Skip to content

Commit fc08aa7

Browse files
yuxingcyxyuxingcyx
authored
Merge branch 'main' into triton-cyx-new
2 parents e2bc5c9 + 56f0182 commit fc08aa7

File tree

8 files changed

+19
-21
lines changed

8 files changed

+19
-21
lines changed

‎docs/source/tutorials/DeepSeek-V3.1.md‎

Lines changed: 3 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -105,7 +105,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
105105
export TP_SOCKET_IFNAME=$nic_name
106106
export HCCL_SOCKET_IFNAME=$nic_name
107107
export OMP_PROC_BIND=false
108-
export OMP_NUM_THREADS=100
108+
export OMP_NUM_THREADS=10
109109
export VLLM_USE_V1=1
110110
export HCCL_BUFFSIZE=200
111111
export VLLM_ASCEND_ENABLE_MLAPO=1
@@ -164,7 +164,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
164164
export TP_SOCKET_IFNAME=$nic_name
165165
export HCCL_SOCKET_IFNAME=$nic_name
166166
export OMP_PROC_BIND=false
167-
export OMP_NUM_THREADS=100
167+
export OMP_NUM_THREADS=10
168168
export VLLM_USE_V1=1
169169
export HCCL_BUFFSIZE=200
170170
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
@@ -220,7 +220,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
220220
export TP_SOCKET_IFNAME=$nic_name
221221
export HCCL_SOCKET_IFNAME=$nic_name
222222
export OMP_PROC_BIND=false
223-
export OMP_NUM_THREADS=100
223+
export OMP_NUM_THREADS=10
224224
export HCCL_BUFFSIZE=200
225225
export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
226226
export VLLM_ASCEND_ENABLE_MLAPO=1

‎docs/source/tutorials/DeepSeek-V3.2-Exp.md‎

Lines changed: 4 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -206,7 +206,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
206206
export TP_SOCKET_IFNAME=$nic_name
207207
export HCCL_SOCKET_IFNAME=$nic_name
208208
export OMP_PROC_BIND=false
209-
export OMP_NUM_THREADS=100
209+
export OMP_NUM_THREADS=10
210210
export HCCL_BUFFSIZE=1024
211211

212212
vllm serve /root/.cache/Modelers_Park/DeepSeek-V3.2-Exp \
@@ -248,7 +248,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
248248
export TP_SOCKET_IFNAME=$nic_name
249249
export HCCL_SOCKET_IFNAME=$nic_name
250250
export OMP_PROC_BIND=false
251-
export OMP_NUM_THREADS=100
251+
export OMP_NUM_THREADS=10
252252
export HCCL_BUFFSIZE=1024
253253

254254
vllm serve /root/.cache/Modelers_Park/DeepSeek-V3.2-Exp \
@@ -295,9 +295,8 @@ export GLOO_SOCKET_IFNAME=$nic_name
295295
export TP_SOCKET_IFNAME=$nic_name
296296
export HCCL_SOCKET_IFNAME=$nic_name
297297
export OMP_PROC_BIND=false
298-
export OMP_NUM_THREADS=100
298+
export OMP_NUM_THREADS=10
299299
export HCCL_BUFFSIZE=1024
300-
export HCCL_OP_EXPANSION_MODE="AIV"
301300
export PYTORCH_NPU_ALLOC_CONF="expandable_segments:True"
302301

303302
vllm serve vllm-ascend/DeepSeek-V3.2-Exp-W8A8 \
@@ -340,9 +339,8 @@ export GLOO_SOCKET_IFNAME=$nic_name
340339
export TP_SOCKET_IFNAME=$nic_name
341340
export HCCL_SOCKET_IFNAME=$nic_name
342341
export OMP_PROC_BIND=false
343-
export OMP_NUM_THREADS=100
342+
export OMP_NUM_THREADS=10
344343
export HCCL_BUFFSIZE=1024
345-
export HCCL_OP_EXPANSION_MODE="AIV"
346344
export PYTORCH_NPU_ALLOC_CONF="expandable_segments:True"
347345

348346
vllm serve vllm-ascend/DeepSeek-V3.2-Exp-W8A8 \

‎docs/source/tutorials/multi_node.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -114,7 +114,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
114114
export TP_SOCKET_IFNAME=$nic_name
115115
export HCCL_SOCKET_IFNAME=$nic_name
116116
export OMP_PROC_BIND=false
117-
export OMP_NUM_THREADS=100
117+
export OMP_NUM_THREADS=10
118118
export HCCL_BUFFSIZE=1024
119119

120120
# The w8a8 weight can be obtained from https://www.modelscope.cn/models/vllm-ascend/DeepSeek-V3.1-W8A8
@@ -159,7 +159,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
159159
export TP_SOCKET_IFNAME=$nic_name
160160
export HCCL_SOCKET_IFNAME=$nic_name
161161
export OMP_PROC_BIND=false
162-
export OMP_NUM_THREADS=100
162+
export OMP_NUM_THREADS=10
163163
export HCCL_BUFFSIZE=1024
164164

