settings should have descriptive hovertext when mouse is over the title of the setting #21

New Issue

hugo · 2026-07-25T20:43:22Z

hugo commented

2026-07-25 20:43:22 +00:00

each setting should have a short paragraph showing in a tooltip when the mouse hovers over the title of the setting, so that the user can actually understand what that setting is. The paragraph should actually be descriptive and not just restate the title of the setting with more words.

hugo added the enhancement label 2026-07-25 20:43:22 +00:00

hugo commented

2026-07-25 22:05:06 +00:00

Implemented in 9bd648d.

Added a hint(title, description) helper in src/app/view.rs wrapping the title in an iced tooltip (bottom-positioned, 320px wide, panel styling); preference_input_row now takes the description as its second argument, so every labelled input gets one by construction.
All 30 settings are covered: model, idle timeout, endpoint/port/CORS, context, max tokens, system prompt, temperature, top-p, min-p, seed, reasoning, CPU threads, GPU power, prefill chunk, quality kernels, warm weights, MTP tokens/margin, GLM MTP and its timing log, DSpark and its threshold/target-only mode, SSD streaming, cold start, expert cache, resident layers, preload count, steering file/FFN/attention scales, simulated memory, expert profile output.
Each text says what the knob actually does and what it costs (memory, throughput, accuracy) rather than restating the label — e.g. min-p is described as a cheaper tail cut than top-p that keeps working at high temperature.

Section headings and the existing footnote/effective-settings lines were left as they are.

Implemented in 9bd648d. - Added a `hint(title, description)` helper in `src/app/view.rs` wrapping the title in an iced tooltip (bottom-positioned, 320px wide, panel styling); `preference_input_row` now takes the description as its second argument, so every labelled input gets one by construction. - All 30 settings are covered: model, idle timeout, endpoint/port/CORS, context, max tokens, system prompt, temperature, top-p, min-p, seed, reasoning, CPU threads, GPU power, prefill chunk, quality kernels, warm weights, MTP tokens/margin, GLM MTP and its timing log, DSpark and its threshold/target-only mode, SSD streaming, cold start, expert cache, resident layers, preload count, steering file/FFN/attention scales, simulated memory, expert profile output. - Each text says what the knob actually does and what it costs (memory, throughput, accuracy) rather than restating the label — e.g. min-p is described as a cheaper tail cut than top-p that keeps working at high temperature. Section headings and the existing footnote/effective-settings lines were left as they are.

hugo closed this issue

2026-07-25 22:05:06 +00:00

hugo referenced this issue

2026-07-26 06:16:39 +00:00

cache management for kvcache disc usage #3

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: hugo/DS4Server#21