Skip to content

fix deepseek #5359

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
xiegegege wants to merge 4 commits into
base: develop
Choose a base branch
from
Open

fix deepseek #5359

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
21c5ca8
fix deepseek
xiegegege Dec 3, 2025
d121ac6
Merge branch 'develop' into develop
EmmonsCurse Dec 3, 2025
19ac723
add fastdeploy benchmark,paddletest-155
xiegegege Dec 9, 2025
957e86a
Merge branch 'PaddlePaddle:develop' into develop
xiegegege Dec 11, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
16 changes: 16 additions & 0 deletions benchmarks/yaml/eb45-32k-wint4-tp4_decode_router.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,16 @@
max_model_len: 32768
max_num_seqs: 256
gpu_memory_utilization: 0.9
kv_cache_ratio: 0.8
tensor_parallel_size: 4
cache_queue_port: 55663
enable_chunked_prefill: True
splitwise_role: decode
engine_worker_queue_port: 6678
cache_transfer_protocol: "rdma,ipc"
rdma_comm_ports: "7671,7672,7673,7674"
pd_comm_port: "2334"
max_num_batched_tokens: 384
max_num_partial_prefills: 3
max_long_partial_prefills: 3
quantization: wint4
13 changes: 13 additions & 0 deletions benchmarks/yaml/eb45-32k-wint4-tp4_prefill_router.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,13 @@
max_model_len: 32768
max_num_seqs: 16
gpu_memory_utilization: 0.9
kv_cache_ratio: 0.9
tensor_parallel_size: 4
splitwise_role: prefill
enable_prefix_caching: True
cache_queue_port: 55664
engine_worker_queue_port: 6677
cache_transfer_protocol: "rdma,ipc"
rdma_comm_ports: "7675,7676,7677,7678"
pd_comm_port: "2333"
quantization: wint4
2 changes: 1 addition & 1 deletion fastdeploy/worker/gpu_worker.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -141,7 +141,7 @@ def determine_available_memory(self) -> int:
paddle_allocated_mem_after_run = paddle.device.cuda.max_memory_allocated(local_rank)

model_block_memory_used = self.cal_theortical_kvcache()
paddle_peak_increase = paddle_allocated_mem_after_run - paddle_allocated_mem_before_run
paddle_peak_increase = paddle_reserved_mem_after_run - paddle_allocated_mem_before_run

paddle.device.cuda.empty_cache()

Expand Down
Loading