Add Kimi K3 B300 1P3D EFA config #2848
There are no files selected for viewing
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,228 @@ | ||
| # Kimi-K3 B300 disaggregated, agentic coding. | ||
| # | ||
| # Mirrors the 1P3D/c32 point from PR 2814, adapted for AWS EFA. NIXL carries | ||
| # P-to-D GPU KV traffic through its LIBFABRIC backend. The ordinary CUDA | ||
| # allocator and DMA-BUF are required because cuMem/VMM-backed KV buffers fail | ||
| # EFA reads with EINVAL on this stack. | ||
| # | ||
| # Mooncake is intentionally instantiated only by the prefill worker. Its large | ||
| # host store uses TCP locally and is not registered with EFA; decode workers use | ||
| # NIXL only and therefore contribute no CPU memory to the Mooncake pool. | ||
|
|
||
| name: kimi-k3-vllm-disagg-b300-1p3d-dcp8-dcp8-dspark4-mooncake-c32-agentic | ||
| model: | ||
| path: moonshotai/Kimi-K3 | ||
| container: vllm/vllm-openai:nightly-dev-x86_64-cu13-3696c77 | ||
| precision: fp4 | ||
| dynamo: | ||
| hash: ba83080ecd31c1ce918559e576d3c5bc9e092ff1 | ||
| install: true | ||
| health_check: | ||
| max_attempts: 720 | ||
| interval_seconds: 10 | ||
| resources: | ||
| gpu_type: b300 | ||
| gpus_per_node: 8 | ||
| prefill_nodes: 1 | ||
| decode_nodes: 3 | ||
| prefill_workers: 1 | ||
| decode_workers: 3 | ||
| gpus_per_prefill: 8 | ||
| gpus_per_decode: 8 | ||
| infra: | ||
| etcd_nats_dedicated_node: false | ||
| nats_max_payload_mb: 32 | ||
| frontend: | ||
| type: dynamo | ||
| enable_multiple_frontends: false | ||
| args: | ||
| router-mode: least-loaded | ||
| router-session-affinity-ttl-secs: 900 | ||
| env: | ||
| DYN_ROUTER_ACTIVE_REQUEST_EXPIRY_SECS: '3600' | ||
| DYN_TOKENIZER_CACHE_BYTES: '8589934592' | ||
| PYTHONPYCACHEPREFIX: /tmp/vllm-pycache-kimi-k3 | ||
| backend: | ||
| type: vllm | ||
| connector: null | ||
| dp_launch_mode: per_node | ||
| kv_events_config: | ||
| prefill: true | ||
| mooncake_kv_store: | ||
| master_extra_args: | ||
| - --default_kv_lease_ttl=60000 | ||
| - --eviction_high_watermark_ratio=0.95 | ||
| - --eviction_ratio=0.10 | ||
| store_config: | ||
| metadata_server: P2PHANDSHAKE | ||
| global_segment_size: 281GB | ||
| local_buffer_size: 4GB | ||
| protocol: tcp | ||
| device_name: '' | ||
| mode: embedded | ||
| enable_offload: false | ||
| prefill_environment: | ||
| VLLM_USE_DIRECT_DCP_A2A: '1' | ||
| VLLM_USE_DIRECT_DCP_Q_GATHER: '1' | ||
| VLLM_USE_DIRECT_DCP_KV_GATHER: '1' | ||
| VLLM_ALLREDUCE_USE_FLASHINFER: '1' | ||
| VLLM_KIMI_K3_SHARD_SP_SHARED_EXPERT: '0' | ||
| DYN_REQUEST_PLANE: tcp | ||
| ETCD_LEASE_TTL: '600' | ||
| VLLM_ENGINE_READY_TIMEOUT_S: '3600' | ||
| VLLM_RPC_TIMEOUT: '600000' | ||
