Modularize fused experts and integrate PPLX kernels (#15956)

This commit is contained in:
bnellnm
2025-05-14 13:11:54 -07:00
committed by GitHub
parent 418d2f8bfb
commit f9c069c85e
42 changed files with 3830 additions and 660 deletions
+2 -1
View File
@@ -390,7 +390,8 @@ class CPUWorker(LocalOrDistributedWorkerBase):
ensure_model_parallel_initialized(
parallel_config.tensor_parallel_size,
parallel_config.pipeline_parallel_size)
parallel_config.pipeline_parallel_size,
parallel_config.enable_expert_parallel)
def get_cache_block_size_bytes(self) -> int:
"""Return the size in bytes of a single KV cache block.