EngineX-Ascend/enginex-ascend-910-vllm
11
0
Fork 1
Code Issues Pull Requests Actions Projects Releases Wiki Activity

init v0.23.0

Browse Source 
Signed-off-by: Sun Ruoxi <sunruoxi@4paradigm.com>
This commit is contained in:
Sun Ruoxi
2026-08-27 15:11:51 +08:00
parent b582a8e7d1
commit 7f8a1b1f7a
2849 changed files with 712887 additions and 22001 deletions
Download Patch File Download Diff File Expand all files Collapse all files

84
tests/e2e/nightly/single_node/models/configs/DeepSeek-R1-0528-W8A8.yaml Normal file
View File

@@ -0,0 +1,84 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
OMP_NUM_THREADS: "10"
OMP_PROC_BIND: "false"
HCCL_BUFFSIZE: "1024"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ENGINE_READY_TIMEOUT_S: "3000"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--data-parallel-size"
- "2"
- "--tensor-parallel-size"
- "8"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--seed"
- "1024"
- "--max-model-len"
- "36864"
- "--max-num-batched-tokens"
- "4096"
- "--max-num-seqs"
- "16"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--speculative-config"
- '{"num_speculative_tokens": 1, "method": "mtp"}'
- "--additional-config"
- '{"enable_weight_nz_layout": true}'
_benchmarks_acc: &benchmarks_acc
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 95
threshold: 10
_benchmarks_perf: &benchmarks_perf
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 400
max_out_len: 1500
batch_size: 1000
baseline: 1
threshold: 0.97
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "DeepSeek-R1-0528-W8A8-single"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--enforce-eager"
benchmarks:
- name: "DeepSeek-R1-0528-W8A8-aclgraph"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
benchmarks:
<<: *benchmarks_acc
<<: *benchmarks_perf

81
tests/e2e/nightly/single_node/models/configs/DeepSeek-V3.2-W8A8-DCP.yaml Normal file
View File

@@ -0,0 +1,81 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "DeepSeek-V3.2-W8A8-DCP-replicated-indexer"
model: "vllm-ascend/DeepSeek-V3.2-W8A8"
envs:
VLLM_ASCEND_ENABLE_NZ: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "20"
HCCL_BUFFSIZE: "768"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_SERVER_DEV_MODE: "1"
VLLM_WORKER_MULTIPROC_METHOD: "spawn"
ASCEND_LAUNCH_BLOCKING: "0"
ASCEND_ENABLE_USE_FABRIC_MEM: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "0"
VLLM_ASCEND_ENABLE_MLAPO: "0"
PYTHONHASHSEED: "0"
ASCEND_A3_ENABLE: "1"
VLLM_ENGINE_READY_TIMEOUT_S: "10000"
VLLM_RPC_TIMEOUT: "3600000"
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: "30000"
TASK_QUEUE_ENABLE: "1"
CPU_AFFINITY_CONF: "1"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--port"
- "$SERVER_PORT"
- "--seed"
- "1024"
- "--max-model-len"
- "8192"
- "--max-num-batched-tokens"
- "1024"
- "--max-num-seqs"
- "32"
- "--trust-remote-code"
- "--quantization"
- "ascend"
- "--data-parallel-size"
- "1"
- "--pipeline-parallel-size"
- "1"
- "--tensor-parallel-size"
- "16"
- "--prefill-context-parallel-size"
- "1"
- "--decode-context-parallel-size"
- "16"
- "--cp-kv-cache-interleave-size"
- "1"
- "--block-size"
- "128"
- "--enable-expert-parallel"
- "--gpu-memory-utilization"
- "0.95"
- "--api-server-count"
- "1"
- "--safetensors-load-strategy"
- "prefetch"
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY", "cudagraph_capture_sizes":[4, 16, 64, 128]}'
- "--additional-config"
- '{"enable_dsa_cp": true, "ascend_compilation_config":{"enable_npugraph_ex": true, "enable_static_kernel": false}, "multistream_overlap_shared_expert": true, "enable_mc2_hierarchy_comm": false, "enable_sparse_sfa_c8": true, "enable_sparse_li_c8": true, "enable_cpu_binding": true, "recompute_scheduler_enable": false}'
- "--speculative-config"
- '{"num_speculative_tokens": 3, "method": "deepseek_mtp"}'
test_content: []
benchmarks:
acc_gsm8k:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 8192
batch_size: 32
baseline: 95
threshold: 5

80
tests/e2e/nightly/single_node/models/configs/DeepSeek-V3.2-W8A8.yaml Normal file
View File

@@ -0,0 +1,80 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "DeepSeek-V3.2-W8A8-TP8-DP2"
model: "vllm-ascend/DeepSeek-V3.2-W8A8"
envs:
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_USE_V1: "1"
HCCL_BUFFSIZE: "256"
ASCEND_AGGREGATE_ENABLE: "1"
ASCEND_TRANSPORT_PRINT: "1"
ACL_OP_INIT_MODE: "1"
ASCEND_A3_ENABLE: "1"
VLLM_NIXL_ABORT_REQUEST_TIMEOUT: "300000"
TASK_QUEUE_ENABLE: "1"
VLLM_ASCEND_ENABLE_MLAPO: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ENGINE_READY_TIMEOUT_S: "3000"
server_cmd:
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--seed"
- "1024"
- "--max-model-len"
- "67000"
- "--max-num-batched-tokens"
- "4096"
- "--max-num-seqs"
- "8"
- "--trust-remote-code"
- "--quantization"
- "ascend"
- "--async-scheduling"
- "--no-enable-prefix-caching"
- "--enable-expert-parallel"
- "--gpu-memory-utilization"
- "0.95"
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,2,4,8,16,24,32,40,48], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--speculative-config"
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}'
- "--reasoning-parser"
- "deepseek_v3"
- "--tokenizer_mode"
- "deepseek_v32"
benchmarks:
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 86.67
temperature: 1.0
top_p: 0.95
thinking: true
threshold: 10
perf_2:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 16
max_out_len: 1500
batch_size: 4
request_rate: 0
baseline: 1
threshold: 0.97

