fp4-mi355x-sglang-agentic-mtp: update conc-list to [48,64] and add hicache dram arm
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
1 parent 10a4f03 commit 43d99c5

1 file changed

Lines changed: 2 additions & 1 deletion

File tree

configs/amd-master.yaml

Lines changed: 2 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1695,7 +1695,8 @@ dsv4-fp4-mi355x-sglang-agentic-mtp:
16951695
agentic-coding:
16961696
- dram-utilization: 0.80
16971697
search-space:
1698-
- { tp: 8, dp-attn: true, kv-offloading: none, conc-list: [4, 8, 16, 32], spec-decoding: mtp }
1698+
- { tp: 8, dp-attn: true, kv-offloading: none, conc-list: [48, 64], spec-decoding: mtp }
1699+
- { tp: 8, dp-attn: true, kv-offloading: dram, kv-offload-backend: { name: hicache }, conc-list: [48, 64], spec-decoding: mtp }
16991700

17001701
# MiniMax-M3 MXFP8 MI355X recipe:
17011702
# https://github.com/vllm-project/recipes/commit/2a3728ed9892debfd767a72a58ebc90b33f186e5

0 commit comments

Comments
 (0)