165165
vllm serve vllm-ascend/DeepSeek-V3.1-W8A8 \

‎docs/source/tutorials/multi_node_kimi.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -69,7 +69,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
6969
export TP_SOCKET_IFNAME=$nic_name
7070
export HCCL_SOCKET_IFNAME=$nic_name
7171
export OMP_PROC_BIND=false
72-
export OMP_NUM_THREADS=100
72+
export OMP_NUM_THREADS=10
7373
export HCCL_BUFFSIZE=1024
7474

7575
# The w8a8 weight can be obtained from https://www.modelscope.cn/models/vllm-ascend/Kimi-K2-Instruct-W8A8
@@ -114,7 +114,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
114114
export TP_SOCKET_IFNAME=$nic_name
115115
export HCCL_SOCKET_IFNAME=$nic_name
116116
export OMP_PROC_BIND=false
117-
export OMP_NUM_THREADS=100
117+
export OMP_NUM_THREADS=10
118118
export HCCL_BUFFSIZE=1024
119119

120120
vllm serve /home/cache/weights/Kimi-K2-Instruct-W8A8 \

‎docs/source/tutorials/multi_node_qwen3vl.md‎

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -69,7 +69,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
6969
export TP_SOCKET_IFNAME=$nic_name
7070
export HCCL_SOCKET_IFNAME=$nic_name
7171
export OMP_PROC_BIND=false
72-
export OMP_NUM_THREADS=100
72+
export OMP_NUM_THREADS=10
7373
export HCCL_BUFFSIZE=1024
7474

7575
vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct \
@@ -110,7 +110,7 @@ export GLOO_SOCKET_IFNAME=$nic_name
110110
export TP_SOCKET_IFNAME=$nic_name
111111
export HCCL_SOCKET_IFNAME=$nic_name
112112
export OMP_PROC_BIND=false
113-
export OMP_NUM_THREADS=100
113+
export OMP_NUM_THREADS=10
114114
export HCCL_BUFFSIZE=1024
115115

116116
vllm serve Qwen/Qwen3-VL-235B-A22B-Instruct \

‎examples/disaggregated_prefill_v1/README.md‎

Lines changed: 4 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -40,7 +40,7 @@ export TP_SOCKET_IFNAME="eth0"
4040
export HCCL_SOCKET_IFNAME="eth0"
4141
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
4242
export OMP_PROC_BIND=false
43-
export OMP_NUM_THREADS=100
43+
export OMP_NUM_THREADS=10
4444
export VLLM_ASCEND_LLMDD_RPC_PORT=5559
4545

4646
vllm serve /models/deepseek_r1_w8a8 \
@@ -80,7 +80,7 @@ export TP_SOCKET_IFNAME="eth0"
8080
export HCCL_SOCKET_IFNAME="eth0"
8181
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
8282
export OMP_PROC_BIND=false
83-
export OMP_NUM_THREADS=100
83+
export OMP_NUM_THREADS=10
8484
export VLLM_ASCEND_LLMDD_RPC_PORT=5659
8585

8686
vllm serve /models/deepseek_r1_w8a8 \
@@ -123,7 +123,7 @@ export TP_SOCKET_IFNAME="eth0"
123123
export HCCL_SOCKET_IFNAME="eth0"
124124
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
125125
export OMP_PROC_BIND=false
126-
export OMP_NUM_THREADS=100
126+
export OMP_NUM_THREADS=10
127127
export VLLM_ASCEND_LLMDD_RPC_PORT=5759
128128

129129
vllm serve /models/deepseek_r1_w8a8 \
@@ -164,7 +164,7 @@ export TP_SOCKET_IFNAME="eth0"
164164
export HCCL_SOCKET_IFNAME="eth0"
165165
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=/vllm-workspace/vllm-ascend/examples/disaggregated_prefill_v1/ranktable.json
166166
export OMP_PROC_BIND=false
167-
export OMP_NUM_THREADS=100
167+
export OMP_NUM_THREADS=10
168168
export VLLM_ASCEND_LLMDD_RPC_PORT=5859
169169

170170
vllm serve /models/deepseek_r1_w8a8 \

‎examples/disaggregated_prefill_v1/run_server.sh‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@ export HCCL_SOCKET_IFNAME="enp48s3u1u1"
55
export DISAGGREGATED_PREFILL_RANK_TABLE_PATH=path-to-rank-table
66

77
export OMP_PROC_BIND=false
8-
export OMP_NUM_THREADS=100
8+
export OMP_NUM_THREADS=10
99

1010
vllm serve model_path \
1111
--host 0.0.0.0 \

‎examples/run_dp_server.sh‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -5,7 +5,7 @@ export TP_SOCKET_IFNAME="eth0"
55
export HCCL_SOCKET_IFNAME="eth0"
66

77
export OMP_PROC_BIND=false
8-
export OMP_NUM_THREADS=100
8+
export OMP_NUM_THREADS=10
99

1010
export VLLM_USE_MODELSCOPE=true
1111

0 commit comments

Comments
 (0)