80
tests/e2e/nightly/single_node/models/configs/DeepSeek-V4-Flash-W8A8-A3.yaml Normal file
View File

@@ -0,0 +1,80 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "DeepSeek-V4-Flash-W8A8-A3"
model: "Eco-Tech/DeepSeek-V4-Flash-w8a8-mtp"
special_dependencies:
transformers: "5.9.0"
envs:
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_BUFFSIZE: "1024"
VLLM_ASCEND_ENABLE_FUSED_MC2: "0"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
ASCEND_LAUNCH_BLOCKING: "0"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ENGINE_READY_TIMEOUT_S: "3000"
server_cmd:
- "--enable-prefix-caching"
- "--max-model-len"
- "1048576"
- "--max-num-batched-tokens"
- "10240"
- "--gpu-memory-utilization"
- "0.9"
- "--max-num-seqs"
- "64"
- "--data-parallel-size"
- "4"
- "--tensor-parallel-size"
- "4"
- "--enable-expert-parallel"
- "--tokenizer-mode"
- "deepseek_v4"
- "--tool-call-parser"
- "deepseek_v4"
- "--enable-auto-tool-choice"
- "--reasoning-parser"
- "deepseek_v4"
- "--safetensors-load-strategy"
- "prefetch"
- "--quantization"
- "ascend"
- "--api-server-count"
- "1"
- "--speculative-config"
- '{"num_speculative_tokens": 1,"method": "mtp","enforce_eager": true}'
- "--port"
- "$SERVER_PORT"
- "--block-size"
- "128"
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
- "--async-scheduling"
- "--additional-config"
- '{"ascend_compilation_config":{"enable_npugraph_ex":true,"enable_static_kernel":false},"enable_cpu_binding":"true","enable_shared_expert_dp":true,"multistream_overlap_shared_expert":true}'
benchmarks:
acc-gpqa:
case_type: accuracy
dataset_path: vllm-ascend/gpqa
request_conf: vllm_api_general_chat
dataset_conf: gpqa/gpqa_gen_0_shot_cot_chat_prompt
max_out_len: 65536
batch_size: 32
baseline: 86.36
threshold: 5
thinking: true
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K_prefix0_in32768_bs1000_deepseek
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 64
max_out_len: 1024
batch_size: 16
request_rate: 0
baseline: 1
threshold: 0.97

66
tests/e2e/nightly/single_node/models/configs/GLM-4.7.yaml Normal file
View File

@@ -0,0 +1,66 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
HCCL_BUFFSIZE: "512"
SERVER_PORT: "DEFAULT_PORT"
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
VLLM_ASCEND_BALANCE_SCHEDULING: "1"
VLLM_ASCEND_ENABLE_TOPK_OPTIMIZE: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
_server_cmd: &server_cmd
- "--enable-expert-parallel"
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "8192"
- "--max-num-batched-tokens"
- "8192"
- "--max-num-seqs"
- "16"
- "--quantization"
- "ascend"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--speculative-config"
- '{"num_speculative_tokens": 3, "method":"mtp"}'
- "--additional-config"
- '{"enable_shared_expert_dp": true, "ascend_fusion_config": {"fusion_ops_gmmswigluquant": false}}'
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 4096
batch_size: 8
baseline: 95
threshold: 10
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "GLM-4.7-TP8-DP2-decodegraph"
model: "Eco-Tech/GLM-4.7-W8A8-floatmtp"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation-config"
- '{"cudagraph_capture_sizes": [1,2,4,8,16,32,64,128,256,512], "cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
<<: *benchmarks

82
tests/e2e/nightly/single_node/models/configs/GLM-5.1-W8A8-PrefillMC2.yaml Normal file
View File

@@ -0,0 +1,82 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "GLM-5.1-W8A8-PrefillMC2"
model: "Eco-Tech/GLM-5.1-w8a8" #need update
envs:
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_USE_V1: "1"
HCCL_BUFFSIZE: "1800"
ASCEND_AGGREGATE_ENABLE: "1"
ASCEND_TRANSPORT_PRINT: "1"
ACL_OP_INIT_MODE: "1"
ASCEND_A3_ENABLE: "1"
VLLM_NIXL_ABORT_REQUEST_TIMEOUT: "300000"
TASK_QUEUE_ENABLE: "1"
VLLM_ASCEND_ENABLE_MLAPO: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "16"
- "--data-parallel-size"
- "1"
- "--port"
- "$SERVER_PORT"
- "--seed"
- "1024"
- "--max-model-len"
- "10240"
- "--max-num-batched-tokens"
- "4096"
- "--max-num-seqs"
- "32"
- "--trust-remote-code"
- "--quantization"
- "ascend"
- "--async-scheduling"
- "--no-enable-prefix-caching"
- "--enable-expert-parallel"
- "--gpu-memory-utilization"
- "0.94"
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--speculative-config"
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp", "enforce_eager": true}'
- "--additional_config"
- '{"enable_prefill_mc2": true}'
- "--reasoning-parser"
- "glm45"
- "--tool-call-parser"
- "glm47"
benchmarks:
acc_gsm8k:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 8192
batch_size: 32
baseline: 96.88
temperature: 1.0
top_p: 0.95
thinking: true
threshold: 5
perf_2:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 64
max_out_len: 1500
batch_size: 32
request_rate: 0
baseline: 1
threshold: 0.97

58
tests/e2e/nightly/single_node/models/configs/Hy3-preview.yaml Normal file
View File

@@ -0,0 +1,58 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
VLLM_USE_MODELSCOPE: "true"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1024"
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--tensor-parallel-size"
- "16"
- "--enable-expert-parallel"
- "--enable-ep-weight-filter"
- "--tool-call-parser"
- "hy_v3"
- "--reasoning-parser"
- "hy_v3"
- "--enable-auto-tool-choice"
- "--max-model-len"
- "32768"
- "--max-num-seqs"
- "8"
- "--port"
- "$SERVER_PORT"
- "--speculative-config"
- '{"method": "mtp", "num_speculative_tokens": 1}'
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
_benchmarks: &benchmarks
acc_gsm8k:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_4_shot_cot_chat_prompt
max_out_len: 4096
batch_size: 8
baseline: 93.07
threshold: 10
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Hy3-preview-TP16-EP-MTP"
model: "Tencent-Hunyuan/Hy3-preview"
envs:
<<: *envs
server_cmd: *server_cmd
benchmarks:
<<: *benchmarks

