biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core][Model] PrithviMAE Enablement on vLLM v1 engine (#20577)

Browse Source 
Signed-off-by: Christian Pinto <christian.pinto@ibm.com>
This commit is contained in:
Christian Pinto
2025-07-23 19:00:23 +01:00
committed by GitHub
parent 316b1bf706
commit 8560a5b258
15 changed files with 704 additions and 238 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/engine/llm_engine.py
View File

@@ -238,14 +238,14 @@ class LLMEngine:
self.log_stats = log_stats
self.use_cached_outputs = use_cached_outputs
if not self.model_config.skip_tokenizer_init:
self.tokenizer = self._init_tokenizer()
self.detokenizer = Detokenizer(self.tokenizer)
tokenizer_group = self.get_tokenizer_group()
else:
if self.model_config.skip_tokenizer_init:
self.tokenizer = None
self.detokenizer = None
tokenizer_group = None
else:
self.tokenizer = self._init_tokenizer()
self.detokenizer = Detokenizer(self.tokenizer)
tokenizer_group = self.get_tokenizer_group()
# Ensure that the function doesn't contain a reference to self,
# to avoid engine GC issues