[Feature] Get Token IDs with Engine.generate() (#2636)

Co-authored-by: Chayenne <zhaochen20@outlook.com>
This commit is contained in:
Shi Shuai
2024-12-29 20:28:27 +00:00
committed by GitHub
parent b085e06b01
commit 35bdb48557
7 changed files with 92 additions and 2 deletions

View File

@@ -323,7 +323,9 @@ class BatchTokenIDOut:
decoded_texts: List[str]
decode_ids: List[int]
read_offsets: List[int]
# Only used when `--skip-tokenizer-init`
# Only used when --return-token-ids` is set
origin_input_ids: Optional[List[int]]
# Only used when `--skip-tokenizer-init` or `--return-token-ids` is set
output_ids: Optional[List[int]]
# Detokenization configs
skip_special_tokens: List[bool]
@@ -354,6 +356,10 @@ class BatchStrOut:
# The output decoded strings
output_strs: List[str]
# The token ids
origin_input_ids: Optional[List[int]]
output_ids: Optional[List[int]]
# Token counts
prompt_tokens: List[int]
completion_tokens: List[int]