52
tests/e2e/nightly/single_node/models/configs/Kimi-K2-Thinking.yaml Normal file
View File

@@ -0,0 +1,52 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Kimi-K2-Thinking-TP16-Case"
model: "moonshotai/Kimi-K2-Thinking"
envs:
HCCL_BUFFSIZE: "1024"
TASK_QUEUE_ENABLE: "1"
OMP_PROC_BIND: "false"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "16"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "8192"
- "--max-num-batched-tokens"
- "8192"
- "--max-num-seqs"
- "12"
- "--gpu-memory-utilization"
- "0.9"
- "--trust-remote-code"
- "--enable-expert-parallel"
- "--no-enable-prefix-caching"
benchmarks:
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 4096
batch_size: 32
baseline: 95
threshold: 10
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 512
max_out_len: 256
batch_size: 64
trust_remote_code: true
request_rate: 11.2
baseline: 1
threshold: 0.97

91
tests/e2e/nightly/single_node/models/configs/Kimi-K2.5.yaml Normal file
View File

@@ -0,0 +1,91 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
HCCL_BUFFSIZE: "512"
SERVER_PORT: "DEFAULT_PORT"
HCCL_OP_EXPANSION_MODE: "AIV"
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
TASK_QUEUE_ENABLE: "1"
VLLM_ASCEND_ENABLE_MLAPO: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM_ASCEND_ENABLE_NZ: "1"
_server_cmd: &server_cmd
- "--enable-expert-parallel"
- "--enable-prefix-caching"
- "--enable-chunked-prefill"
- "--allowed-local-media-path"
- "/"
- "--tensor-parallel-size"
- "4"
- "--data-parallel-size"
- "4"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "133120"
- "--max-num-batched-tokens"
- "8192"
- "--max-num-seqs"
- "16"
- "--quantization"
- "ascend"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--seed"
- "42"
- "--compilation-config"
- '{"cudagraph_capture_sizes":[4,8,12,16,32], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--speculative-config"
- '{"method":"eagle3", "model":"lightseekorg/kimi-k2.5-eagle3", "num_speculative_tokens":3}'
- "--additional-config"
- '{"enable_shared_expert_dp":true}'
- "--mm-processor-cache-gb"
- "0"
- "--mm-encoder-tp-mode"
- "data"
_benchmarks: &benchmarks
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 65536
temperature: 0.0
top_p: 1
top_k: -1
repetition_penalty: 1.0
batch_size: 32
baseline: 95
threshold: 10
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K_prefix90_in131072_bs1000_kimi
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 8
max_out_len: 1024
batch_size: 2
trust_remote_code: true
request_rate: 0
baseline: 1
threshold: 0.97
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Kimi-K2.5-W4A8-Case"
model: "Eco-Tech/Kimi-K2.5-W4A8"
envs:
<<: *envs
server_cmd: *server_cmd
benchmarks:
<<: *benchmarks

70
tests/e2e/nightly/single_node/models/configs/Kimi-K2.6-w4a8-A3.yaml Normal file
View File

@@ -0,0 +1,70 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Kimi-K2.6-W4A8-in3.5k-out1.5k-TPOT50-0-128-32"
model: "Eco-Tech/Kimi-K2.6-w4a8"
envs:
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
HCCL_BUFFSIZE: "800"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM_ASCEND_ENABLE_MLAPO: "1"
VLLM_ASCEND_BALANCE_SCHEDULING: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
DYNAMIC_EPLB: "true"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--quantization"
- "ascend"
- "--port"
- "$SERVER_PORT"
- "--allowed-local-media-path"
- "/"
- "--trust-remote-code"
- "--safetensors-load-strategy"
- 'prefetch'
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "2"
- "--no-enable-prefix-caching"
- "--enable-expert-parallel"
- "--max-num-seqs"
- "24"
- "--max-model-len"
- "6144"
- "--max-num-batched-tokens"
- "4096"
- "--gpu-memory-utilization"
- "0.85"
- "--seed"
- "42"
- "--async-scheduling"
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--additional-config"
- '{"eplb_config": {"dynamic_eplb": true},"ascend_compilation_config": {"enable_static_kernel": false}}'
- "--profiler-config"
- '{"profiler": "torch", "torch_profiler_dir": "./vllm_profile", "torch_profiler_with_stack": true}'
- "--mm-processor-cache-gb"
- "0"
- "--mm-encoder-tp-mode"
- "data"
- "--speculative-config"
- '{"method": "dflash","model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 7}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs4096-prefix0-kimi
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 128
max_out_len: 1500
batch_size: 32
request_rate: 0
baseline: 1433.4454
threshold: 0.97

91
tests/e2e/nightly/single_node/models/configs/MTPX-DeepSeek-R1-0528-W8A8.yaml Normal file
View File

