[WIP] Add GLM-5.2-NVFP4 B300 AgentX recipes #2654
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 +1,162 @@ | ||
| base: | ||
| name: b300-fp4-glm5.2-agentx-agg | ||
| model: | ||
| path: glm-5.2-fp4 | ||
| container: lmsysorg/sglang:nightly-dev-cu13-20260824-95f5ecd3 | ||
| precision: fp4 | ||
| identity: | ||
| model: | ||
| repo: nvidia/GLM-5.2-NVFP4 | ||
| revision: aec724e | ||
| frameworks: | ||
| sglang: nightly-dev-cu13-20260824-95f5ecd3 | ||
| resources: | ||
| gpu_type: b300 | ||
| gpus_per_node: 8 | ||
| dynamo: | ||
| hash: 71eb001e17fa73c742f0afe1a6ed96836cb135fd | ||
| install: true | ||
| backend: | ||
| type: sglang | ||
| sglang_config: | ||
| aggregated: | ||
| served-model-name: GLM-5.2-FP4 | ||
| trust-remote-code: true | ||
| quantization: modelopt_fp4 | ||
| kv-cache-dtype: fp8_e4m3 | ||
| tensor-parallel-size: 8 | ||
| data-parallel-size: 1 | ||
| expert-parallel-size: 1 | ||
| enable-dp-attention: false | ||
| enable-dp-lm-head: false | ||
| max-running-requests: 10 | ||
| cuda-graph-max-bs: 10 | ||
| chunked-prefill-size: 8192 | ||
| max-prefill-tokens: 8192 | ||
| context-length: 1048576 | ||
| speculative-algorithm: EAGLE | ||
| speculative-num-steps: 4 | ||
| speculative-eagle-topk: 1 | ||
| speculative-num-draft-tokens: 5 | ||
| nsa-prefill-backend: trtllm | ||
| nsa-decode-backend: trtllm | ||
| moe-runner-backend: flashinfer_trtllm | ||
| fp4-gemm-backend: flashinfer_trtllm | ||
| disable-shared-experts-fusion: true | ||
| enable-flashinfer-allreduce-fusion: true | ||
| weight-loader-prefetch-checkpoints: true | ||
| model-loader-extra-config: '{"enable_multithread_load": true}' | ||
| mem-fraction-static: 0.8 | ||
| enable-hierarchical-cache: true | ||
| hicache-write-policy: write_back | ||
| hicache-size: 270 | ||
| hicache-io-backend: direct | ||
| enable-metrics: true | ||
| enable-cache-report: true | ||
| aggregated_environment: | ||
| TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800' | ||
| PYTHONUNBUFFERED: '1' | ||
| DYN_SKIP_SGLANG_LOG_FORMATTING: '1' | ||
| MC_TE_METRIC: 'true' | ||
| NCCL_CUMEM_ENABLE: '1' | ||
| SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0' | ||
| SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1' | ||
| SGLANG_ENABLE_THINKING: '1' | ||
| SGLANG_REASONING_EFFORT: max | ||
| SGLANG_ENABLE_UNIFIED_RADIX_TREE: '1' | ||
| SGLANG_HICACHE_DEBUG_LOG: '1' | ||
| SGLANG_HICACHE_DEBUG_SAMPLE_RATE: '16384' | ||
| SGLANG_MOE_NVFP4_DISPATCH: '1' | ||
| SGLANG_NVFP4_CKPT_FP8_NEXTN_MOE: '1' | ||
| SGLANG_CLIP_MAX_NEW_TOKENS_ESTIMATION: '8' | ||
| SGLANG_SIMULATE_ACC_LEN: '3.33' | ||
| SGLANG_SIMULATE_ACC_METHOD: match-expected | ||
| SGLANG_SIMULATE_ACC_TOKEN_MODE: real-draft-token | ||
| PIP_BREAK_SYSTEM_PACKAGES: '1' | ||
| benchmark: | ||
| type: custom | ||
| env: | ||
| CONCURRENCY: '2' | ||
| DURATION: '3600' | ||
| AIPERF_HF_HOME: /hf_home | ||
| HF_HOME: /hf_home | ||
| HF_HUB_CACHE: /hf_home/hub | ||
| HF_DATASETS_CACHE: /hf_home/datasets | ||
| HF_HUB_OFFLINE: '1' | ||
| HF_DATASETS_OFFLINE: '1' | ||
| HF_HUB_DISABLE_XET: '1' | ||
| XDG_CACHE_HOME: /tmp/cache | ||
| AIPERF_DATASET_MMAP_CACHE_DIR: /hf_home/mmap_cache | ||
| AIPERF_DATASET_CONFIGURATION_TIMEOUT: '1800' | ||
| AIPERF_SERVICE_PROFILE_CONFIGURE_TIMEOUT: '1800' | ||
| AIPERF_FAILED_REQUEST_THRESHOLD: '0.1' | ||
| AIPERF_RECORD_STRIP_PAYLOAD_BYTES: 'true' | ||
| AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true' | ||
| AIPERF_REQUIRED_SERVER_METRIC_PREFIX: 'sglang:' | ||
| command: | | ||
| set -euo pipefail | ||
| export HF_HOME="${AIPERF_HF_HOME:-/hf_home}" | ||
| export HF_HUB_CACHE="${HF_HOME}/hub" | ||
| export HF_DATASETS_CACHE="${HF_HOME}/datasets" | ||
| mkdir -p "${XDG_CACHE_HOME}" "${AIPERF_DATASET_MMAP_CACHE_DIR}" /logs/aiperf | ||
| /aiperf-venv/bin/aiperf profile \ | ||
| --scenario inferencex-agentx-mvp \ | ||
| --url http://localhost:8000 \ | ||
| --endpoint /v1/chat/completions \ | ||
| --endpoint-type chat \ | ||
| --streaming \ | ||
| --model GLM-5.2-FP4 \ | ||
| --tokenizer /model \ | ||
| --tokenizer-trust-remote-code \ | ||
| --concurrency "${CONCURRENCY}" \ | ||
| --benchmark-duration "${DURATION}" \ | ||
| --random-seed 42 \ | ||
| --failed-request-threshold "${AIPERF_FAILED_REQUEST_THRESHOLD}" \ | ||
| --stats-interval 30 \ | ||
| --warmup-requests-per-lane 10 \ | ||
| --warmup-grace-period 1800 \ | ||
| --trajectory-start-min-ratio 0.25 \ | ||
| --trajectory-start-max-ratio 0.75 \ | ||
| --use-server-token-count \ | ||
| --no-gpu-telemetry \ | ||
| --num-dataset-entries 393 \ | ||
| --slice-duration 1.0 \ | ||
| --public-dataset semianalysis_cc_traces_weka_062126 \ | ||
| --trace-idle-gap-cap-seconds 300 \ | ||
| --unsafe-override \ | ||
| --ui simple \ | ||
| --server-metrics-formats json \ | ||
| --output-artifact-dir /logs/aiperf | ||
| rm -f /logs/aiperf/inputs.json | ||
| health_check: | ||
| max_attempts: 1440 | ||
| interval_seconds: 10 | ||
| infra: | ||
| etcd_nats_dedicated_node: false | ||
| nats_max_payload_mb: 64 | ||
| frontend: | ||
| type: dynamo | ||
| env: | ||
| PIP_BREAK_SYSTEM_PACKAGES: '1' | ||
| args: | ||
| router-mode: kv | ||
| active-decode-blocks-threshold: None | ||
| active-prefill-tokens-threshold: None | ||
| active-prefill-tokens-threshold-frac: None | ||
| router-session-affinity-ttl-secs: 3600 | ||
| sbatch_directives: | ||
| mem: '0' | ||
| srun_options: | ||
| mem: '0' | ||
| container-remap-root: '' | ||
| # ################# agentx ################# | ||
| zip_override_mtp_agentx_lowlat: | ||
| name: [agentx-agg1_tp8-c2, agentx-agg1_tp8-c4] | ||
| benchmark: | ||
| env: | ||
| CONCURRENCY: ['2', '4'] | ||
| resources: | ||
| agg_nodes: 1 | ||
| agg_workers: 1 | ||
| gpus_per_decode: 8 | ||
| gpus_per_prefill: 8 | ||
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 +1,233 @@ | ||
| base: | ||
| name: b300-fp4-glm5.2-agentx | ||
| model: | ||
| path: glm-5.2-fp4 | ||
| container: lmsysorg/sglang:nightly-dev-cu13-20260824-95f5ecd3 | ||
| precision: fp4 | ||
| identity: | ||
| model: | ||
| repo: nvidia/GLM-5.2-NVFP4 | ||
| revision: aec724e | ||
| frameworks: | ||
| dynamo: 1.2.1 | ||
| sglang: nightly-dev-cu13-20260824-95f5ecd3 | ||
| resources: | ||
| gpu_type: b300 | ||
| gpus_per_node: 8 | ||
| frontend: | ||
| type: dynamo | ||
| nginx_session_affinity: true | ||
| nginx_session_affinity_header: X-Dynamo-Session-ID | ||
| env: | ||
| PIP_BREAK_SYSTEM_PACKAGES: '1' | ||
| args: | ||
| router-mode: kv | ||
| router-session-affinity-ttl-secs: '3600' | ||
| active-decode-blocks-threshold: None | ||
| active-prefill-tokens-threshold: None | ||
| active-prefill-tokens-threshold-frac: None | ||
| dynamo: | ||
| hash: 71eb001e17fa73c742f0afe1a6ed96836cb135fd | ||
| install: true | ||
| backend: | ||
| type: sglang | ||
| prefill_environment: | ||
| TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800' | ||
| PYTHONUNBUFFERED: '1' | ||
| DYN_SKIP_SGLANG_LOG_FORMATTING: '1' | ||
| SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000' | ||
| SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000' | ||
| SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000' | ||
| MC_TE_METRIC: 'true' | ||
| NCCL_CUMEM_ENABLE: '1' | ||
| SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0' | ||
| SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1' | ||
| SGLANG_ENABLE_THINKING: '1' | ||
| SGLANG_ENABLE_UNIFIED_RADIX_TREE: '1' | ||
| SGLANG_HICACHE_DEBUG_LOG: '1' | ||
| SGLANG_HICACHE_DEBUG_SAMPLE_RATE: '16384' | ||
| SGLANG_REASONING_EFFORT: max | ||
| PIP_BREAK_SYSTEM_PACKAGES: '1' | ||
| SGLANG_SIMULATE_ACC_LEN: '2.5' | ||
| decode_environment: | ||
| TORCH_DISTRIBUTED_DEFAULT_TIMEOUT: '1800' | ||
| PYTHONUNBUFFERED: '1' | ||
| DYN_SKIP_SGLANG_LOG_FORMATTING: '1' | ||
| SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE: '100000' | ||
| SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT: '100000' | ||
| SGLANG_DISAGGREGATION_WAITING_TIMEOUT: '100000' | ||
| MC_TE_METRIC: 'true' | ||
| NCCL_CUMEM_ENABLE: '1' | ||
| SGLANG_USE_MESSAGE_QUEUE_BROADCASTER: '0' | ||
| SGLANG_DISABLE_TP_MEMORY_INBALANCE_CHECK: '1' | ||
| SGLANG_MOE_NVFP4_DISPATCH: '1' | ||
| SGLANG_NVFP4_CKPT_FP8_NEXTN_MOE: '1' | ||
| SGLANG_ENABLE_THINKING: '1' | ||
| SGLANG_CLIP_MAX_NEW_TOKENS_ESTIMATION: '8' | ||
| SGLANG_REASONING_EFFORT: max | ||
| PIP_BREAK_SYSTEM_PACKAGES: '1' | ||
| SGLANG_SIMULATE_ACC_LEN: '2.5' | ||
| SGLANG_SIMULATE_ACC_METHOD: match-expected | ||
| SGLANG_SIMULATE_ACC_TOKEN_MODE: real-draft-token | ||
| sglang_config: | ||
| prefill: | ||
| served-model-name: GLM-5.2-FP4 | ||
| trust-remote-code: true | ||
| quantization: modelopt_fp4 | ||
| kv-cache-dtype: fp8_e4m3 | ||
| disaggregation-mode: prefill | ||
| disaggregation-transfer-backend: nixl | ||
| tensor-parallel-size: 4 | ||
| data-parallel-size: 4 | ||
| expert-parallel-size: 4 | ||
| enable-dp-attention: true | ||
| enable-dp-lm-head: true | ||
| load-balance-method: total_tokens | ||
| chunked-prefill-size: 65536 | ||
| max-prefill-tokens: 16384 | ||
| max-running-requests: 16 | ||
| cuda-graph-max-bs: 16 | ||
| disable-cuda-graph: true | ||
| nsa-prefill-backend: trtllm | ||
