Include tokens from prompt phase in counter_generation_tokens (#2802)
This commit is contained in:
@@ -872,6 +872,9 @@ class LLMEngine:
|
||||
num_prompt_tokens = sum(
|
||||
len(seq_group.prompt_token_ids)
|
||||
for seq_group in scheduler_outputs.scheduled_seq_groups)
|
||||
num_generation_tokens = sum(
|
||||
seq_group.num_seqs()
|
||||
for seq_group in scheduler_outputs.scheduled_seq_groups)
|
||||
else:
|
||||
num_generation_tokens = scheduler_outputs.num_batched_tokens
|
||||
|
||||
|
||||
Reference in New Issue
Block a user