@@ -0,0 +1,91 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
OMP_NUM_THREADS: "100"
OMP_PROC_BIND: "false"
HCCL_BUFFSIZE: "1024"
VLLM_RPC_TIMEOUT: "3600000"
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: "3600000"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ENGINE_READY_TIMEOUT_S: "7200"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--seed"
- "1024"
- "--no-enable-prefix-caching"
- "--data-parallel-size"
- "2"
- "--tensor-parallel-size"
- "8"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "40960"
- "--max-num-seqs"
- "14"
- "--trust-remote-code"
_benchmarks_gsm8k: &benchmarks_gsm8k
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 95
threshold: 10
_benchmarks_aime: &benchmarks_aime
acc:
case_type: accuracy
dataset_path: vllm-ascend/aime2024
request_conf: vllm_api_general_chat
dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 86.67
threshold: 10
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "MTPX-DeepSeek-R1-0528-W8A8-mtp2"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--max-num-batched-tokens"
- "4096"
- "--speculative-config"
- '{"num_speculative_tokens": 2, "method": "mtp"}'
- "--gpu-memory-utilization"
- "0.92"
benchmarks:
<<: *benchmarks_gsm8k
- name: "MTPX-DeepSeek-R1-0528-W8A8-mtp3"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
envs:
<<: *envs
HCCL_OP_EXPANSION_MODE: "AIV"
server_cmd: *server_cmd
server_cmd_extra:
- "--max-num-batched-tokens"
- "2048"
- "--speculative-config"
- '{"num_speculative_tokens": 3, "method": "mtp"}'
- "--gpu-memory-utilization"
- "0.9"
- "--compilation-config"
- '{"cudagraph_capture_sizes": [56], "cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
<<: *benchmarks_aime

88
tests/e2e/nightly/single_node/models/configs/MiniMax-M2.5-w8a8-QuaRot-A2.yaml Normal file
View File

@@ -0,0 +1,88 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "MiniMax-M2.5-w8a8"
model: "Eco-Tech/MiniMax-M2.5-w8a8-QuaRot"
envs:
HCCL_BUFFSIZE: "512"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
VLLM-ASCEND_BALANCE_SCHEDULING: "1"
HCCL_INTRA_PCIE_ENABLE: "1"
HCCL_INTRA_ROCE_ENABLE: "0"
OMP_PROC_BIND: "false"
VLLM_TORCH_PROFILER_WITH_STACK: "0"
VLLM_TORCH_PROFILER_DIR: "./profile"
VLLM_USE_MODELSCOPE: "true"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "8"
- "--port"
- "$SERVER_PORT"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--quantization"
- "ascend"
- "--additional-config"
- '{"enable_cpu_binding":true}'
- "--model-loader-extra-config"
- '{"enable_multithread_load":true,"num_threads":16}'
- "--speculative_config"
- '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
- "--enable-expert-parallel"
- "--enable-chunked-prefill"
- "--enable-prefix-caching"
- "--max-num-seqs"
- "100"
- "--max-model-len"
- "196608"
- "--seed"
- "1024"
- "--max-num-batched-tokens"
- "6144"
- "--enable-auto-tool-choice"
- "--tool-call-parser"
- "minimax_m2"
- "--reasoning-parser"
- "minimax_m2_append_think"
- "--enable-force-include-usage"
- "--profiler-config"
- '{"profiler":"torch","torch_profiler_dir":"./profile","torch_profiler_with_stack":false}'
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[4,16,40,80,160,256,400]}'
benchmarks:
acc:
case_type: accuracy
dataset_path: vllm-ascend/gpqa
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gpqa_gen_0_shot_str
max_out_len: 131072
batch_size: 64
baseline: 83
threshold: 5
bos_token_id: 200019
do_sample: true
eos_token_id: 200020
temperature: 1.0
top_p: 0.95
top_k: 40
transformers_version: 4.46.1
ignore_eos: false
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 360
max_out_len: 1500
batch_size: 120
request_rate: 0
baseline: 2042
threshold: 0.97

73
tests/e2e/nightly/single_node/models/configs/MiniMax-M2.5-w8a8-QuaRot-A3.yaml Normal file
View File

@@ -0,0 +1,73 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "MiniMax-M2.5-w8a8"
model: "Eco-Tech/MiniMax-M2.5-w8a8-QuaRot"
envs:
HCCL_BUFFSIZE: "512"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
VLLM-ASCEND_ENABLE_NZ: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM-ASCEND_BALANCE_SCHEDULING: "1"
VLLM_USE_MODELSCOPE: "true"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "1"
- "--port"
- "$SERVER_PORT"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.85"
- "--quantization"
- "ascend"
- "--no-enable-prefix-caching"
- "--additional-config"
- '{"enable_cpu_binding":true}'
- "--speculative_config"
- '{"method":"eagle3","model":"vllm-ascend/MiniMax-M2.5-eagle-model-0318","num_speculative_tokens":3}'
- "--enable-expert-parallel"
- "--max-num-seqs"
- "128"
- "--max-num-batched-tokens"
- "16384"
- "--max-model-len"
- "196608"
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
benchmarks:
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 131072
batch_size: 32
baseline: 90
threshold: 10
bos_token_id: 200019
do_sample: true
eos_token_id: 200020
temperature: 1.0
top_p: 0.95
top_k: 40
transformers_version: 4.46.1
ignore_eos: false
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs2800
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 512
max_out_len: 1500
batch_size: 128
request_rate: 0
baseline: 1116
threshold: 0.97

85
tests/e2e/nightly/single_node/models/configs/Minimax_m2.7_w8a8_A3.yaml Normal file
View File

@@ -0,0 +1,85 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
SERVER_PORT: "DEFAULT_PORT"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1200"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
OMP_NUM_THREADS: "1"
LD_PRELOAD: "/usr/lib/aarch64-linux-gnu/libjemalloc.so.2:$LD_PRELOAD"
TASK_QUEUE_ENABLE: "1"
_server_cmd: &server_cmd
- "--port"
- "$SERVER_PORT"
- "--host"
- "0.0.0.0"
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "2"
- "--enable-expert-parallel"
- "--async-scheduling"
- "--max-num-seqs"
- "128"
- "--safetensors-load-strategy"
- 'prefetch'
- "--max-num-batched-tokens"
- "16384"
- "--trust-remote-code"
- "--quantization"
- "ascend"
- "--enable-auto-tool-choice"
- "--tool-call-parser"
- "minimax_m2"
- "--speculative-config"
- '{"method":"eagle3","model":"Eco-Tech/MiniMax-M2.7-eagle-model-short","num_speculative_tokens":3}'
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
- "--additional-config"
- '{"enable_cpu_binding": true, "enable_npugraph_ex": true, "enable_static_kernel": true,"enable_fused_mc2":true,"weight_nz_mode":true,"enable_flashcomm1":true}'
_benchmarks_3500: &benchmarks_3500
perf_50:
case_type: performance
dataset_path: vllm-ascend/GSM8K_in3500_bs4000_minimax
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 760
max_out_len: 1500
batch_size: 190
request_rate: 0
baseline: 4573.02
threshold: 0.97
perf_20:
case_type: performance
dataset_path: vllm-ascend/GSM8K_in3500_bs4000_minimax
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 192
max_out_len: 1500
batch_size: 48
request_rate: 0
baseline: 2229.147
threshold: 0.97
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "MiniMax-M2.7-3500"
model: "vllm-ascend/MiniMax-M2.7-w8a8-QuaRot"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--max-model-len"
- "70000"
- "--gpu-memory-utilization"
- "0.8"
- "--no-enable-prefix-caching"
benchmarks:
<<: *benchmarks_3500

