A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 09:00 UTC
[CI] Bump `num_speculative_tokens` to 3 in nightly DeepSeek tests (#35882)
Signed-off-by: Matthew Bonanni <mbonanni@redhat.com>
Matthew Bonanni committed
Mar 3, 2026 at 12:26 UTC
8e1fd5baf0ff272936618bf578533d9aa7080a27
4 files changed
+4
-4
tests/evals/gsm8k/configs/DeepSeek-R1-DP.yaml
+1
-1
@@ -8,4 +8,4 @@ server_args: >-
8
--max-model-len 4096
9
--data-parallel-size 8
10
--enable-expert-parallel
11
- --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
11
+ --speculative-config '{"method":"mtp","num_speculative_tokens":3}'
tests/evals/gsm8k/configs/DeepSeek-R1-TP.yaml
+1
-1
@@ -8,4 +8,4 @@ server_args: >-
8
--max-model-len 4096
9
--tensor-parallel-size 8
10
--enable-expert-parallel
11
- --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
11
+ --speculative-config '{"method":"mtp","num_speculative_tokens":3}'
tests/evals/gsm8k/configs/DeepSeek-V3.2-DP.yaml
+1
-1
@@ -8,4 +8,4 @@ server_args: >-
8
--max-model-len 4096
9
--data-parallel-size 8
10
--enable-expert-parallel
11
- --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
11
+ --speculative-config '{"method":"mtp","num_speculative_tokens":3}'
tests/evals/gsm8k/configs/DeepSeek-V3.2-TP.yaml
+1
-1
@@ -8,4 +8,4 @@ server_args: >-
8
--max-model-len 4096
9
--tensor-parallel-size 8
10
--enable-expert-parallel
11
- --speculative-config '{"method":"mtp","num_speculative_tokens":1}'
11
+ --speculative-config '{"method":"mtp","num_speculative_tokens":3}'