340 lines
9.2 KiB
YAML
340 lines
9.2 KiB
YAML
test_name: "GLM-5.2-w4a8c8-128k-1k-90-50-PD"
|
|
model: "Eco-Tech/GLM-5.2-w4a8c8"
|
|
num_nodes: 4
|
|
npu_per_node: 16
|
|
|
|
|
|
routing:
|
|
type: "disaggregated_prefill"
|
|
groups:
|
|
prefiller: [ 0, 1 ]
|
|
decoder: [ 2, 3 ]
|
|
|
|
config:
|
|
- node_index: 0
|
|
port_start: 7100
|
|
dp_rpc_port: 13321
|
|
dp_size: 4
|
|
dp_size_local: 2
|
|
dp_rank_start: 0
|
|
tp_size: 8
|
|
dp_address: "${NODE_0_IP}"
|
|
|
|
- node_index: 1
|
|
port_start: 7100
|
|
dp_rpc_port: 13321
|
|
dp_size: 4
|
|
dp_size_local: 2
|
|
dp_rank_start: 2
|
|
tp_size: 8
|
|
dp_address: "${NODE_0_IP}"
|
|
|
|
- node_index: 2
|
|
port_start: 7100
|
|
dp_rpc_port: 13321
|
|
dp_size: 8
|
|
dp_size_local: 4
|
|
dp_rank_start: 0
|
|
tp_size: 4
|
|
dp_address: "${NODE_2_IP}"
|
|
|
|
- node_index: 3
|
|
port_start: 7100
|
|
dp_rpc_port: 13321
|
|
dp_size: 8
|
|
dp_size_local: 4
|
|
dp_rank_start: 4
|
|
tp_size: 4
|
|
dp_address: "${NODE_2_IP}"
|
|
|
|
env_common: &env_common
|
|
VLLM_USE_MODELSCOPE: "true"
|
|
SERVER_PORT: "${PORT}"
|
|
ASCEND_RT_VISIBLE_DEVICES: "${VISIBLE_DEVICES}"
|
|
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
|
|
HCCL_OP_EXPANSION_MODE: "AIV"
|
|
OMP_PROC_BIND: "false"
|
|
OMP_NUM_THREADS: "1"
|
|
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
|
|
ACL_OP_INIT_MODE: "1"
|
|
ASCEND_A3_ENABLE: "1"
|
|
|
|
templates:
|
|
- node_index: 0
|
|
envs:
|
|
<<: *env_common
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
HCCL_BUFFSIZE: "400"
|
|
server_cmd_template:
|
|
- --host
|
|
- "0.0.0.0"
|
|
- --port
|
|
- $SERVER_PORT
|
|
- --data-parallel-size
|
|
- ${DP_SIZE}
|
|
- --data-parallel-rank
|
|
- ${DP_RANK}
|
|
- --data-parallel-address
|
|
- ${DP_ADDRESS}
|
|
- --data-parallel-rpc-port
|
|
- ${DP_RPC_PORT}
|
|
- --tensor-parallel-size
|
|
- ${TP_SIZE}
|
|
- --enable-expert-parallel
|
|
- --hf-overrides
|
|
- '{"use_index_cache": true, "index_topk_freq": 4}'
|
|
- --speculative-config
|
|
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp","enforce_eager":true}'
|
|
- --seed
|
|
- "1024"
|
|
- --max-model-len
|
|
- "202752"
|
|
- --additional-config
|
|
- '{"fuse_muls_add":true,"recompute_scheduler_enable" : false,"multistream_overlap_shared_expert": true, "enable_dsa_cp":true,"enable_sparse_sfa_c8": true, "enable_sparse_li_c8": true,"c8_enable_reshape_optim":true}'
|
|
- --max-num-batched-tokens
|
|
- "8192"
|
|
- --trust-remote-code
|
|
- --enable-prefix-caching
|
|
- --max-num-seqs
|
|
- "64"
|
|
- --quantization
|
|
- "ascend"
|
|
- --gpu-memory-utilization
|
|
- "0.92"
|
|
- --enforce-eager
|
|
- --enable-auto-tool-choice
|
|
- --tool-call-parser
|
|
- "glm47"
|
|
- --reasoning-parser
|
|
- "glm45"
|
|
- --kv-transfer-config
|
|
- '{"kv_connector": "MooncakeConnectorV1",
|
|
"kv_role": "kv_producer",
|
|
"kv_port": "30000",