78
tests/e2e/nightly/single_node/models/configs/Prefix-Cache-DeepSeek-R1-0528-W8A8.yaml Normal file
View File

@@ -0,0 +1,78 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "prefix-cache-deepseek-r1-0528-w8a8"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
envs:
OMP_NUM_THREADS: "10"
OMP_PROC_BIND: "false"
HCCL_BUFFSIZE: "1024"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ENGINE_READY_TIMEOUT_S: "7200"
server_cmd:
- "--quantization"
- "ascend"
- "--data-parallel-size"
- "2"
- "--tensor-parallel-size"
- "8"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--seed"
- "1024"
- "--max-model-len"
- "5200"
- "--max-num-batched-tokens"
- "4096"
- "--max-num-seqs"
- "16"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"enable_weight_nz_layout": true}'
- "--speculative-config"
- '{"num_speculative_tokens": 1, "method": "mtp"}'
test_content:
- "benchmark_comparisons"
benchmark_comparisons_args:
- metric: "TTFT"
baseline: "prefix0"
target: "prefix75"
ratio: 0.5
operator: "<"
benchmarks:
warm_up:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in1024-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 1000
baseline: 0
threshold: 0.97
prefix0:
case_type: performance
dataset_path: vllm-ascend/prefix0-in3500-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 18
baseline: 1
threshold: 0.97
prefix75:
case_type: performance
dataset_path: vllm-ascend/prefix75-in3500-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 18
baseline: 1
threshold: 0.97

70
tests/e2e/nightly/single_node/models/configs/Prefix-Cache-Qwen3-32B-Int8.yaml Normal file
View File

@@ -0,0 +1,70 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "prefix-cache-qwen3-32b-w8a8"
model: "vllm-ascend/Qwen3-32B-W8A8"
envs:
TASK_QUEUE_ENABLE: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--quantization"
- "ascend"
- "--reasoning-parser"
- "qwen3"
- "--tensor-parallel-size"
- "4"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "8192"
- "--max-num-batched-tokens"
- "8192"
- "--max-num-seqs"
- "256"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"enable_weight_nz_layout": true}'
test_content:
- "benchmark_comparisons"
benchmark_comparisons_args:
- metric: "TTFT"
baseline: "prefix0"
target: "prefix75"
ratio: 0.4
operator: "<"
benchmarks:
warm_up:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in1024-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 1000
baseline: 0
threshold: 0.97
prefix0:
case_type: performance
dataset_path: vllm-ascend/prefix0-in3500-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 48
baseline: 1
threshold: 0.97
prefix75:
case_type: performance
dataset_path: vllm-ascend/prefix75-in3500-bs210
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 210
max_out_len: 1
batch_size: 48
baseline: 1
threshold: 0.97

84
tests/e2e/nightly/single_node/models/configs/Qwen3-235B-A22B-W8A8.yaml Normal file
View File

@@ -0,0 +1,84 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
OMP_NUM_THREADS: "10"
OMP_PROC_BIND: "false"
HCCL_BUFFSIZE: "1024"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--data-parallel-size"
- "4"
- "--tensor-parallel-size"
- "4"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "40960"
- "--max-num-batched-tokens"
- "8192"
- "--max-num-seqs"
- "12"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 32
top_k: 20
baseline: 95
threshold: 10
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-235B-A22B-W8A8-full_graph"
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
<<: *benchmarks
- name: "Qwen3-235B-A22B-W8A8-piecewise"
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation-config"
- '{"cudagraph_mode": "PIECEWISE"}'
benchmarks:
<<: *benchmarks
- name: "Qwen3-235B-A22B-W8A8-EPLB"
model: "vllm-ascend/Qwen3-235B-A22B-W8A8"
envs:
<<: *envs
DYNAMIC_EPLB: "true"
server_cmd: *server_cmd
server_cmd_extra:
- "--additional-config"
- '{"eplb_config": {"dynamic_eplb": "true", "expert_heat_collection_interval": 600, "algorithm_execution_interval": 50, "num_redundant_experts": 16, "eplb_policy_type": 2}}'
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
<<: *benchmarks

41
tests/e2e/nightly/single_node/models/configs/Qwen3-30B-A3B-W4A8-llm-compressor.yaml Normal file
View File

@@ -0,0 +1,41 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-30B-A3B-W4A8-llm-compressor"
model: "vllm-ascend/Qwen3-30B-A3B-Instruct-2507-quantized.w4a8"
envs:
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "1"
HCCL_BUFFSIZE: "1024"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--no-enable-prefix-caching"
- "--tensor-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "40960"
- "--max-num-batched-tokens"
- "16384"
- "--max-num-seqs"
- "128"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.8"
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 95
threshold: 10

43
tests/e2e/nightly/single_node/models/configs/Qwen3-30B-A3B-W8A8.yaml Normal file
View File

@@ -0,0 +1,43 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-30B-A3B-W8A8-TP1"
model: "vllm-ascend/Qwen3-30B-A3B-W8A8"
envs:
OMP_PROC_BIND: "false"
OMP_NUM_THREADS: "10"
HCCL_BUFFSIZE: "1024"
HCCL_OP_EXPANSION_MODE: "AIV"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--no-enable-prefix-caching"
- "--tensor-parallel-size"
- "1"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "5600"
- "--max-num-batched-tokens"
- "16384"
- "--max-num-seqs"
- "100"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--compilation-config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 180
max_out_len: 1500
batch_size: 45
request_rate: 0
baseline: 1
threshold: 0.97