| TILELANG_CLEANUP_TEMP_FILES: '1' | ||
| VLLM_USE_NCCL_SYMM_MEM: '0' | ||
| NCCL_CUMEM_ENABLE: '1' | ||
| NCCL_MNNVL_ENABLE: '0' | ||
| NCCL_NVLS_ENABLE: '1' | ||
| VLLM_SERVER_DEV_MODE: '1' | ||
| VLLM_USE_V2_MODEL_RUNNER: '1' | ||
| VLLM_USE_RUST_FRONTEND: '1' | ||
| VLLM_MOONCAKE_LOAD_RECV_THREADS: '4' | ||
| MC_SLICE_SIZE: '1048576' | ||
| VLLM_CONNECTOR_PREFETCH_DEPTH: '8' | ||
| VLLM_CONNECTOR_PREFETCH_KV_CAP: '0.65' | ||
| VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: '1800' | ||
| VLLM_PREFIX_CACHE_RETENTION_INTERVAL: '0' | ||
| NCCL_P2P_LEVEL: NVL | ||
| MC_ENABLE_DEST_DEVICE_AFFINITY: '1' | ||
| WITH_NVIDIA_PEERMEM: '0' | ||
| NCCL_NET_PLUGIN: none | ||
| UCX_MEMTYPE_CACHE: n | ||
| UCX_MEMTYPE_REG_WHOLE: n | ||
| UCX_RCACHE_MAX_UNRELEASED: '1024' | ||
| UCX_TCP_AF_PRIO: inet | ||
| FI_PROVIDER: efa | ||
| FI_EFA_USE_DEVICE_RDMA: '1' | ||
| FI_EFA_ENABLE_SHM_TRANSFER: '0' | ||
| FI_EFA_ENABLE_SHM: '0' | ||
| FI_EFA_USE_HUGE_PAGE: '0' | ||
| FI_HMEM: cuda | ||
| FI_HMEM_CUDA_ENABLE_XFER: '1' | ||
| FI_HMEM_CUDA_USE_DMABUF: '1' | ||
| VLLM_SSM_CONV_STATE_LAYOUT: DS | ||
| DG_JIT_CACHE_DIR: /tmp/dg-cache-kimi-k3-gb300-pd-dspark-mooncake-{job_id} | ||
| PYTHONHASHSEED: '42' | ||
| PYTHONNOUSERSITE: '1' | ||
| PYTHONUNBUFFERED: '1' | ||
| PYTHONPATH: /nixl_overlay/lib/python3/dist-packages | ||
| LD_LIBRARY_PATH: /nixl_overlay/lib/x86_64-linux-gnu:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/efa_system_libs:/efa_system_libs/libibverbs:/usr/local/cuda/lib64:/usr/local/nvidia/lib64 | ||
| IBV_DRIVERS_PATH: /efa_system_libs/libibverbs | ||
| NIXL_PLUGIN_DIR: /nixl_overlay/lib/x86_64-linux-gnu/plugins | ||
| TORCH_CUDA_ARCH_LIST: '10.0' | ||
| MC_STORE_CLIENT_METRIC: '1' | ||
| MC_STORE_CLIENT_METRIC_INTERVAL: '5' | ||
| MC_TE_METRIC: '0' | ||
| decode_environment: | ||
| VLLM_USE_DIRECT_DCP_A2A: '1' | ||
| VLLM_USE_DIRECT_DCP_Q_GATHER: '1' | ||
| VLLM_USE_DIRECT_DCP_KV_GATHER: '1' | ||
| VLLM_ALLREDUCE_USE_FLASHINFER: '1' | ||
| VLLM_KIMI_K3_SHARD_SP_SHARED_EXPERT: '0' | ||
| DYN_REQUEST_PLANE: tcp | ||
| ETCD_LEASE_TTL: '600' | ||
| VLLM_ENGINE_READY_TIMEOUT_S: '3600' | ||
| VLLM_RPC_TIMEOUT: '600000' | ||
| TILELANG_CLEANUP_TEMP_FILES: '1' | ||
| VLLM_USE_NCCL_SYMM_MEM: '0' | ||
| NCCL_CUMEM_ENABLE: '1' | ||
| NCCL_MNNVL_ENABLE: '0' | ||
| NCCL_NVLS_ENABLE: '1' | ||
| VLLM_SERVER_DEV_MODE: '1' | ||
| VLLM_USE_V2_MODEL_RUNNER: '1' | ||
| VLLM_USE_RUST_FRONTEND: '1' | ||
| VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: '1800' | ||
| VLLM_PREFIX_CACHE_RETENTION_INTERVAL: '0' | ||
| NCCL_P2P_LEVEL: NVL | ||
| WITH_NVIDIA_PEERMEM: '0' | ||
| NCCL_NET_PLUGIN: none | ||
| UCX_MEMTYPE_CACHE: n | ||
| UCX_MEMTYPE_REG_WHOLE: n | ||
| UCX_RCACHE_MAX_UNRELEASED: '1024' | ||
| UCX_TCP_AF_PRIO: inet | ||
| FI_PROVIDER: efa | ||
| FI_EFA_USE_DEVICE_RDMA: '1' | ||
| FI_EFA_ENABLE_SHM_TRANSFER: '0' | ||
| FI_EFA_ENABLE_SHM: '0' | ||
| FI_EFA_USE_HUGE_PAGE: '0' | ||
| FI_HMEM: cuda | ||
| FI_HMEM_CUDA_ENABLE_XFER: '1' | ||
| FI_HMEM_CUDA_USE_DMABUF: '1' | ||
| VLLM_SSM_CONV_STATE_LAYOUT: DS | ||
| DG_JIT_CACHE_DIR: /tmp/dg-cache-kimi-k3-gb300-pd-dspark-mooncake-{job_id} | ||
| PYTHONHASHSEED: '42' | ||
| PYTHONNOUSERSITE: '1' | ||
| PYTHONUNBUFFERED: '1' | ||
| PYTHONPATH: /nixl_overlay/lib/python3/dist-packages | ||
| LD_LIBRARY_PATH: /nixl_overlay/lib/x86_64-linux-gnu:/opt/amazon/efa/lib:/opt/amazon/efa/lib64:/efa_system_libs:/efa_system_libs/libibverbs:/usr/local/cuda/lib64:/usr/local/nvidia/lib64 | ||
| IBV_DRIVERS_PATH: /efa_system_libs/libibverbs | ||
| NIXL_PLUGIN_DIR: /nixl_overlay/lib/x86_64-linux-gnu/plugins | ||
| TORCH_CUDA_ARCH_LIST: '10.0' | ||
| vllm_config: | ||
| prefill: | ||
| kv-transfer-config: '{"kv_connector":"MultiConnector","kv_role":"kv_both","kv_load_failure_policy":"recompute","kv_connector_extra_config":{"connectors":[{"kv_connector":"NixlConnector","kv_role":"kv_both","kv_load_failure_policy":"fail","kv_buffer_device":"cuda","kv_connector_extra_config":{"backends":["LIBFABRIC"],"enforce_handshake_compat":false,"enable_cross_layers_blocks":false}},{"kv_connector":"MooncakeStoreConnector","kv_role":"kv_both","kv_load_failure_policy":"recompute","kv_connector_extra_config":{"load_async":true,"lookup_async":true,"compact_group_io":true,"max_load_batch_keys":2,"enable_cross_layers_blocks":false,"enable_offload":false}}]}}' | ||
| served-model-name: moonshotai/Kimi-K3 | ||
| prefix-match-unit: 128 | ||
| load-format: fastsafetensors | ||
| kv-cache-dtype: fp8 | ||
| gpu-memory-utilization: 0.94 | ||
| tensor-parallel-size: 8 | ||
| decode-context-parallel-size: 8 | ||
| dcp-comm-backend: a2a | ||
| enable-cumem-allocator: false | ||
| trust-remote-code: true | ||
| max-cudagraph-capture-size: 512 | ||
| stream-interval: 10 | ||
| language-model-only: true | ||
| attention-backend: TOKENSPEED_MLA | ||
| attention-config: '{"mla_prefill_backend": "TRTLLM_RAGGED", "use_prefill_query_quantization": true}' | ||
| speculative-config: '{"model":"Inferact/Kimi-K3-DSpark","attention_backend":"TOKENSPEED_MLA","method":"dspark","num_speculative_tokens":4,"rejection_sample_method":"block","draft_sample_method":"probabilistic"}' | ||
| enable-prefix-caching: true | ||
| decode: | ||
| kv-transfer-config: '{"kv_connector":"NixlConnector","kv_role":"kv_both","kv_load_failure_policy":"fail","kv_buffer_device":"cuda","kv_connector_extra_config":{"backends":["LIBFABRIC"],"enforce_handshake_compat":false,"enable_cross_layers_blocks":false}}' | ||
| served-model-name: moonshotai/Kimi-K3 | ||
| prefix-match-unit: 128 | ||
| load-format: fastsafetensors | ||
| kv-cache-dtype: fp8 | ||
| gpu-memory-utilization: 0.94 | ||
| tensor-parallel-size: 8 | ||
| decode-context-parallel-size: 8 | ||
| dcp-comm-backend: a2a | ||
| enable-cumem-allocator: false | ||
| trust-remote-code: true | ||
| max-cudagraph-capture-size: 512 | ||
| stream-interval: 10 | ||
| language-model-only: true | ||
| attention-backend: TOKENSPEED_MLA | ||
| attention-config: '{"mla_prefill_backend": "TRTLLM_RAGGED", "use_prefill_query_quantization": true}' | ||
| speculative-config: '{"model":"Inferact/Kimi-K3-DSpark","attention_backend":"TOKENSPEED_MLA","method":"dspark","num_speculative_tokens":4,"rejection_sample_method":"block","draft_sample_method":"probabilistic"}' | ||
| enable-prefix-caching: true | ||
| sbatch_directives: | ||
| cpus-per-task: '72' | ||
| mem: '0' | ||
| comment: '''{"OccupiedIdleGPUsJobReaper":{"exemptIdleTimeMins":"60","reason":"model_loading","description":"Very large model will take extra loading time."}}''' | ||
| srun_options: | ||
| mem: '0' | ||
| container-remap-root: '' | ||
| benchmark: | ||
| type: custom | ||
| command: bash /infmax-workspace/benchmarks/multi_node/agentic_srt.sh | ||
| env: | ||
| INFMAX_CONTAINER_WORKSPACE: /infmax-workspace | ||
| RESULT_DIR: /logs/agentic | ||
| PORT: '8000' | ||
| IS_MULTINODE: 'true' | ||
| AIPERF_USE_DYNAMO_CONV_AWARE_ROUTING: '0' | ||
| AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true' | ||
| AIPERF_DYNAMO_SESSION_TIMEOUT_SECONDS: '14400' | ||
| AIPERF_DATASET_MMAP_CACHE_DIR: /aiperf_mmap_cache | ||
| HF_HUB_CACHE: /hf_hub_cache | ||
| WEKA_LOADER_OVERRIDE: semianalysis_cc_traces_weka_062126 | ||
| identity: | ||
| model: | ||
| repo: moonshotai/Kimi-K3 | ||
| container: | ||
| image: vllm/vllm-openai:nightly-dev-x86_64-cu13-3696c77 | ||
| frameworks: | ||
| dynamo: ba83080ecd31c1ce918559e576d3c5bc9e092ff1 |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.
Oops, something went wrong.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Overlay setup failure is ignored
High Severity
The new EFA/NIXL overlay setup is invoked without checking its exit status, and this launcher never enables
set -e. A failed first-time build still leavesUSES_EFA_NIXL=1and continues intosrtctl apply, so the job mounts an incomplete overlay and a 4-node B300 allocation starts without theLIBFABRICbackend it needs.Reviewed by Cursor Bugbot for commit ce0e5b7. Configure here.