Elgato_dark/void
1
0
Fork 0
mirror of https://github.com/voideditor/void synced 2026-05-23 01:18:25 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Update default MLX model to use quantized weights for improved performance and adjust setup instructions accordingly

Browse source
This commit is contained in:
Jérôme Commaret 2026-05-20 15:28:41 +02:00
parent a5ed7b6af7
commit d666c68241
2 changed files with 3 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/vs/workbench/contrib/void/browser/react/src/void-settings-tsx/Settings.tsx
View file

@ -890,7 +890,7 @@ export const OllamaSetupInstructions = ({ sayWeAutoDetect }: { sayWeAutoDetect?:
export const MlxSetupInstructions = () => {
return <div className='prose-p:my-0 prose-ol:list-decimal prose-p:py-0 prose-ol:my-0 prose-ol:py-0 text-void-fg-3 text-sm list-decimal select-text mb-4'>
<div><ChatMarkdownRender string={`MLX (one model at a time)`} chatMessageLocation={undefined} /></div>
<div className='pl-6'><ChatMarkdownRender string={`1. Void can run \`pip install mlx-lm\` and start \`mlx_lm.server\` for you (toggle in Settings → Models). Default model: \`mlx-community/Qwen2.5-Coder-1.5B-Instruct\` on port 8080.`} chatMessageLocation={undefined} /></div>
<div className='pl-6'><ChatMarkdownRender string={`1. Void can run \`pip install mlx-lm\` and start \`mlx_lm.server\` for you (toggle in Settings → Models). Default model: \`mlx-community/Qwen2.5-Coder-1.5B-Instruct-4bit\` on port 8080.`} chatMessageLocation={undefined} /></div>
<div className='pl-6'><ChatMarkdownRender string={`2. Void shows only **one** autodetected entry: the model currently loaded by the server.`} chatMessageLocation={undefined} /></div>
<div className='pl-6'><ChatMarkdownRender string={`3. **Switch models**: stop the server, run \`mlx_lm.server --model <other-hf-repo>\`, then click **Refresh** next to MLX in Settings → Models.`} chatMessageLocation={undefined} /></div>
<div className='pl-6'><ChatMarkdownRender string={`4. **Second model in parallel**: run another server on a different port (e.g. \`mlx_lm.server --model … --port 8081\`), then at the bottom of the model list use **Add Model** → **OpenAI-Compatible** with \`http://127.0.0.1:8081/v1\`, or add the model id manually under MLX if the endpoint exposes multiple ids.`} chatMessageLocation={undefined} /></div>

3
src/vs/workbench/contrib/void/common/mlxTypes.ts
View file

@ -8,7 +8,8 @@ import { createDecorator } from '../../../../platform/instantiation/common/insta
export const MLX_DEFAULT_ENDPOINT = 'http://127.0.0.1:8080';
export const MLX_DEFAULT_PORT = 8080;
/** Small default model; first start may download weights from Hugging Face. */
export const MLX_DEFAULT_MODEL = 'mlx-community/Qwen2.5-Coder-1.5B-Instruct';
/** Quantized MLX weights on Hugging Face; the unsuffixed repo id does not exist. */
export const MLX_DEFAULT_MODEL = 'mlx-community/Qwen2.5-Coder-1.5B-Instruct-4bit';
export type MlxEnsureAction = 'already-running' | 'started' | 'installed-and-started';