40
tests/e2e/nightly/single_node/models/configs/Qwen3-30B-QuaRot-eagle3.yaml Normal file
View File

@@ -0,0 +1,40 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-30B-QuaRot"
model: "vllm-ascend/Qwen3-30B-A3B-W8A8-QuaRot"
envs:
VLLM_WORKER_MULTIPROC_METHOD: "spawn"
SERVER_PORT: "DEFAULT_PORT"
HCCL_BUFFSIZE: "768"
server_cmd:
- "--enforce-eager"
- "--no-enable-prefix-caching"
- "--enable-expert-parallel"
- "--tensor-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "8192"
- "--trust-remote-code"
- "--distributed-executor-backend"
- "mp"
- "--gpu-memory-utilization"
- "0.9"
- "--speculative-config"
- '{"method": "eagle3", "model": "AngelSlim/Qwen3-a3B_eagle3", "num_speculative_tokens": 3}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 80
max_out_len: 1500
batch_size: 20
request_rate: 0
baseline: 1
threshold: 0.97

78
tests/e2e/nightly/single_node/models/configs/Qwen3-32B-Int8-A2.yaml Normal file
View File

@@ -0,0 +1,78 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
TASK_QUEUE_ENABLE: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--no-enable-prefix-caching"
- "--tensor-parallel-size"
- "4"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "40960"
- "--max-num-batched-tokens"
- "40960"
- "--block-size"
- "128"
- "--trust-remote-code"
- "--reasoning-parser"
- "qwen3"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"weight_prefetch_config":{"enabled":true}}'
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/aime2024
request_conf: vllm_api_general_chat
dataset_conf: aime2024/aime2024_gen_0_shot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 83.33
threshold: 10
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 288
max_out_len: 1500
batch_size: 72
baseline: 1
threshold: 0.97
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-32B-W8A8-aclgraph-a2"
model: "vllm-ascend/Qwen3-32B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[1,12,16,20,24,32,48,60,64,68,72,76,80]}'
benchmarks:
<<: *benchmarks
- name: "Qwen3-32B-W8A8-single-a2"
model: "vllm-ascend/Qwen3-32B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--enforce-eager"
benchmarks:

78
tests/e2e/nightly/single_node/models/configs/Qwen3-32B-Int8.yaml Normal file
View File

@@ -0,0 +1,78 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
TASK_QUEUE_ENABLE: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
VLLM_ASCEND_ENABLE_FLASHCOMM: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--no-enable-prefix-caching"
- "--tensor-parallel-size"
- "4"
- "--max-num-seqs"
- "80"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "40960"
- "--max-num-batched-tokens"
- "40960"
- "--block-size"
- "128"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"weight_prefetch_config":{"enabled":true}}'
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/gsm8k-lite
request_conf: vllm_api_general_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_noncot_chat_prompt
max_out_len: 10240
batch_size: 32
baseline: 96
threshold: 10
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 304
max_out_len: 1500
batch_size: 76
baseline: 1
threshold: 0.97
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-32B-W8A8-aclgraph-a3"
model: "vllm-ascend/Qwen3-32B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY","cudagraph_capture_sizes":[1,12,16,20,24,32,48,60,64,68,72,76,80]}'
benchmarks:
<<: *benchmarks
- name: "Qwen3-32B-W8A8-single-a3"
model: "vllm-ascend/Qwen3-32B-W8A8"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--enforce-eager"
benchmarks:

38
tests/e2e/nightly/single_node/models/configs/Qwen3-32B-QuaRot-eagle3.yaml Normal file
View File

@@ -0,0 +1,38 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-32B-QuaRot"
model: "vllm-ascend/Qwen3-32B-W8A8-QuaRot"
envs:
VLLM_WORKER_MULTIPROC_METHOD: "spawn"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--enforce-eager"
- "--no-enable-prefix-caching"
- "--tensor-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "8192"
- "--trust-remote-code"
- "--distributed-executor-backend"
- "mp"
- "--gpu-memory-utilization"
- "0.9"
- "--speculative-config"
- '{"method": "eagle3", "model": "RedHatAI/Qwen3-32B-speculator.eagle3", "num_speculative_tokens": 3}'
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs400
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 80
max_out_len: 1500
batch_size: 20
request_rate: 0
baseline: 1
threshold: 0.97

70
tests/e2e/nightly/single_node/models/configs/Qwen3-VL-235B-A22B-Instruct-W8A8.yaml Normal file
View File

@@ -0,0 +1,70 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
OMP_NUM_THREADS: "1"
OMP_PROC_BIND: "false"
TASK_QUEUE_ENABLE: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1536"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
VLLM_ASCEND_ENABLE_NZ: "2"
VLLM_ASCEND_BALANCE_SCHEDULING: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--no-enable-prefix-caching"
- "--mm-processor-cache-gb"
- "0"
- "--tensor-parallel-size"
- "4"
- "--data-parallel-size"
- "4"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "32768"
- "--max-num-batched-tokens"
- "16384"
- "--max-num-seqs"
- "32"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.92"
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/textvqa-lite
request_conf: vllm_api_stream_chat
dataset_conf: textvqa/textvqa_gen_base64
max_out_len: 2048
batch_size: 128
baseline: 83
temperature: 0
top_k: -1
top_p: 1
repetition_penalty: 1
threshold: 5
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-VL-235B-A22B-Instruct-W8A8"
model: "Eco-Tech/Qwen3-VL-235B-A22B-Instruct-w8a8-QuaRot"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation_config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY", "cudagraph_capture_sizes": [1,2,4,8,16,24,32]}'
benchmarks:
<<: *benchmarks

67
tests/e2e/nightly/single_node/models/configs/Qwen3-VL-32B-Instruct-W8A8.yaml Normal file
View File

