xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-07-08 14:57:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

skip detokenize

Browse Source
This commit is contained in:
Woosuk Kwon 2025-10-23 05:40:20 +00:00
parent a1cac48477
commit cb439737db
1 changed files with 4 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/benchmarks/datasets.py
View File

@ -75,7 +75,7 @@ class SampleRequest:
Represents a single inference request for benchmarking. Represents a single inference request for benchmarking.
""" """
prompt: str | list[str] prompt: str | list[str] | list[int]
prompt_len: int prompt_len: int
expected_output_len: int expected_output_len: int
multi_modal_data: MultiModalDataDict | dict | list[dict] | None = None multi_modal_data: MultiModalDataDict | dict | list[dict] | None = None
@ -402,8 +402,9 @@ def gen_prompt_decode_to_target_len(
remain_num_try = max_retry remain_num_try = max_retry
token_mismatch = 0 token_mismatch = 0
while True: while True:
prompt = tokenizer.decode(token_sequence) # prompt = tokenizer.decode(token_sequence)
token_sequence = tokenizer.encode(prompt, add_special_tokens=add_special_tokens) # token_sequence = tokenizer.encode(prompt, add_special_tokens=add_special_tokens)
prompt = token_sequence
if remain_num_try <= 0: if remain_num_try <= 0:
if len(token_sequence) != target_token_len: if len(token_sequence) != target_token_len:
token_mismatch = len(token_sequence) - target_token_len token_mismatch = len(token_sequence) - target_token_len