Skip to content

[AgentX GB300] refresh the full Kimi-K3 curve so every point carries measured power / 重测完整曲线使每个点都带实测功耗 #3315

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
edwingao28 wants to merge 4 commits into
base: main
Choose a base branch
from
Open

[AgentX GB300] refresh the full Kimi-K3 curve so every point carries measured power / 重测完整曲线使每个点都带实测功耗 #3315

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
833de7a
chore: refresh the complete GB300 Kimi-K3 AgentX curve
edwingao28 Sep 20, 2026
ac57ada
fix: migrate the GB300 Kimi-K3 recipes to the current telemetry schema
edwingao28 Sep 20, 2026
ec8f11a
docs: trim GB300 Kimi-K3 changelog
edwingao28 Sep 21, 2026
a7dc0a8
chore: resolve GB300 Kimi-K3 changelog conflict
edwingao28 Sep 21, 2026
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
3 changes: 1 addition & 2 deletions ...rt-slurm-recipes/kimik3/vllm/gb300-fp4/agentx/disagg-1p1d-dcp8-dcp8-dspark4-mooncake.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,8 +174,7 @@ srun_options:

telemetry:
enabled: true
provider: dcgm-power
default_frequency: 1.0
collect_interval_ms: 1000
storage_subdir: power
required: true
startup_timeout_seconds: 120
Expand Down
3 changes: 1 addition & 2 deletions ...rt-slurm-recipes/kimik3/vllm/gb300-fp4/agentx/disagg-1p2d-dcp8-dcp8-dspark4-mooncake.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,8 +174,7 @@ srun_options:

telemetry:
enabled: true
provider: dcgm-power
default_frequency: 1.0
collect_interval_ms: 1000
storage_subdir: power
required: true
startup_timeout_seconds: 120
Expand Down
3 changes: 1 addition & 2 deletions ...rt-slurm-recipes/kimik3/vllm/gb300-fp4/agentx/disagg-1p3d-dcp8-dcp8-dspark4-mooncake.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,8 +174,7 @@ srun_options:

telemetry:
enabled: true
provider: dcgm-power
default_frequency: 1.0
collect_interval_ms: 1000
storage_subdir: power
required: true
startup_timeout_seconds: 120
Expand Down
3 changes: 1 addition & 2 deletions ...rt-slurm-recipes/kimik3/vllm/gb300-fp4/agentx/disagg-1p3d-dcp8-dcp8-dspark7-mooncake.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -174,8 +174,7 @@ srun_options:

telemetry:
enabled: true
provider: dcgm-power
default_frequency: 1.0
collect_interval_ms: 1000
storage_subdir: power
required: true
startup_timeout_seconds: 120
Expand Down
12 changes: 12 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8531,3 +8531,15 @@
- "为 GB300 vLLM DeepSeek-V4.1-Flash AgentX 配方在现有 TP4 臂旁新增 TP2 臂，Engram 表继续通过 --engram-config cpu_offload 放在固定页主机 DRAM；每张 277 GiB GPU 的权重升至约 175 GiB"
- "在 dsv41flash_fp4_vllm_mtp.sh 中将 B200 TP2 的上限推广到所有 TP2 臂：--max-num-batched-tokens 4096（上游 16384 时 indexer 的 [batched-tokens, 1M] fp8 缓冲区达 32 GiB）、--max-num-seqs 为并发的两倍（16-256）、CUDA graph 捕获上限 512，为每张 GPU 留出约 36 GiB KV；TP4 与 TP8 臂沿用上游默认值"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3321

- config-keys:
- kimik3-fp4-gb300-dynamo-vllm-agentic-dspark-mooncake-dcp8-agg
- kimik3-fp4-gb300-dynamo-vllm-agentic-dspark-mooncake-dcp8-disagg
scenario-type:
- agentic-coding
description:
- "Refresh the full GB300 Kimi-K3 AgentX curve with measured power across aggregated and disaggregated deployments."
- "Update four disaggregated recipes to use srtctl's dcgm_exporter and collect_interval_ms telemetry fields."
- "重测 GB300 Kimi-K3 AgentX 完整曲线，为聚合和分离部署补齐实测功耗。"
- "将四个分离部署配方的 telemetry 字段更新为 srtctl 的 dcgm_exporter 和 collect_interval_ms。"
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3315
Loading