Skip to content

[Klaud Cold] Update kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg vLLM image to nightly-dev-arm64-cu13-3696c77 (digest-pinned) / 将 kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg 的 vLLM 镜像更新至 nightly-dev-arm64-cu13-3696c77（按 digest 固定） #2956

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

Closed
Klaud-Cold wants to merge 4 commits into from
Closed

[Klaud Cold] Update kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg vLLM image to nightly-dev-arm64-cu13-3696c77 (digest-pinned) / 将 kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg 的 vLLM 镜像更新至 nightly-dev-arm64-cu13-3696c77（按 digest 固定） #2956

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
58c3e91
feat(config): refresh kimik3-fp4-gb200 DSpark DCP16 AgentX vLLM image…
Klaud-Cold Sep 10, 2026
a090cd6
fix(recipe): use the native Dynamo chat processor for kimik3 GB200 DC…
Klaud-Cold Sep 10, 2026
e9e068a
chore(changelog): record kimik3 GB200 DSpark DCP16 AgentX vLLM image …
Klaud-Cold Sep 10, 2026
2566897
chore: merge main into the kimik3 GB200 DCP16 image refresh branch
Klaud-Cold Sep 10, 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
13 changes: 7 additions & 6 deletions ...-slurm-recipes/vllm/kimi-k3/agentic/agg-gb200-dcp16-dspark4-maxseq2-mooncake-agentic.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3,14 +3,14 @@ name: "kimi-k3-vllm-agg-gb200-dcp16-dspark4-maxseq2-mooncake-agentic"

model:
path: "kimi-k3"
container: "vllm/vllm-openai:nightly-dev-arm64-cu13.0.1-75c2eef"
container: "vllm/vllm-openai:nightly-dev-arm64-cu13-3696c77@sha256:42e17a3c600c043c45e0623e4568cd4d5903dd76525a85f5df6ca6fa65358b12"
precision: "fp4"

identity:
model:
repo: "moonshotai/Kimi-K3"
container:
image: "vllm/vllm-openai:nightly-dev-arm64-cu13.0.1-75c2eef"
image: "vllm/vllm-openai:nightly-dev-arm64-cu13-3696c77@sha256:42e17a3c600c043c45e0623e4568cd4d5903dd76525a85f5df6ca6fa65358b12"
frameworks:
dynamo: "ba83080ecd31c1ce918559e576d3c5bc9e092ff1"

Expand Down Expand Up @@ -40,11 +40,12 @@ frontend:
type: dynamo
enable_multiple_frontends: false
args:
dyn-chat-processor: "vllm"
# vLLM moved vllm/entrypoints/openai/cli_args.py (vllm-project/vllm#53659,
# 2026-08-25), which Dynamo's `--dyn-chat-processor vllm` path imports, so
# this image runs the native Dynamo chat processor. Tool-call and reasoning
# parsing stay on kimi_k3 through the workers' dyn-tool-call-parser and
# dyn-reasoning-parser registration below.
trust-remote-code: true
tool-call-parser: "kimi_k3"
reasoning-parser: "kimi_k3"
enable-auto-tool-choice: true
router-mode: "random"
router-session-affinity-ttl-secs: 900
env:
Expand Down
2 changes: 1 addition & 1 deletion configs/nvidia-master.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8537,7 +8537,7 @@ kimik3-fp4-gb200-dynamo-vllm-agentic:
- "CONFIG_FILE=recipes/vllm/kimi-k3/agentic/agg-gb200-dep16-throughput-vllm-simple-offload-agentic.yaml"
# Kimi-K3 GB200 TP16/DCP16 profiles using Mooncake DRAM offload.
kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg:
image: vllm/vllm-openai:nightly-dev-arm64-cu13.0.1-75c2eef
image: vllm/vllm-openai:nightly-dev-arm64-cu13-3696c77@sha256:42e17a3c600c043c45e0623e4568cd4d5903dd76525a85f5df6ca6fa65358b12
model: moonshotai/Kimi-K3
model-prefix: kimik3
runner: cluster:gb200-nv
Expand Down
7 changes: 7 additions & 0 deletions perf-changelog.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7070,3 +7070,10 @@
- "MTP draft length by concurrency: speculative-num-steps 3 (golden AL 2.49) below conc 256, and 1 (golden AL 1.79) at and above it."
- "Trim the search space to TP8 no-offload conc [1, 4, 16], TP8 hicache conc [32, 48], and TP8 DP-attention hicache conc [128, 256]."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2885

- config-keys:
- kimik3-fp4-gb200-dynamo-vllm-agentic-dspark-mooncake-dcp16-agg
description:
- "Bump the vLLM image from vllm/vllm-openai:nightly-dev-arm64-cu13.0.1-75c2eef to vllm/vllm-openai:nightly-dev-arm64-cu13-3696c77 pinned by digest sha256:42e17a3c600c043c45e0623e4568cd4d5903dd76525a85f5df6ca6fa65358b12."
- "Run the Dynamo frontend with its native chat processor because vLLM moved vllm/entrypoints/openai/cli_args.py (vllm-project/vllm#53659), which the pinned Dynamo ba83080 vLLM chat-processor path imports; tool-call and reasoning parsing stay on kimi_k3 through the worker dyn-tool-call-parser and dyn-reasoning-parser registration."
pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2956
Loading