apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-22 16:19:37 +08:00
Code Issues Actions 5 Packages Projects Releases Wiki Activity

fix DP&&TP (#3872)
Some checks failed
CE Compile Job / ce_job_pre_check (push) Has been cancelled
Details
CE Compile Job / print_ce_job_pre_check_outputs (push) Has been cancelled
Details
CE Compile Job / FD-Clone-Linux (push) Has been cancelled
Details
CE Compile Job / Show Code Archive Output (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8090 (push) Has been cancelled
Details
CE Compile Job / BUILD_SM8689 (push) Has been cancelled
Details
CE Compile Job / CE_UPLOAD (push) Has been cancelled
Details

Browse Source
This commit is contained in:
lizhenyun01
2025-09-04 14:38:26 +08:00
committed by GitHub
parent b8d0f1c081
commit d40d3a5a4f
2 changed files with 3 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
fastdeploy/config.py
View File

@@ -1284,10 +1284,6 @@ class FDConfig:
f"be less than or equal to max_num_partial_prefills: {self.max_num_partial_prefills}"
)
assert self.splitwise_role in ["mixed", "prefill", "decode"]
# TODO(@wufeisheng): TP and EP need to be supported simultaneously.
assert (self.parallel_config.tensor_parallel_size == 1 and self.parallel_config.expert_parallel_size >= 1) or (
self.parallel_config.tensor_parallel_size >= 1 and self.parallel_config.expert_parallel_size == 1
), "TP and EP cannot be enabled at the same time"
if not self.cache_config.enable_chunked_prefill:
if not envs.ENABLE_V1_KVCACHE_SCHEDULER:

4
fastdeploy/worker/worker_process.py
View File

@@ -655,7 +655,9 @@ def initialize_fd_config(args, ranks: int = 1, local_rank: int = 0) -> FDConfig:
num_experts_per_rank = num_experts // parallel_config.expert_parallel_size
num_experts_start_offset = expert_parallel_rank * num_experts_per_rank
max_chips_per_node = 16 if current_platform.is_iluvatar() else 8
parallel_config.local_data_parallel_id = expert_parallel_rank % max_chips_per_node
parallel_config.local_data_parallel_id = parallel_config.data_parallel_rank % (
max_chips_per_node // parallel_config.tensor_parallel_size
)
parallel_config.expert_parallel_rank = expert_parallel_rank
parallel_config.num_experts_per_rank = num_experts_per_rank