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

fix missing revision arg when loading tokenizer (#2982)

Browse Source
This commit is contained in:
giorgiopiatti-dfinity
2025-01-19 20:36:07 +01:00
committed by GitHub
parent a69cb5cff7
commit 8b6a4486ec
5 changed files with 8 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
python/sglang/srt/managers/detokenizer_manager.py
View File

@@ -71,6 +71,7 @@ class DetokenizerManager:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
self.decode_status = LimitedCapacityDict() self.decode_status = LimitedCapacityDict()

2
python/sglang/srt/managers/scheduler.py
View File

@@ -206,6 +206,7 @@ class Scheduler:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
self.tokenizer = self.processor.tokenizer self.tokenizer = self.processor.tokenizer
else: else:
@@ -213,6 +214,7 @@ class Scheduler:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
# Check whether overlap can be enabled # Check whether overlap can be enabled

2
python/sglang/srt/managers/tokenizer_manager.py
View File

@@ -158,6 +158,7 @@ class TokenizerManager:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
self.tokenizer = self.processor.tokenizer self.tokenizer = self.processor.tokenizer
os.environ["TOKENIZERS_PARALLELISM"] = "false" os.environ["TOKENIZERS_PARALLELISM"] = "false"
@@ -171,6 +172,7 @@ class TokenizerManager:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
# Store states # Store states

2
python/sglang/srt/managers/tp_worker.py
View File

@@ -83,6 +83,7 @@ class TpModelWorker:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
self.tokenizer = self.processor.tokenizer self.tokenizer = self.processor.tokenizer
else: else:
@@ -90,6 +91,7 @@ class TpModelWorker:
server_args.tokenizer_path, server_args.tokenizer_path,
tokenizer_mode=server_args.tokenizer_mode, tokenizer_mode=server_args.tokenizer_mode,
trust_remote_code=server_args.trust_remote_code, trust_remote_code=server_args.trust_remote_code,
revision=server_args.revision,
) )
self.device = self.model_runner.device self.device = self.model_runner.device

1
python/sglang/srt/server.py
View File

@@ -1027,6 +1027,7 @@ class Runtime:
self.server_args.tokenizer_path, self.server_args.tokenizer_path,
tokenizer_mode=self.server_args.tokenizer_mode, tokenizer_mode=self.server_args.tokenizer_mode,
trust_remote_code=self.server_args.trust_remote_code, trust_remote_code=self.server_args.trust_remote_code,
revision=self.server_args.revision,
) )
async def async_generate( async def async_generate(