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 14, 05:30 UTC
[CI][Fix] deterministic seed for flaky CI runs on structured outputs (#24380)
Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
Aaron Pham committed
Sep 6, 2025 at 23:10 UTC
e67597545b0162e8f2e0d4f8186fb2e546c04128
1 file changed
+3
-3
tests/v1/entrypoints/llm/test_struct_output_generate.py
+3
-3
@@ -41,9 +41,8 @@ EAGLE_SPEC_CONFIG = {
41
PARAMS_MODELS_BACKENDS_TOKENIZER_MODE = [
42
("mistralai/Ministral-8B-Instruct-2410", "xgrammar", "auto", None),
43
("mistralai/Ministral-8B-Instruct-2410", "guidance", "auto", None),
44
- #FIXME: This test is flaky on CI thus disabled
45
- #("mistralai/Ministral-8B-Instruct-2410", "lm-format-enforcer", "auto",
46
- # None),
44
+ ("mistralai/Ministral-8B-Instruct-2410", "lm-format-enforcer", "auto",
45
+ None),
46
("mistralai/Ministral-8B-Instruct-2410", "xgrammar", "mistral", None),
47
("Qwen/Qwen2.5-1.5B-Instruct", "xgrammar", "auto", None),
48
("Qwen/Qwen2.5-1.5B-Instruct", "lm-format-enforcer", "auto", None),
@@ -123,6 +122,7 @@ def test_structured_output(
122
guided_decoding_backend=guided_decoding_backend,
123
guided_decoding_disable_any_whitespace=(guided_decoding_backend
124
in {"xgrammar", "guidance"}),
125
+ seed=120,
126
tokenizer_mode=tokenizer_mode,
127
speculative_config=speculative_config)
128