apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2025-10-06 17:17:14 +08:00
Code Issues Actions 2 Packages Projects Releases Wiki Activity

[FDConfig]Remove max_num_batched_tokens/max_num_seqs in parallel config (#4116)

Browse Source 
* remove max_num_batched_tokens in parallel config

* remove max_num_seqs

* update test case

* fix test

* fix

---------

Co-authored-by: Jiang-Jia-Jun <163579578+Jiang-Jia-Jun@users.noreply.github.com>
This commit is contained in:
YuanRisheng
2025-09-17 10:43:35 +08:00
committed by GitHub
parent c01a756912
commit 2e9e53ff7e
30 changed files with 169 additions and 131 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
fastdeploy/worker/xpu_model_runner.py
View File

@@ -353,12 +353,12 @@ class XPUModelRunner(ModelRunnerBase):
self.graph_opt_level = self.graph_opt_config.graph_opt_level
self.use_cudagraph = False
self.sot_warmup_sizes = self.graph_opt_config.sot_warmup_sizes
self.input_ids = paddle.zeros(self.parallel_config.max_num_seqs, dtype="int32")
self.input_ids = paddle.zeros(self.scheduler_config.max_num_seqs, dtype="int32")
# Initialize share inputs
self._init_share_inputs(self.fd_config.parallel_config.max_num_seqs)
self._init_share_inputs(self.fd_config.scheduler_config.max_num_seqs)
self.infer_seed_increment = paddle.full(
shape=[self.parallel_config.max_num_seqs, 1],
shape=[self.scheduler_config.max_num_seqs, 1],
fill_value=4,
dtype="int64",
).cpu()
@@ -812,7 +812,7 @@ class XPUModelRunner(ModelRunnerBase):
start_time = time.perf_counter()
for batch_size in self.sot_warmup_sizes:
self._dummy_run(
num_tokens=self.parallel_config.max_num_batched_tokens,
num_tokens=self.scheduler_config.max_num_batched_tokens,
batch_size=batch_size,
)
logger.info(f"SOT warmup the model with the batch size:{batch_size}")
@@ -987,8 +987,8 @@ class XPUModelRunner(ModelRunnerBase):
"""Execute a forward pass with dummy inputs to profile the memory usage of the model."""
self._dummy_run(
num_tokens=int(self.parallel_config.max_num_batched_tokens),
batch_size=min(self.parallel_config.max_num_seqs, 1),
num_tokens=int(self.scheduler_config.max_num_batched_tokens),
batch_size=min(self.scheduler_config.max_num_seqs, 1),
)
def clear_block_table(self) -> None: