[Model Runner V2] Minor refactoring for penalties (#34662)

Signed-off-by: Woosuk Kwon <woosuk@inferact.ai>
This commit is contained in:
Woosuk Kwon
2026-02-16 21:43:00 -08:00
committed by GitHub
parent 9752da9d9c
commit d00df624f3
4 changed files with 94 additions and 75 deletions
+2 -8
View File
@@ -155,9 +155,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
max_num_reqs=self.max_num_reqs,
vocab_size=self.vocab_size,
device=self.device,
all_token_ids=self.req_states.all_token_ids.gpu,
prompt_len=self.req_states.prompt_len.gpu,
total_len=self.req_states.total_len.gpu,
req_states=self.req_states,
logprobs_mode=self.model_config.logprobs_mode,
num_speculative_tokens=self.num_speculative_steps + 1,
)
@@ -528,11 +526,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
if scheduler_output.scheduled_new_reqs:
self.req_states.apply_staged_writes()
self.sampler.apply_staged_writes(
self.req_states.all_token_ids.gpu,
self.req_states.prefill_len.np,
self.req_states.prompt_len.np,
)
self.sampler.apply_staged_writes()
if self.uses_mrope:
self.mrope_states.apply_staged_writes()