ggml-org/whisper.cpp

SetMaxTokensPerSegment in go binding causes the segements to end prematurely

Open

#984 创建于 2023年6月3日

在 GitHub 查看
 (1 评论) (1 反应) (0 负责人)C++ (49,693 star) (5,535 fork)batch import
buggood first issuehelp wanted

描述

When I use context.SetMaxTokensPerSegment and context.NextSegment, it generates incomplete segments. e.g. when tested using the jfk.wav sample, it gives the following output:

wspr_test.go:33: {Text:And so my fellow Americans ask Start:0s End:30s}

when this value is not set, it gives the correct output:

wspr_test.go:33: {Text:And so my fellow Americans ask not what your country can do for you Start:0s End:8s}
wspr_test.go:33: {Text:ask what you can do for your country. Start:8s End:11s}

test case: https://github.com/JackKCWong/go-whisper-test/blob/main/pkg/whisper/wspr_test.go#L29

贡献者指南

技术栈
gocpp
领域
backend
议题类型
bug
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
3
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
1-3 hours
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
stale
清晰度议题是否清楚说明期望改动、验收标准和下一步。
clear
前置要求
Gobasic understanding of whisper.cpp
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
50
研究方向
Look at the Go binding implementation, specifically the SetMaxTokensPerSegment and NextSegment functions in the whisper package. The test case in the issue provides a reproduction. Investigate how the max tokens per segment parameter is passed to the C++ code and whether there is an off by one or incorrect handling that causes segments to end prematurely. Compare with the behavior when the parameter is not set.