| nsa-decode-backend: trtllm | ||
| moe-runner-backend: flashinfer_cutlass | ||
| fp4-gemm-backend: flashinfer_cutlass | ||
| enable-flashinfer-allreduce-fusion: true | ||
| weight-loader-prefetch-checkpoints: true | ||
| model-loader-extra-config: '{"enable_multithread_load": true}' | ||
| mem-fraction-static: 0.8 | ||
| context-length: 1048576 | ||
| enable-hierarchical-cache: true | ||
| hicache-write-policy: write_back | ||
| hicache-size: 270 | ||
| hicache-io-backend: direct | ||
| speculative-algorithm: EAGLE | ||
| speculative-num-steps: 1 | ||
| speculative-eagle-topk: 1 | ||
| speculative-num-draft-tokens: 2 | ||
| enable-metrics: true | ||
| enable-cache-report: true | ||
| decode: | ||
| served-model-name: GLM-5.2-FP4 | ||
| trust-remote-code: true | ||
| quantization: modelopt_fp4 | ||
| kv-cache-dtype: fp8_e4m3 | ||
| disaggregation-mode: decode | ||
| disaggregation-transfer-backend: nixl | ||
| disable-radix-cache: true | ||
| speculative-algorithm: EAGLE | ||
| speculative-num-steps: 2 | ||
| speculative-eagle-topk: 1 | ||
| speculative-num-draft-tokens: 3 | ||
| tensor-parallel-size: 4 | ||
| data-parallel-size: 1 | ||
| expert-parallel-size: 1 | ||
| enable-dp-attention: false | ||
| enable-dp-lm-head: false | ||
| max-running-requests: 16 | ||
| cuda-graph-max-bs: 16 | ||
| chunked-prefill-size: 64 | ||
| context-length: 1048576 | ||
| nsa-prefill-backend: trtllm | ||
| nsa-decode-backend: trtllm | ||
| moe-runner-backend: flashinfer_trtllm | ||
| fp4-gemm-backend: flashinfer_cutlass | ||
| skip-tokenizer-init: true | ||
| stream-interval: 30 | ||
| enable-flashinfer-allreduce-fusion: true | ||
| weight-loader-prefetch-checkpoints: true | ||
| model-loader-extra-config: '{"enable_multithread_load": true}' | ||
| mem-fraction-static: 0.92 | ||
| disaggregation-decode-extra-slots: 0 | ||
| enable-metrics: true | ||
| enable-cache-report: true | ||
| health_check: | ||
| max_attempts: 1440 | ||
| interval_seconds: 10 | ||
| benchmark: | ||
| type: custom | ||
| env: | ||
| CONCURRENCY: '63' | ||
| DURATION: '3600' | ||
| AIPERF_HF_HOME: /hf_home | ||
| HF_HOME: /hf_home | ||
| HF_HUB_CACHE: /hf_home/hub | ||
| HF_DATASETS_CACHE: /hf_home/datasets | ||
| HF_HUB_OFFLINE: '1' | ||
| HF_DATASETS_OFFLINE: '1' | ||
| HF_HUB_DISABLE_XET: '1' | ||
| XDG_CACHE_HOME: /tmp/cache | ||
| AIPERF_DATASET_MMAP_CACHE_DIR: /hf_home/mmap_cache | ||
| AIPERF_DATASET_CONFIGURATION_TIMEOUT: '1800' | ||
| AIPERF_SERVICE_PROFILE_CONFIGURE_TIMEOUT: '1800' | ||
| AIPERF_FAILED_REQUEST_THRESHOLD: '0.1' | ||
| AIPERF_RECORD_STRIP_PAYLOAD_BYTES: 'true' | ||
| AIPERF_HTTP_X_DYNAMO_SESSION_ID_FROM_CORRELATION_ID: 'true' | ||
| AIPERF_REQUIRED_SERVER_METRIC_PREFIX: 'sglang:' | ||
| command: | | ||
| set -euo pipefail | ||
| export HF_HOME="${AIPERF_HF_HOME:-/hf_home}" | ||
| export HF_HUB_CACHE="${HF_HOME}/hub" | ||
| export HF_DATASETS_CACHE="${HF_HOME}/datasets" | ||
| mkdir -p "${XDG_CACHE_HOME}" "${AIPERF_DATASET_MMAP_CACHE_DIR}" /logs/aiperf | ||
| /aiperf-venv/bin/aiperf profile \ | ||
