Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 7 additions & 6 deletions configs/nvidia-master.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -7176,7 +7176,7 @@ qwen3.5-fp4-gb200-dynamo-sglang-agentic-mtp:
decode: { num-worker: 0, tp: 2, ep: 2, dp-attn: false }

minimaxm3-fp4-b300-vllm-agentic-mtp:
image: vllm/vllm-openai:nightly-5e35a6f4f9bbc217c599692157ca985c894373f7
image: vllm/vllm-openai:nightly-ac7509e2b1db40fec2f03dde1ed4e9dfdc2338c9
model: nvidia/MiniMax-M3-NVFP4
model-prefix: minimaxm3
runner: cluster:b300-nv
Expand All @@ -7185,13 +7185,14 @@ minimaxm3-fp4-b300-vllm-agentic-mtp:
multinode: false
scenarios:
agentic-coding:
# 0.683 maps the B300 TP4 SimpleCPU point to a 1,024 GB
# engine-level CPU KV budget. GPU-resident points receive a zero budget.
- dram-utilization: 0.683
search-space:
- { tp: 4, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 5, 10, 15, 20] }
- { tp: 2, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 2, 5] }
- { tp: 4, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [30, 40, 50, 60, 65, 70, 75] }
- { tp: 8, spec-decoding: mtp, kv-offloading: none, conc-list: [1] }
- { tp: 4, spec-decoding: mtp, kv-offloading: none, conc-list: [1, 5, 10, 15, 20] }
- { tp: 4, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [30] }
- dram-utilization: 1.0
search-space:
- { tp: 2, spec-decoding: mtp, kv-offloading: dram, kv-offload-backend: { name: vllm-simple }, conc-list: [24] }
Comment thread
Ankur-singh marked this conversation as resolved.

# GB200 DeepSeek-V4 disaggregated AgentX frontier. The 3P/2D TEP8/TP8 curve
# covers the middle/high-interactivity range omitted by the one-decode DEP
Expand Down
8 changes: 8 additions & 0 deletions perf-changelog.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -5972,6 +5972,14 @@
- "Use KV-aware Dynamo routing with 4-hour correlation-ID affinity, authoritative vLLM KV events, KV-cache token metrics, and every logical vLLM server-metrics endpoint."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2567

- config-keys:
- minimaxm3-fp4-b300-vllm-agentic-mtp
scenario-type:
- agentic-coding
description:
- "Refresh the MiniMax-M3 B300 AgentX submission as eight EAGLE3-GQA points: no-offload TP8 C1 and TP4 C1/C5/C10/C15/C20, TP4 C30 with 1,024 GB SimpleCPU KV offload, and TP2 C24 with the proportional-policy 750 GB SimpleCPU KV-offload cap."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2610

- config-keys:
- dsv4-fp4-gb300-dynamo-vllm-agentic-mtp-agg
- dsv4-fp4-gb300-dynamo-vllm-agentic-mtp-disagg
Expand Down