|
|
"engine_id": "0",
|
|
"kv_connector_extra_config": {
|
|
"use_ascend_direct": true,
|
|
"prefill": {
|
|
"dp_size": 4,
|
|
"tp_size": 8
|
|
},
|
|
"decode": {
|
|
"dp_size": 8,
|
|
"tp_size": 4
|
|
}
|
|
}
|
|
}'
|
|
- node_index: 1
|
|
envs:
|
|
<<: *env_common
|
|
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
|
|
HCCL_BUFFSIZE: "400"
|
|
server_cmd_template:
|
|
- --host
|
|
- "0.0.0.0"
|
|
- --port
|
|
- $SERVER_PORT
|
|
- --data-parallel-size
|
|
- ${DP_SIZE}
|
|
- --data-parallel-rank
|
|
- ${DP_RANK}
|
|
- --data-parallel-address
|
|
- ${DP_ADDRESS}
|
|
- --data-parallel-rpc-port
|
|
- ${DP_RPC_PORT}
|
|
- --tensor-parallel-size
|
|
- ${TP_SIZE}
|
|
- --enable-expert-parallel
|
|
- --hf-overrides
|
|
- '{"use_index_cache": true, "index_topk_freq": 4}'
|
|
- --speculative-config
|
|
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp","enforce_eager":true}'
|
|
- --seed
|
|
- "1024"
|
|
- --max-model-len
|
|
- "202752"
|
|
- --additional-config
|
|
- '{"fuse_muls_add":true,"recompute_scheduler_enable" : false,"multistream_overlap_shared_expert": true, "enable_dsa_cp":true,"enable_sparse_sfa_c8": true, "enable_sparse_li_c8": true,"c8_enable_reshape_optim":true}'
|
|
- --max-num-batched-tokens
|
|
- "8192"
|
|
- --trust-remote-code
|
|
- --enable-prefix-caching
|
|
- --max-num-seqs
|
|
- "64"
|
|
- --quantization
|
|
- "ascend"
|
|
- --gpu-memory-utilization
|
|
- "0.92"
|
|
- --enforce-eager
|
|
- --enable-auto-tool-choice
|
|
- --tool-call-parser
|
|
- "glm47"
|
|
- --reasoning-parser
|
|
- "glm45"
|
|
- --kv-transfer-config
|
|
- '{"kv_connector": "MooncakeConnectorV1",
|
|
"kv_role": "kv_producer",
|
|
"kv_port": "30000",
|
|
"engine_id": "1",
|
|
"kv_connector_extra_config": {
|
|
"use_ascend_direct": true,
|
|
"prefill": {
|
|
"dp_size": 4,
|
|
"tp_size": 8
|
|
},
|
|
"decode": {
|
|
"dp_size": 8,
|
|
"tp_size": 4
|
|
}
|
|
}
|
|
}'
|
|
- node_index: 2
|
|
envs:
|
|
<<: *env_common
|
|
HCCL_BUFFSIZE: "256"
|
|
TASK_QUEUE_ENABLE: "1"
|
|
VLLM_ASCEND_ENABLE_MLAPO: "1"
|
|
server_cmd_template:
|
|
- --host
|
|
- "0.0.0.0"
|
|
- --port
|
|
- $SERVER_PORT
|
|
- --data-parallel-size
|
|
- ${DP_SIZE}
|
|
- --data-parallel-rank
|
|
- ${DP_RANK}
|
|
- --data-parallel-address
|
|
- ${DP_ADDRESS}
|
|
- --data-parallel-rpc-port
|
|
- ${DP_RPC_PORT}
|
|
- --tensor-parallel-size
|
|
- ${TP_SIZE}
|
|
- --enable-expert-parallel
|
|
- --hf-overrides
|
|
- '{"use_index_cache": true, "index_topk_freq": 4}'
|
|
- --seed
|
|
- "1024"
|
|
- --max-model-len
|
|
- "202752"
|
|
- --max-num-batched-tokens
|
|
- "164"
|
|
- --compilation-config
|
|
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- --speculative-config
|
|
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp","enforce_eager":true}'
|
|
- --additional-config
|
|
