Skip to content

Commit a8cfc19

Browse files
guxin108guxin108
authored
[CI] update weekly cases: Deepseekv3.2-w8a8-A3.yaml GLM5.1-W8A8-a3-weekly.yaml Qwen3.5-397B-A17B-w8a8-mtp-A3-weekly.yaml (vllm-project#11518)
### What this PR does / why we need it? we update weekly cases: Deepseekv3.2-w8a8-A3.yaml GLM5.1-W8A8-a3-weekly.yaml Qwen3.5-397B-A17B-w8a8-mtp-A3-weekly.yaml ### Does this PR introduce _any_ user-facing change? no ### How was this patch tested? run the cases weekly - vLLM version: v0.23.0 - vLLM main: vllm-project/vllm@1f486d9 --------- Signed-off-by: guxin108 <1252896542@qq.com>
1 parent 41b3a23 commit a8cfc19

5 files changed

Lines changed: 25 additions & 25 deletions

File tree

‎tests/e2e/weekly/multi_node/external_dp/config/GLM_5_1_PD_in32k_bs16-0.yaml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -224,7 +224,7 @@ templates:
224224
- --compilation-config
225225
- '{"cudagraph_mode":"FULL_DECODE_ONLY"}'
226226
- --additional-config
227-
- '{"fuse_muls_add":true,"recompute_scheduler_enable": true,"multistream_overlap_shared_expert": true,"eplb_config":{"dynamic_eplb":true,"expert_heat_collection_interval":2000,"algorithm_execution_interval":50, "eplb_policy_type":3, "num_redundant_experts":32}}'
227+
- '{"fuse_muls_add":true,"recompute_scheduler_enable": true,"multistream_overlap_shared_expert": true}'
228228
- --trust-remote-code
229229
- --max-num-seqs
230230
- "32"

‎tests/e2e/weekly/multi_node/external_dp/config/Kimi-K2.5-W4A8-16k-1k-TPOT50.yaml‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -57,7 +57,7 @@ env_common: &env_common
5757
TASK_QUEUE_ENABLE: "1"
5858
ASCEND_BUFFER_POOL: "4:8"
5959
ASCEND_RT_VISIBLE_DEVICES: "${VISIBLE_DEVICES}"
60-
60+
VLLM_USE_MODELSCOPE: "true"
6161
env_prefill: &env_prefill
6262
<<: *env_common
6363
HCCL_BUFFSIZE: "256"

‎tests/e2e/weekly/single_node/configs/DeepSeek-V3.2-W8A8_A3_weekly.yaml‎

Lines changed: 6 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -59,7 +59,7 @@ _benchmarks: &benchmarks
5959
max_out_len: 1024
6060
batch_size: 1
6161
request_rate: 0
62-
baseline: 1
62+
baseline: 26.10
6363
threshold: 0.97
6464
perf_16384_bs16:
6565
case_type: performance
@@ -70,7 +70,7 @@ _benchmarks: &benchmarks
7070
max_out_len: 1024
7171
batch_size: 4
7272
request_rate: 0.5
73-
baseline: 1
73+
baseline: 60.53
7474
threshold: 0.97
7575
perf_32768_bs4:
7676
case_type: performance
@@ -81,7 +81,7 @@ _benchmarks: &benchmarks
8181
max_out_len: 512
8282
batch_size: 1
8383
request_rate: 0
84-
baseline: 1
84+
baseline: 22.68
8585
threshold: 0.97
8686
perf_32768_bs8:
8787
case_type: performance
@@ -92,7 +92,7 @@ _benchmarks: &benchmarks
9292
max_out_len: 512
9393
batch_size: 2
9494
request_rate: 0.2
95-
baseline: 1
95+
baseline: 32.73
9696
threshold: 0.97
9797
perf_65536_bs4:
9898
case_type: performance
@@ -103,7 +103,7 @@ _benchmarks: &benchmarks
103103
max_out_len: 1024
104104
batch_size: 1
105105
request_rate: 0
106-
baseline: 1
106+
baseline: 22.77
107107
threshold: 0.97
108108
perf_65536_bs8:
109109
case_type: performance
@@ -114,7 +114,7 @@ _benchmarks: &benchmarks
114114
max_out_len: 1024
115115
batch_size: 2
116116
request_rate: 1
117-
baseline: 1
117+
baseline: 33.80
118118
threshold: 0.97
119119

120120
test_cases:

‎tests/e2e/weekly/single_node/configs/GLM-5_1-W8A8_A3_weekly.yaml‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -33,7 +33,7 @@ _server_cmd: &server_cmd
3333
- "--compilation-config"
3434
- '{"cudagraph_mode": "FULL_DECODE_ONLY"}'
3535
- "--speculative-config"
36-
- '{"num_speculative_tokens": 3, "method": "deepseek_mtp"}'
36+
- '{"num_speculative_tokens": 3, "method": "deepseek_mtp","enforce_eager": true}'
3737

