Repository navigation

[AgentX] DeepSeek-V4-Pro-0813 llm-d on GB200 - #2719

Closed

ilmarkov wants to merge 86 commits into

mainfrom

imarkov/dsv4-pro-agentic

Closed

[AgentX] DeepSeek-V4-Pro-0813 llm-d on GB200#2719
ilmarkov wants to merge 86 commits into
mainfrom
imarkov/dsv4-pro-agentic

Commits on Aug 24, 2026
- Add DeepSeek-V4-Pro FP4 B200 llmd-vllm P/D disagg (agentX 1P-DEP8/1D-DEP8)
  
  ilmarkov
  and
  cursoragent
  committed
  8cf10ab
Commits on Aug 26, 2026
- Trim disagg conc-list to 192 and script the v0.10.0 EPP binary pull
  
  ilmarkov
  and
  cursoragent
  committed
  83c3060
- Add DeepSeek-V4-Pro FP4 B200 llmd-vllm aggregated (TP8/DEP8) benchmark
  
  ilmarkov
  and
  cursoragent
  committed
  fe7e680
Commits on Aug 31, 2026
- Update configs
  ilmarkov
  committed
  79f3e12
- Merge upstream/main into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  839c220
- Cleanup comments
  ilmarkov
  committed
  8136823
- Address review comments
  ilmarkov
  committed
  af3818d
- Update DSV4-Pro model to 0813 version
  ilmarkov
  committed
  e648883
Commits on Sep 2, 2026
- fix: run llm-d DSpark with AgentX golden acceptance and explicit metadata
  cquil11
  committed
  a4edd45
- merge: resolve main conflicts while preserving DSpark AgentX fixes
  cquil11
  committed
  a437eb1
- fix: accept cluster runner labels in llm-d container selection
  cquil11
  committed
  e3d6d0c
- chore: remove standalone llm-d recipe and redundant config tests
  cquil11
  committed
  8d00f59
- docs: consolidate DSpark performance changelog
  cquil11
  committed
  0391477
- fix: expose and verify llm-d worker metrics for AIPerf
  cquil11
  committed
  a65c1e3
- fix(llm-d): use the B200 Nscale shared home for cached artifacts
  cquil11
  committed
  f6cfad9
- fix(llm-d): match AgentX router metadata to the bundled version
  cquil11
  committed
  1758eac
- fix(llm-d): move AgentX to GB200 and simplify launch paths
  cquil11
  committed
  4e5f6c6
- Adapt optimized configs and map to agentic gb200
  ilmarkov
  committed
  775e5e9
- merge: resolve conflicts with upstream/imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  claude
  committed
  07c6cfd
- Fix recipe
  ilmarkov
  committed
  95586e0
- Fix spec decoding
  ilmarkov
  committed
  6365015
- Address cursor's comment
  ilmarkov
  committed
  11b8af5
Commits on Sep 3, 2026
- fix: keep GB200 import setup explicit and correct DSpark changelog
  cquil11
  committed
  81060c2
- merge: sync main while preserving the GB200 AgentX changelog entry
  cquil11
  committed
  e8e1594
- fix: preserve fixed DSpark draft budgets for golden AL
  cquil11
  committed
  8493a29
- fix: use GB200 InfiniBand adapters for Mooncake transfers
  cquil11
  committed
  6394609
- fix: provide DEP8 router token-load dependencies
  cquil11
  committed
  73476be
- fix: start a shared Mooncake master for llm-d jobs
  cquil11
  committed
  3ba226b
- fix: reserve GPU memory for long-context llm-d replay
  cquil11
  committed
  bc55cca
- fix: cover TP8 and DEP8 long-context allocation peaks
  cquil11
  committed
  9c1c0c4
- fix: prevent duplicate llm-d frontend metric scrapes
  cquil11
  committed
  7b65775
- fix: retain P/D KV blocks across long decode stalls
  cquil11
  committed
  4a6231e
- fix: validate llm-d metrics targets and retain endpoint roles
  cquil11
  committed
  9411ed9
- fix: copy configs for p1d1
  ilmarkov
  committed
  1a31973
- Add concurrency
  ilmarkov
  committed
  70c8a39
Commits on Sep 4, 2026
- Adjust decode utilization
  ilmarkov
  committed
  02ed639
- fix
  ilmarkov
  committed
  8080915
- fix prefill
  ilmarkov
  committed
  da74471
- Disable AV fin sake of golden acceptence
  ilmarkov
  committed
  109f597
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  claude
  committed
  51255cb
Commits on Sep 14, 2026
- Fix disagg prefill OOM and add SimpleCPUOffloadConnector to mooncake disagg
  
  ilmarkov
  and
  claude
  committed
  4366bc2
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  claude
  committed
  3e1b813
- Move dsv4-gb200 changelog entry to end of file
  
  ilmarkov
  and
  claude
  committed
  5be3fcf
- Fix mock curl in test_llmd_agentic_adapter to handle --write-out preflight
  
  ilmarkov
  and
  claude
  committed
  b089e16
- Fix test_llmd_agentic_adapter endpoint fixture broken by 9411ed99b
  
  ilmarkov
  and
  claude
  committed
  884b32b
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  claude
  committed
  36c6f4d
Commits on Sep 15, 2026
- Fix mooncake KV pollution and EPP routing for agentic p1d1 and DEP8
  
  ilmarkov
  and
  claude
  committed
  23dae65
- Merge remote-tracking branch 'upstream/main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  893ba8c
- Correct after merge
  ilmarkov
  committed
  d6a6d82
Commits on Sep 16, 2026
- fix(llm-d): use external LB and per-rank EPP endpoints for DEP8 AgentX
  
  ilmarkov
  and
  cursoragent
  committed
  ad34e7f
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  fd8e4d5
- Merge upstream/main into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  a5eac8c
Commits on Sep 17, 2026
- fix(llm-d): restore GB200 disagg wrapper and re-deprecate 8k1k key
  
  ilmarkov
  and
  cursoragent
  committed
  f88fe3c
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  73cb898
- fix(llm-d): forward INFMAX_CONTAINER_WORKSPACE for agentic throughput
  ilmarkov
  committed
  3d1ea2b
- debug(llm-d): enable VLLM_LOGGING_LEVEL=DEBUG for DSpark AgentX bring-up
  
  ilmarkov
  and
  cursoragent
  committed
  545e5e6
- fix(llm-d): shrink Mooncake DRAM budget and restore INFO logging
  ilmarkov
  committed
  061901e
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  7456ad7
Commits on Sep 18, 2026
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  authored and
  cquil11
  committed
  4742446
Commits on Sep 21, 2026
- Add prefiller=2 and 3 configs
  ilmarkov
  committed
  ca22a3c
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  5bc5142
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  2a4cd24
- Update the nvhsmem and nccl flags
  ilmarkov
  committed
  5daf09e
- Merge upstream/main into imarkov/dsv4-pro-agentic
  
  ilmarkov
  and
  cursoragent
  committed
  99f14d6
Commits on Sep 23, 2026
- Add adaptive verification, and new configs
  ilmarkov
  committed
  574db81
- Merge branch 'main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  e42b819
Commits on Sep 24, 2026
- Merge branch 'main' into imarkov/dsv4-pro-agentic
  ilmarkov
  authored
  877c094
Commits on Sep 25, 2026
- fix(llm-d): extend prefill health wait and enable agg DEP8 adaptive verification
  ilmarkov
  committed
  57c6fdf
- Merge branch 'main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  f7289ed
- Fix CI tests
  ilmarkov
  committed
  5b1229e
Commits on Sep 28, 2026
- Merge branch 'main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  7892b37
- Fix
  ilmarkov
  committed
  b60604b
- Fix server.sh
  ilmarkov
  committed
  ee4dd0a
- Fix after rename
  ilmarkov
  committed
  916f0b3
Commits on Sep 29, 2026
- fix(gb200-agentx): align agg DEP8 DSpark K=3, add 2P1D c576
  ilmarkov
  committed
  d5a4d3a
- Merge branch 'main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  a9134f0
- Upd disagg
  ilmarkov
  committed
  1320dde
- Merge remote-tracking branch 'upstream/main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  3093354
Commits on Sep 30, 2026
- Merge remote-tracking branch 'upstream/main' into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  67dc45b
- Fix test
  ilmarkov
  committed
  13953f8
- Fix lint
  ilmarkov
  committed
  ba1978d
Commits on Oct 2, 2026
- feat(agentx): add DeepSeek-V4-Pro-0813 llm-d on GB200
  
  ilmarkov
  and
  cquil11
  committed
  139313e
Commits on Oct 6, 2026
- Merge branch 'imarkov/dsv4-pro-agentic' of https://github.com/SemiAnalysisAI/InferenceX into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  0a8b277
Commits on Oct 7, 2026
- Merge upstream/main into imarkov/dsv4-pro-agentic
  ilmarkov
  committed
  aac33a7
- Fix tests
  ilmarkov
  committed
  aed750f
- Fix server.sh
  ilmarkov
  committed
  5f8a118

Commits

Commits on Aug 24, 2026

Commits on Aug 26, 2026

Commits on Aug 31, 2026

Commits on Sep 2, 2026

Commits on Sep 3, 2026

Commits on Sep 4, 2026

Commits on Sep 14, 2026

Commits on Sep 15, 2026

Commits on Sep 16, 2026

Commits on Sep 17, 2026

Commits on Sep 18, 2026

Commits on Sep 21, 2026

Commits on Sep 23, 2026

Commits on Sep 24, 2026

Commits on Sep 25, 2026

Commits on Sep 28, 2026

Commits on Sep 29, 2026

Commits on Sep 30, 2026

Commits on Oct 2, 2026

Commits on Oct 6, 2026

Commits on Oct 7, 2026