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
[Bugfix][CI] Machete kernels: deterministic ordering for more cache hits (#23055)
Signed-off-by: Andy Lo <andy@mistral.ai>
Andy Lo committed
Aug 18, 2025 at 07:10 UTC
b2fd0b81e065c677ceebecb9a0e1ee6f226b7cec
1 file changed
+6
-3
csrc/quantization/machete/generate.py
+6
-3
@@ -349,9 +349,12 @@ def to_cute_constant(value: list[int]):
349
350
351
def unique_schedules(impl_configs: list[ImplConfig]):
352
- return list(
353
- set(sch for impl_config in impl_configs
354
- for sch in impl_config.schedules))
352
+ # Use dict over set for deterministic ordering
353
+ return list({
354
+ sch: None
355
+ for impl_config in impl_configs
356
+ for sch in impl_config.schedules
357
+ }.keys())
358
359
360
def unsigned_type_with_bitwidth(num_bits):