xinyin025/LangBot
1
0
Fork 0
mirror of https://github.com/langbot-app/LangBot.git synced 2026-07-22 04:16:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

feat: 不再预先计算前文token数而是在报错时提醒用户重置

Browse Source
This commit is contained in:
RockChinQ
2024-03-12 16:04:11 +08:00
parent a398c6f311
commit 1d963d0f0c
16 changed files with 25 additions and 144 deletions
Download Patch File Download Diff File Expand all files Collapse all files
pkg/pipeline/preproc/preproc.py
-22
View File
@@ -51,28 +51,6 @@ class PreProcessor(stage.PipelineStage):
query.prompt.messages = event_ctx.event.default_prompt
query.messages = event_ctx.event.prompt
# 根据模型max_tokens剪裁
max_tokens = min(query.use_model.max_tokens, self.ap.pipeline_cfg.data['submit-messages-tokens'])
test_messages = query.prompt.messages + query.messages + [query.user_message]
while await query.use_model.tokenizer.count_token(test_messages, query.use_model) > max_tokens:
# 前文都pop完了，还是大于max_tokens，由于prompt和user_messages不能删减，报错
if len(query.prompt.messages) == 0:
return entities.StageProcessResult(
result_type=entities.ResultType.INTERRUPT,
new_query=query,
user_notice='输入内容过长，请减少情景预设或者输入内容长度',
console_notice='输入内容过长，请减少情景预设或者输入内容长度，或者增大配置文件中的 submit-messages-tokens 项（但不能超过所用模型最大tokens数）'
)
query.messages.pop(0) # pop第一个肯定是role=user的
# 继续pop到第二个role=user前一个
while len(query.messages) > 0 and query.messages[0].role != 'user':
query.messages.pop(0)
test_messages = query.prompt.messages + query.messages + [query.user_message]
return entities.StageProcessResult(
result_type=entities.ResultType.CONTINUE,
new_query=query