- '{"fuse_muls_add":true,"recompute_scheduler_enable":true,"multistream_overlap_shared_expert":true,"enable_sparse_sfa_c8": true, "enable_sparse_li_c8": true}'
|
|
- --trust-remote-code
|
|
- --max-num-seqs
|
|
- "32"
|
|
- --gpu-memory-utilization
|
|
- "0.92"
|
|
- --async-scheduling
|
|
- --quantization
|
|
- "ascend"
|
|
- --enable-auto-tool-choice
|
|
- --tool-call-parser
|
|
- "glm47"
|
|
- --reasoning-parser
|
|
- "glm45"
|
|
- --enable-prefix-caching
|
|
- --kv-transfer-config
|
|
- '{"kv_connector": "MooncakeConnectorV1",
|
|
"kv_role": "kv_consumer",
|
|
"kv_port": "30200",
|
|
"engine_id": "2",
|
|
"kv_connector_extra_config": {
|
|
"use_ascend_direct": true,
|
|
"prefill": {
|
|
"dp_size": 4,
|
|
"tp_size": 8
|
|
},
|
|
"decode": {
|
|
"dp_size": 8,
|
|
"tp_size": 4
|
|
}
|
|
}
|
|
}'
|
|
- node_index: 3
|
|
envs:
|
|
<<: *env_common
|
|
HCCL_BUFFSIZE: "256"
|
|
TASK_QUEUE_ENABLE: "1"
|
|
VLLM_ASCEND_ENABLE_MLAPO: "1"
|
|
server_cmd_template:
|
|
- --host
|
|
- "0.0.0.0"
|
|
- --port
|
|
- $SERVER_PORT
|
|
- --data-parallel-size
|
|
- ${DP_SIZE}
|
|
- --data-parallel-rank
|
|
- ${DP_RANK}
|
|
- --data-parallel-address
|
|
- ${DP_ADDRESS}
|
|
- --data-parallel-rpc-port
|
|
- ${DP_RPC_PORT}
|
|
- --tensor-parallel-size
|
|
- ${TP_SIZE}
|
|
- --enable-expert-parallel
|
|
- --hf-overrides
|
|
- '{"use_index_cache": true, "index_topk_freq": 4}'
|
|
- --seed
|
|
- "1024"
|
|
- --max-model-len
|
|
- "202752"
|
|
- --max-num-batched-tokens
|
|
- "164"
|
|
- --compilation-config
|
|
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
|
|
- --speculative-config
|
|
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp","enforce_eager":true}'
|
|
- --additional-config
|
|
- '{"fuse_muls_add":true,"recompute_scheduler_enable":true,"multistream_overlap_shared_expert":true,"enable_sparse_sfa_c8": true, "enable_sparse_li_c8": true}'
|
|
- --trust-remote-code
|
|
- --max-num-seqs
|
|
- "32"
|
|
- --gpu-memory-utilization
|
|
- "0.92"
|
|
- --async-scheduling
|
|
- --quantization
|
|
- "ascend"
|
|
- --enable-auto-tool-choice
|
|
- --tool-call-parser
|
|
- "glm47"
|
|
- --reasoning-parser
|
|
- "glm45"
|
|
- --enable-prefix-caching
|
|
- --kv-transfer-config
|
|
- '{"kv_connector": "MooncakeConnectorV1",
|
|
"kv_role": "kv_consumer",
|
|
"kv_port": "30200",
|
|
"engine_id": "3",
|
|
"kv_connector_extra_config": {
|
|
"use_ascend_direct": true,
|
|
"prefill": {
|
|
"dp_size": 4,
|
|
"tp_size": 8
|
|
},
|
|
"decode": {
|
|
"dp_size": 8,
|
|
"tp_size": 4
|
|
}
|
|
}
|
|
}'
|
|
|
|
|
|
|
|
benchmarks:
|
|
perf:
|
|
case_type: performance
|
|
dataset_path: vllm-ascend/GSM8K-in131072-bs500-prefix90-glm51
|
|
request_conf: vllm_api_stream_chat
|
|
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
|
|
num_prompts: 256
|
|
max_out_len: 1024
|
|
batch_size: 64
|
|
request_rate: 1.52
|
|
baseline: 1294.76
|
|
threshold: 0.97
|