3838

3939
_benchmarks: &benchmarks
@@ -81,7 +81,7 @@ _benchmarks_bs10: &benchmarks_bs10
8181
dataset_path: vllm-ascend/GSM8K_prefix90_in65536_bs100_glm
8282
request_conf: vllm_api_stream_chat
8383
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
84-
num_prompts: 10
84+
num_prompts: 20
8585
max_out_len: 1024
8686
batch_size: 5
8787
request_rate: 0

‎tests/e2e/weekly/single_node/configs/Qwen3.5-397B-A17B-W8A8-mtp-A3_weekly.yaml‎

Lines changed: 15 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -42,14 +42,14 @@ _server_cmd: &server_cmd
4242
_benchmarks: &benchmarks
4343
perf:
4444
case_type: performance
45-
dataset_path: vllm-ascend/GSM8K-in131072-bs100-qwen3
45+
dataset_path: vllm-ascend/GSM8K_prefix90_in131072_bs100_qwen
4646
request_conf: vllm_api_stream_chat
4747
dataset_conf: gsm8k/gsm8k_gen_0_shot_cot_str_perf
48-
num_prompts: 8
48+
num_prompts: 16
4949
max_out_len: 1024
50-
batch_size: 2
50+
batch_size: 4
5151
request_rate: 0
52-
baseline: 1
52+
baseline: 165.23
5353
threshold: 0.97
5454

5555

@@ -63,7 +63,7 @@ _benchmarks_bs144: &benchmarks_bs144
6363
max_out_len: 1024
6464
batch_size: 36
6565
request_rate: 0
66-
baseline: 1
66+
baseline: 630.88
6767
threshold: 0.97
6868

6969
_benchmarks_bs36: &benchmarks_bs36
@@ -76,7 +76,7 @@ _benchmarks_bs36: &benchmarks_bs36
7676
max_out_len: 1024
7777
batch_size: 9
7878
request_rate: 0
79-
baseline: 1
79+
baseline: 336.80
8080
threshold: 0.97
8181

8282
_benchmarks_bs48: &benchmarks_bs48
@@ -89,7 +89,7 @@ _benchmarks_bs48: &benchmarks_bs48
8989
max_out_len: 512
9090
batch_size: 12
9191
request_rate: 0
92-
baseline: 1
92+
baseline: 180.60
9393
threshold: 0.97
9494

9595
_benchmarks_bs16: &benchmarks_bs16
@@ -102,7 +102,7 @@ _benchmarks_bs16: &benchmarks_bs16
102102
max_out_len: 512
103103
batch_size: 4
104104
request_rate: 0
105-
baseline: 1
105+
baseline: 118.17
106106
threshold: 0.97
107107

108108

@@ -116,7 +116,7 @@ _benchmarks_bs160: &benchmarks_bs160
116116
max_out_len: 512
117117
batch_size: 40
118118
request_rate: 0
119-
baseline: 1
119+
baseline: 689.71
120120
threshold: 0.97
121121

122122
_benchmarks_bs32: &benchmarks_bs32
@@ -129,7 +129,7 @@ _benchmarks_bs32: &benchmarks_bs32
129129
max_out_len: 512
130130
batch_size: 8
131131
request_rate: 0
132-
baseline: 1
132+
baseline: 303.07
133133
threshold: 0.97
134134

135135
_benchmarks_bs8: &benchmarks_bs8
@@ -142,7 +142,7 @@ _benchmarks_bs8: &benchmarks_bs8
142142
max_out_len: 1024
143143
batch_size: 2
144144
request_rate: 0
145-
baseline: 1
145+
baseline: 93.10
146146
threshold: 0.97
147147

148148
_benchmarks_bs32_in65536: &benchmarks_bs32_in65536
@@ -155,7 +155,7 @@ _benchmarks_bs32_in65536: &benchmarks_bs32_in65536
155155
max_out_len: 1024
156156
batch_size: 8
157157
request_rate: 0
158-
baseline: 1
158+
baseline: 160.55
159159
threshold: 0.97
160160

161161
_benchmarks_bs136: &benchmarks_bs136
@@ -168,7 +168,7 @@ _benchmarks_bs136: &benchmarks_bs136
168168
max_out_len: 1024
169169
batch_size: 34
170170
request_rate: 0
171-
baseline: 1
171+
baseline: 618.66
172172
threshold: 0.97
173173

174174
_benchmarks_bs16_in65536: &benchmarks_bs16_in65536
@@ -181,7 +181,7 @@ _benchmarks_bs16_in65536: &benchmarks_bs16_in65536
181181
max_out_len: 1024
182182
batch_size: 4
183183
request_rate: 0
184-
baseline: 1
184+
baseline: 198.87
185185
threshold: 0.97
186186

187187
_benchmarks_bs80: &benchmarks_bs80
@@ -194,7 +194,7 @@ _benchmarks_bs80: &benchmarks_bs80
194194
max_out_len: 1024
195195
batch_size: 20
196196
request_rate: 0
197-
baseline: 1
197+
baseline: 335.95
198198
threshold: 0.97
199199

200200
# ==========================================

0 commit comments

Comments
 (0)