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

[CI]Fix the error caused by layer_sharding in dsv32 (#8719)

Browse Source 
### What this PR does / why we need it?

This PR fixes the error in DSV32 mixed deployment caused by enabling
layer_sharding.

- Currently, mixed deployment no longer supports the enabling of
layer_sharding. Therefore, it has been removed from the service-oriented
configuration.
- The error "RPC call to sample_tokens timed out" occurred because the
dshm size limit was set too small. Therefore, it was increased to 512
Gi.

### Does this PR introduce _any_ user-facing change?

no.

### How was this patch tested?

The nightly test has passed.

Signed-off-by: wyh145 <1987244901@qq.com>
This commit is contained in:
Nagisa125
2026-04-30 10:35:48 +08:00
committed by GitHub
parent 483f7d8188
commit 600bf80c6d
2 changed files with 0 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/e2e/nightly/multi_node/config/DeepSeek-V3_2-W8A8-A3-dual-nodes.yaml
View File

@@ -39,7 +39,6 @@ deployment:
--trust-remote-code --trust-remote-code
--speculative-config '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}' --speculative-config '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}'
--compilation-config '{"cudagraph_capture_sizes": [8, 16, 24, 32, 40, 48], "cudagraph_mode": "FULL_DECODE_ONLY"}' --compilation-config '{"cudagraph_capture_sizes": [8, 16, 24, 32, 40, 48], "cudagraph_mode": "FULL_DECODE_ONLY"}'
--additional-config '{"layer_sharding": ["q_b_proj", "o_proj"]}'
--tokenizer-mode deepseek_v32 --tokenizer-mode deepseek_v32
--reasoning-parser deepseek_v3 --reasoning-parser deepseek_v3
@@ -64,7 +63,6 @@ deployment:
--trust-remote-code --trust-remote-code
--speculative-config '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}' --speculative-config '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}'
--compilation-config '{"cudagraph_capture_sizes": [8, 16, 24, 32, 40, 48], "cudagraph_mode": "FULL_DECODE_ONLY"}' --compilation-config '{"cudagraph_capture_sizes": [8, 16, 24, 32, 40, 48], "cudagraph_mode": "FULL_DECODE_ONLY"}'
--additional-config '{"layer_sharding": ["q_b_proj", "o_proj"]}'
--tokenizer-mode deepseek_v32 --tokenizer-mode deepseek_v32
--reasoning-parser deepseek_v3 --reasoning-parser deepseek_v3
benchmarks: benchmarks:

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

@@ -38,8 +38,6 @@ test_cases:
- '{"cudagraph_capture_sizes":[4, 8, 16, 20, 24, 28, 32], "cudagraph_mode":"FULL_DECODE_ONLY"}' - '{"cudagraph_capture_sizes":[4, 8, 16, 20, 24, 28, 32], "cudagraph_mode":"FULL_DECODE_ONLY"}'
- "--speculative-config" - "--speculative-config"
- '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}' - '{"num_speculative_tokens": 3, "method":"deepseek_mtp"}'
- "--additional-config"
- '{"layer_sharding": ["q_b_proj", "o_proj"]}'
- "--reasoning-parser" - "--reasoning-parser"
- "deepseek_v3" - "deepseek_v3"
- "--tokenizer_mode" - "--tokenizer_mode"