@@ -0,0 +1,67 @@
# ==========================================
# Shared Configurations
# ==========================================
_envs: &envs
OMP_NUM_THREADS: "1"
OMP_PROC_BIND: "false"
TASK_QUEUE_ENABLE: "1"
HCCL_OP_EXPANSION_MODE: "AIV"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
VLLM_ASCEND_ENABLE_PREFETCH_MLP: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--quantization"
- "ascend"
- "--no-enable-prefix-caching"
- "--mm-processor-cache-gb"
- "0"
- "--tensor-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "20000"
- "--max-num-batched-tokens"
- "8192"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
_benchmarks: &benchmarks
acc:
case_type: accuracy
dataset_path: vllm-ascend/textvqa-lite
request_conf: vllm_api_stream_chat
dataset_conf: textvqa/textvqa_gen_base64
max_out_len: 2048
batch_size: 128
baseline: 80
temperature: 0
top_k: -1
top_p: 1
repetition_penalty: 1
threshold: 5
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3-VL-32B-Instruct-W8A8"
model: "Eco-Tech/Qwen3-VL-32B-Instruct-w8a8-QuaRot"
envs:
<<: *envs
server_cmd: *server_cmd
server_cmd_extra:
- "--compilation_config"
- '{"cudagraph_mode": "FULL_DECODE_ONLY", "cudagraph_capture_sizes": [1,12,16,20,24,32,48,64,68,72,76,80,128]}'
benchmarks:
<<: *benchmarks

79
tests/e2e/nightly/single_node/models/configs/Qwen3.5-122B-A10B-W8A8-A3.yaml Normal file
View File

@@ -0,0 +1,79 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
_envs: &envs
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1024"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
VLLM_ASCEND_ENABLE_FLASHCOMM1: "1"
SERVER_PORT: "DEFAULT_PORT"
_server_cmd: &server_cmd
- "--port"
- "$SERVER_PORT"
- "--max-model-len"
- "131072"
- "--max-num-batched-tokens"
- "16384"
- "--data-parallel-size"
- "1"
- "--tensor-parallel-size"
- "4"
- "--enable-expert-parallel"
- "--max-num-seqs"
- "128"
- "--gpu-memory-utilization"
- "0.9"
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
- "--trust-remote-code"
- "--allowed-local-media-path"
- "/"
- "--quantization"
- "ascend"
- "--mm-processor-cache-gb"
- "0"
- "--additional-config"
- '{"enable_cpu_binding": true, "enable_shared_expert_dp": true}'
test_cases:
- name: "Qwen3.5-122B-A10B-W8A8-A3"
model: "Eco-Tech/Qwen3.5-122B-A10B-w8a8-mtp"
envs:
<<: *envs
server_cmd: *server_cmd
benchmarks:
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 65536
batch_size: 32
baseline: 90
threshold: 10
thinking: true
temperature: 1.0
top_p: 0.95
top_k: 20
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
ignore_eos: false
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs8000-qwen3
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 320
max_out_len: 1500
batch_size: 80
request_rate: 0
baseline: 1
threshold: 0.97

57
tests/e2e/nightly/single_node/models/configs/Qwen3.5-27B-w8a8-A2.yaml Normal file
View File

@@ -0,0 +1,57 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3.5-27B-w8a8"
model: "Eco-Tech/Qwen3.5-27B-w8a8-mtp"
envs:
VLLM_USE_MODELSCOPE: "true"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_BUFFSIZE: "1024"
HCCL_OP_EXPANSION_MODE: "AIV"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
VLLM_ASCEND_ENABLE_PREFETCH_MLP: "1"
VLLM_ASCEND_ENABLE_DENSE_OPTIMIZE: "1"
VLLM_ASCEND_ENABLE_NZ: "1"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "2"
- "--port"
- "$SERVER_PORT"
- "--max-num-seqs"
- "128"
- "--quantization"
- "ascend"
- "--max-model-len"
- "262144"
- "--max-num-batched-tokens"
- "8192"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.95"
- "--additional-config"
- '{"enable_cpu_binding":true, "enable_weight_nz_layout":true}'
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3,"enforce_eager": true}'
- "--compilation-config"
- '{"cudagraph_mode":"FULL_DECODE_ONLY", "cudagraph_capture_sizes":[4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,104,108,112,116,120,124,128,132,136,140,144]}'
- "--mm-processor-cache-gb"
- "0"
- "--mm_processor_cache_type"
- "shm"
benchmarks:
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs8000-qwen3
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 128
max_out_len: 1500
batch_size: 32
request_rate: 0
baseline: 604
threshold: 0.97

73
tests/e2e/nightly/single_node/models/configs/Qwen3.5-27B-w8a8-A3.yaml Normal file
View File

@@ -0,0 +1,73 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3.5-27B-w8a8"
model: "Eco-Tech/Qwen3.5-27B-w8a8-mtp"
envs:
VLLM_USE_MODELSCOPE: "true"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_BUFFSIZE: "1024"
HCCL_OP_EXPANSION_MODE: "AIV"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "2"
- "--data-parallel-size"
- "1"
- "--port"
- "$SERVER_PORT"
- "--max-num-seqs"
- "128"
- "--quantization"
- "ascend"
- "--max-model-len"
- "196608"
- "--max-num-batched-tokens"
- "16384"
- "--trust-remote-code"
- "--async-scheduling"
- "--allowed-local-media-path"
- "/"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"enable_cpu_binding":true}'
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3,"enforce_eager": true}'
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,4,8,12,16,24,32,48,56,64,72,84,96,108,112,128,160,172,196,200,212,232,256,272,288,312,328,344,360,384,400,416,432,448,480,512], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--mm-processor-cache-gb"
- "0"
benchmarks:
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 65536
batch_size: 32
baseline: 90
threshold: 10
ignore_eos: false
thinking: true
temperature: 1.0
top_p: 0.95
top_k: 20
min_p: 0.0
presence_penalty: 1.5
repetition_penalty: 1.0
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs8000-qwen3
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 140
max_out_len: 1500
batch_size: 35
request_rate: 0
baseline: 610.22
threshold: 0.97

76
tests/e2e/nightly/single_node/models/configs/Qwen3.5-397B-A17B-W8A8-mtp-A3.yaml Normal file
View File

@@ -0,0 +1,76 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3.5-397B-A17B-w8a8-mtp"
model: "Eco-Tech/Qwen3.5-397B-A17B-w8a8-mtp"
envs:
VLLM_USE_MODELSCOPE: "true"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1024"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
VLLM_ENGINE_READY_TIMEOUT_S: "3000"
VLLM_RPC_TIMEOUT: "6000"
server_cmd:
- "--tensor-parallel-size"
- "16"
- "--data-parallel-size"
- "1"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-num-seqs"
- "128"
- "--quantization"
- "ascend"
- "--max-model-len"
- "133120"
- "--max-num-batched-tokens"
- "16384"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--additional-config"
- '{"enable_cpu_binding":true}'
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 5, "enforce_eager": true}'
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,6,12,18,24,30,36,42,48,54,72,78,84,90,96,102,108,144,192], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--allowed-local-media-path"
- "/"
- "--mm-processor-cache-gb"
- "0"
- "--safetensors-load-strategy"
- "lazy"
benchmarks:
acc_aime2025:
case_type: accuracy
dataset_path: vllm-ascend/aime2025
request_conf: vllm_api_general_chat
dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 90
threshold: 10
temperature: 0.6
top_p: 0.95
top_k: 20
min_p: 0.0
presence_penalty: 0.0
repetition_penalty: 1.0
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in131072-bs100-qwen3
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 8
max_out_len: 1024
batch_size: 2
request_rate: 0
baseline: 56.357
threshold: 0.97

74
tests/e2e/nightly/single_node/models/configs/Qwen3.5-397B-A17B-w4a8-mtp-A2.yaml Normal file
View File

@@ -0,0 +1,74 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3.5-397B-A17B-w4a8-mtp"
model: "Eco-Tech/Qwen3.5-397B-A17B-w4a8-mtp"
envs:
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1024"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
SERVER_PORT: "DEFAULT_PORT"
server_cmd:
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "1"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-num-seqs"
- "128"
- "--quantization"
- "ascend"
- "--max-model-len"
- "133120"
- "--max-num-batched-tokens"
- "16384"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--no-enable-prefix-caching"
- "--additional-config"
- '{"enable_cpu_binding":true,"multistream_overlap_shared_expert":true}'
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,4,8,12,16,20,24,28,32,36,40,44,48,52,56,60,64,68,72,76,80,84,88,92,96,100,108,112,128,160,172,196,200,212,232,256,260,288,320,360,400], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--allowed-local-media-path"
- "/"
- "--mm-processor-cache-gb"
- "0"
benchmarks:
acc_GPQA:
case_type: accuracy
dataset_path: vllm-ascend/gpqa
request_conf: vllm_api_general_chat
dataset_conf: gpqa/gpqa_gen_0_shot_cot_chat_prompt
max_out_len: 32768
batch_size: 32
baseline: 88.38
threshold: 5
# acc_aime2025:
# case_type: accuracy
# dataset_path: vllm-ascend/aime2025
# request_conf: vllm_api_general_chat
# dataset_conf: aime2025/aime2025_gen_0_shot_chat_prompt
# max_out_len: 72348
# batch_size: 32
# baseline: 93.33
# threshold: 5
perf:
case_type: performance
dataset_path: vllm-ascend/GSM8K-in3500-bs8000-qwen3
request_conf: vllm_api_stream_chat
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
num_prompts: 256
max_out_len: 1500
batch_size: 64
request_rate: 0
baseline: 1040
threshold: 0.97

62
tests/e2e/nightly/single_node/models/configs/Qwen3.5-397B-A17B-w8a8-mtp-longseq-A3.yaml Normal file
View File

@@ -0,0 +1,62 @@
# ==========================================
# ACTUAL TEST CASES
# ==========================================
test_cases:
- name: "Qwen3.5-397B-A17B-w8a8-mtp-longseq"
model: "Eco-Tech/Qwen3.5-397B-A17B-w8a8-mtp"
envs:
VLLM_USE_MODELSCOPE: "true"
PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
HCCL_OP_EXPANSION_MODE: "AIV"
HCCL_BUFFSIZE: "1024"
OMP_NUM_THREADS: "1"
TASK_QUEUE_ENABLE: "1"
SERVER_PORT: "DEFAULT_PORT"
VLLM_ASCEND_ENABLE_FUSED_MC2: "1"
VLLM_ENGINE_READY_TIMEOUT_S: "3000"
VLLM_RPC_TIMEOUT: "6000"
server_cmd:
- "--tensor-parallel-size"
- "8"
- "--data-parallel-size"
- "1"
- "--prefill-context-parallel-size"
- "2"
- "--decode-context-parallel-size"
- "4"
- "--enable-expert-parallel"
- "--port"
- "$SERVER_PORT"
- "--max-num-seqs"
- "32"
- "--quantization"
- "ascend"
- "--max-model-len"
- "133120"
- "--max-num-batched-tokens"
- "16384"
- "--trust-remote-code"
- "--gpu-memory-utilization"
- "0.9"
- "--speculative_config"
- '{"method": "qwen3_5_mtp", "num_speculative_tokens": 3, "enforce_eager": true}'
- "--compilation-config"
- '{"cudagraph_capture_sizes":[1,4,8,16,32,64], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--allowed-local-media-path"
- "/"
- "--mm-processor-cache-gb"
- "0"
- "--safetensors-load-strategy"
- "lazy"
benchmarks:
acc_GPQA:
case_type: accuracy
dataset_path: vllm-ascend/gpqa
request_conf: vllm_api_general_chat
dataset_conf: gpqa/gpqa_gen_0_shot_cot_chat_prompt
num_prompts: 50
max_out_len: 32768
batch_size: 32
baseline: 84
threshold: 8