EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Tiny refactor ModelConfig.from_server_args (#5219)

Browse Source
This commit is contained in:
fzyzcjy
2025-05-08 16:02:43 +08:00
committed by GitHub
parent 3b2680a44d
commit b6cf3532b5
6 changed files with 23 additions and 53 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
python/sglang/srt/managers/scheduler.py
View File

@@ -455,17 +455,7 @@ class Scheduler(
def init_tokenizer(self):
server_args = self.server_args
self.model_config = ModelConfig(
server_args.model_path,
trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
context_length=server_args.context_length,
model_override_args=server_args.json_model_override_args,
is_embedding=server_args.is_embedding,
enable_multimodal=server_args.enable_multimodal,
dtype=server_args.dtype,
quantization=server_args.quantization,
)
self.model_config = ModelConfig.from_server_args(server_args)
self.is_generation = self.model_config.is_generation
if server_args.skip_tokenizer_init:

12
python/sglang/srt/managers/tokenizer_manager.py
View File

@@ -165,17 +165,7 @@ class TokenizerManager:
# Read model args
self.model_path = server_args.model_path
self.served_model_name = server_args.served_model_name
self.model_config = ModelConfig(
server_args.model_path,
trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
context_length=server_args.context_length,
model_override_args=server_args.json_model_override_args,
is_embedding=server_args.is_embedding,
enable_multimodal=server_args.enable_multimodal,
dtype=server_args.dtype,
quantization=server_args.quantization,
)
self.model_config = ModelConfig.from_server_args(server_args)
self.is_generation = self.model_config.is_generation
self.is_image_gen = self.model_config.is_image_gen

13
python/sglang/srt/managers/tp_worker.py
View File

@@ -65,20 +65,13 @@ class TpModelWorker:
self.pp_rank = pp_rank
# Init model and tokenizer
self.model_config = ModelConfig(
(
self.model_config = ModelConfig.from_server_args(
server_args,
model_path=(
server_args.model_path
if not is_draft_worker
else server_args.speculative_draft_model_path
),
trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
context_length=server_args.context_length,
model_override_args=server_args.json_model_override_args,
is_embedding=server_args.is_embedding,
enable_multimodal=server_args.enable_multimodal,
dtype=server_args.dtype,
quantization=server_args.quantization,
is_draft_model=is_draft_worker,
)