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 12, 11:00 UTC
[Bugfix] Fix pre-commit (#29601)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Nov 27, 2025 at 18:23 UTC
e6d4f3c254a215e75b4d76d531176e242fe62a1f
2 files changed
+9
-7
tests/v1/ec_connector/integration/test_epd_correctness.py
+2
-3
@@ -237,9 +237,8 @@ def main():
237
238
for i, prompt_data in enumerate(test_prompts):
239
print(
240
- f"\nRunning prompt {i + 1}/{len(test_prompts)}: {
241
- prompt_data['description']
242
- }"
240
+ f"\nRunning prompt {i + 1}/{len(test_prompts)}: "
241
+ f"{prompt_data['description']}"
242
)
243
244
output_str = run_chat_completion(
vllm/v1/worker/gpu/spec_decode/eagle_cudagraph.py
+7
-4
@@ -35,13 +35,16 @@ class EagleCudaGraphManager:
35
self.compilation_config = vllm_config.compilation_config
36
assert self.compilation_config is not None
37
38
+ cudagraph_mode: CUDAGraphMode
39
if self.compilation_config.cudagraph_mode is None:
39
- self.cudagraph_mode = CUDAGraphMode.NONE
40
+ cudagraph_mode = CUDAGraphMode.NONE
41
else:
41
- self.cudagraph_mode = self.compilation_config.cudagraph_mode
42
- if self.cudagraph_mode == CUDAGraphMode.FULL:
42
+ cudagraph_mode = self.compilation_config.cudagraph_mode
43
+ if cudagraph_mode == CUDAGraphMode.FULL:
44
# NOTE(woosuk): For Eagle, we only use CUDA graphs for decode.
44
- self.cudagraph_mode = CUDAGraphMode.FULL_DECODE_ONLY
45
+ cudagraph_mode = CUDAGraphMode.FULL_DECODE_ONLY
46
+
47
+ self.cudagraph_mode = cudagraph_mode
48
49
self.cudagraph_sizes = get_cudagraph_sizes(
50
self.compilation_config.cudagraph_capture_sizes,