| --scenario inferencex-agentx-mvp \ | ||
| --url http://localhost:8000 \ | ||
| --endpoint /v1/chat/completions \ | ||
| --endpoint-type chat \ | ||
| --streaming \ | ||
| --model GLM-5.2-FP4 \ | ||
| --tokenizer /model \ | ||
| --tokenizer-trust-remote-code \ | ||
| --concurrency "${CONCURRENCY}" \ | ||
| --benchmark-duration "${DURATION}" \ | ||
| --random-seed 42 \ | ||
| --failed-request-threshold "${AIPERF_FAILED_REQUEST_THRESHOLD}" \ | ||
| --stats-interval 30 \ | ||
| --warmup-requests-per-lane 10 \ | ||
| --warmup-grace-period 1800 \ | ||
| --trajectory-start-min-ratio 0.25 \ | ||
| --trajectory-start-max-ratio 0.75 \ | ||
| --use-server-token-count \ | ||
| --no-gpu-telemetry \ | ||
| --num-dataset-entries 393 \ | ||
| --slice-duration 1.0 \ | ||
| --public-dataset semianalysis_cc_traces_weka_062126 \ | ||
| --trace-idle-gap-cap-seconds 300 \ | ||
| --unsafe-override \ | ||
| --ui simple \ | ||
| --server-metrics-formats json \ | ||
| --output-artifact-dir /logs/aiperf | ||
| rm -f /logs/aiperf/inputs.json | ||
|
There was a problem hiding this comment. Recipes skip InferenceX agentic harnessHigh Severity Both recipes invoke Additional Locations (1)  Reviewed by Cursor Bugbot for commit 6588cb2. Configure here. |
||
| infra: | ||
| etcd_nats_dedicated_node: false | ||
| nats_max_payload_mb: 64 | ||
| sbatch_directives: | ||
| mem: '0' | ||
| srun_options: | ||
| mem: '0' | ||
| container-remap-root: '' | ||
| # ################# agentx ################# | ||
| zip_override_mtp_agentx_lowlat: | ||
| name: [agentx-1p2d_tp4-c63, agentx-1p3d_tp4-c61, agentx-1p4d_tp4-c59, agentx-1p6d_tp4-c58, agentx-2p1d_dtp8-c139] | ||
| backend: | ||
| sglang_config: | ||
| decode: | ||
| cuda-graph-max-bs: [16, 16, 16, 16, 139] | ||
| data-parallel-size: [1, 1, 1, 1, 8] | ||
| enable-dp-attention: [false, false, false, false, true] | ||
| enable-dp-lm-head: [false, false, false, false, true] | ||
| max-running-requests: [16, 16, 16, 16, 139] | ||
| tensor-parallel-size: [4, 4, 4, 4, 8] | ||
|
There was a problem hiding this comment. DEP8 decode missing expert parallelHigh Severity The throughput point enables decode Additional Locations (1)Reviewed by Cursor Bugbot for commit e8c0a18. Configure here. |
||
| benchmark: | ||
| env: | ||
| CONCURRENCY: ['63', '61', '59', '58', '139'] | ||
| frontend: | ||
| enable_multiple_frontends: true | ||
| resources: | ||
| decode_nodes: [1, 2, 2, 3, 1] | ||
| decode_workers: [2, 3, 4, 6, 1] | ||
| gpus_per_decode: [4, 4, 4, 4, 8] | ||
| gpus_per_prefill: 4 | ||
| prefill_nodes: 1 | ||
| prefill_workers: [1, 1, 1, 1, 2] | ||
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.
Aggregated recipe missing GPU field
High Severity
The aggregated zip override sets
agg_nodes/agg_workersbut usesgpus_per_decodeandgpus_per_prefillinstead ofgpus_per_agg. Every other aggregated recipe in-repo (GB200/GB300 GLM, DeepSeek H200/GB300) usesgpus_per_agg. Aggregated worker GPU allocation is likely wrong or rejected.Reviewed by Cursor Bugbot for commit e8c0a18. Configure here.