EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

add disable cuda graph unit test for eagle 2 (#3412)

Browse Source
This commit is contained in:
Yineng Zhang
2025-02-09 08:02:56 +08:00
committed by GitHub
parent fad315cb8e
commit 6222e1c228
1 changed files with 61 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

36
test/srt/test_eagle_infer.py
View File

@@ -30,16 +30,34 @@ class TestEAGLEEngine(unittest.TestCase):
ref_output = ref_engine.generate(prompt, sampling_params)["text"]
ref_engine.shutdown()
# Test cases with different configurations
configs = [
# Original config
{
"model_path": DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
"speculative_draft_model_path": DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
"speculative_algorithm": "EAGLE",
"speculative_num_steps": 5,
"speculative_eagle_topk": 8,
"speculative_num_draft_tokens": 64,
"mem_fraction_static": 0.7,
},
# Config with CUDA graph disabled
{
"model_path": DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
"speculative_draft_model_path": DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
"speculative_algorithm": "EAGLE",
"speculative_num_steps": 5,
"speculative_eagle_topk": 8,
"speculative_num_draft_tokens": 64,
"mem_fraction_static": 0.7,
"disable_cuda_graph": True,
},
]
for config in configs:
# Launch EAGLE engine
engine = sgl.Engine(
model_path=DEFAULT_EAGLE_TARGET_MODEL_FOR_TEST,
speculative_draft_model_path=DEFAULT_EAGLE_DRAFT_MODEL_FOR_TEST,
speculative_algorithm="EAGLE",
speculative_num_steps=5,
speculative_eagle_topk=8,
speculative_num_draft_tokens=64,
mem_fraction_static=0.7,
)
engine = sgl.Engine(**config)
# Case 1: Test the output of EAGLE engine is the same as normal engine
out1 = engine.generate(prompt, sampling_params)["text"]