EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[CI] Add multi-node test case for a2 (#3805)

Browse Source 
### What this PR does / why we need it?
This patch add multi-node test case for a2
### Does this PR introduce _any_ user-facing change?

### How was this patch tested?

- vLLM version: v0.11.0rc3
- vLLM main:
c9461e05a4

---------

Signed-off-by: wangli <wangli858794774@gmail.com>
This commit is contained in:
Li Wang
2025-10-27 23:10:17 +08:00
committed by GitHub
parent 9030106a14
commit f846bd20e4
5 changed files with 97 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
tests/e2e/nightly/multi_node/config/models/DeepSeep-R1-W8A8-A2.yaml Normal file
View File

@@ -0,0 +1,57 @@
test_name: "test DeepSeek-R1-W8A8 on A2"
model: "vllm-ascend/DeepSeek-R1-0528-W8A8"
num_nodes: 2
npu_per_node: 8
env_common:
VLLM_USE_MODELSCOPE: true
HCCL_BUFFSIZE: 1024
SERVER_PORT: 8080
OMP_PROC_BIND: false
OMP_NUM_THREADS: 10
deployment:
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--host 0.0.0.0
--port $SERVER_PORT
--data-parallel-size 4
--data-parallel-size-local 2
--data-parallel-address $LOCAL_IP
--data-parallel-rpc-port 13399
--no-enable-prefix-caching
--max-num-seqs 16
--tensor-parallel-size 4
--max-model-len 36864
--max-num-batched-tokens 6000
--enable-expert-parallel
--trust-remote-code
--quantization ascend
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
--additional-config '{"ascend_scheduler_config":{"enabled":false},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
-
server_cmd: >
vllm serve vllm-ascend/DeepSeek-R1-0528-W8A8
--headless
--data-parallel-size 4
--data-parallel-rpc-port 13399
--data-parallel-size-local 2
--data-parallel-start-rank 2
--data-parallel-address $MASTER_IP
--no-enable-prefix-caching
--max-num-seqs 16
--tensor-parallel-size 4
--max-model-len 36864
--max-num-batched-tokens 6000
--enable-expert-parallel
--trust-remote-code
--quantization ascend
--gpu-memory-utilization 0.9
--enforce-eager
--speculative-config '{"num_speculative_tokens": 1, "method":"deepseek_mtp"}' \
--additional-config '{"ascend_scheduler_config":{"enabled":false},"chunked_prefill_for_mla":true,"enable_weight_nz_layout":true}'
benchmarks:

8
tests/e2e/nightly/multi_node/scripts/lws.yaml.jinja2
View File

@@ -37,11 +37,11 @@ spec:
bash /root/.cache/tests/run.sh
resources:
limits:
huawei.com/ascend-1980: "16"
huawei.com/ascend-1980: {{ npu_per_node | default("16") }}
memory: 512Gi
ephemeral-storage: 100Gi
requests:
huawei.com/ascend-1980: "16"
huawei.com/ascend-1980: {{ npu_per_node | default("16") }}
ephemeral-storage: 100Gi
cpu: 125
ports:
@@ -95,11 +95,11 @@ spec:
bash /root/.cache/tests/run.sh
resources:
limits:
huawei.com/ascend-1980: "16"
huawei.com/ascend-1980: {{ npu_per_node | default("16") }}
memory: 512Gi
ephemeral-storage: 100Gi
requests:
huawei.com/ascend-1980: "16"
huawei.com/ascend-1980: {{ npu_per_node | default("16") }}
ephemeral-storage: 100Gi
cpu: 125
volumeMounts: