Free's top bar shows a quota figure instead of the model in use #364

Closed
opened 2026-11-03 09:31:53 -05:00 by cmoriarty · 1 comment
Owner

What happens
On a Free run the top bar says lowest: judge tier 99% left today · refills 19:00. It names the tier with the least left, often not the one in use, and the operator can't tell what it means (#194 already tried to make it clearer). The bar never says which model is working, though the Inference panel lists it per step.

Beside it, the run's green dot and Free 1 running read like the run's health. The dot only means the backend answers, and the count is agent steps on that backend, from any run.

Fix

  • The cell shows the model serving now, e.g. MODEL reflection/beam-free, updating as the router changes model. With several steps running: one model plus +1 more, the rest on hover.
  • Paused until … stays for spent quota. The percentages stay in the Inference panel's quota section only.
  • Make the dot and count say what they are, or drop them.

Check first
osf.free.served is written when a request completes, so the name may lag a switch, and a step retrying may keep showing the last model that answered. If the gap is long, show that it is waiting for a model.

Done when

  • A browser test: the cell follows a change of model and shows Paused until … while quota is spent.
  • The top bar has no quota percentage.
**What happens** On a Free run the top bar says `lowest: judge tier 99% left today · refills 19:00`. It names the tier with the least left, often not the one in use, and the operator can't tell what it means (#194 already tried to make it clearer). The bar never says which model is working, though the Inference panel lists it per step. Beside it, the run's green dot and `Free 1 running` read like the run's health. The dot only means the backend answers, and the count is agent steps on that backend, from any run. **Fix** - The cell shows the model serving now, e.g. `MODEL reflection/beam-free`, updating as the router changes model. With several steps running: one model plus `+1 more`, the rest on hover. - `Paused until …` stays for spent quota. The percentages stay in the Inference panel's quota section only. - Make the dot and count say what they are, or drop them. **Check first** `osf.free.served` is written when a request completes, so the name may lag a switch, and a step retrying may keep showing the last model that answered. If the gap is long, show that it is waiting for a model. **Done when** - A browser test: the cell follows a change of model and shows `Paused until …` while quota is spent. - The top bar has no quota percentage.
Author
Owner

Shipped in PR 468 (main ddfd8a93), archived as batch-2026-11-09b (PR 469).

What changed. The top bar's Free cell is captioned Free model and names the model serving now, from the slot holders: hetzner/Qwen3.8-27B, with +1 more when steps are on several and each model's step count in the hover. waiting for a model while a step has no answer yet, none serving with nothing at work. Paused until … stays for spent quota, with its hover. No percentage on the bar; the shares stay on the Inference page. The backend cell's hover says its count (2 of 6 slots in use across every run) is the backend's, not the run's.

Done when, checked.

  • Browser test free.spec.ts: on free_serving the cell reads hetzner/Qwen3.8-27B +1 more with both models in the hover; on the new free_paused fixture it reads Paused until … with data-paused; the top bar contains no % in either state. Unit tests in free.test.ts cover the change of model, the waiting and the empty cases.
  • Screenshots of the real console on the test server: Free model / hetzner/Qwen3.8-27B +1 more, and Free quota / Paused until 16:29.

The name lags a switch by one answer (it is the model that last answered), as the issue's "Check first" expected; a step with no answer yet says so.

Shipped in PR 468 (main `ddfd8a93`), archived as `batch-2026-11-09b` (PR 469). **What changed.** The top bar's Free cell is captioned `Free model` and names the model serving now, from the slot holders: `hetzner/Qwen3.8-27B`, with `+1 more` when steps are on several and each model's step count in the hover. `waiting for a model` while a step has no answer yet, `none serving` with nothing at work. `Paused until …` stays for spent quota, with its hover. No percentage on the bar; the shares stay on the Inference page. The backend cell's hover says its count (`2 of 6 slots in use across every run`) is the backend's, not the run's. **Done when, checked.** - Browser test `free.spec.ts`: on `free_serving` the cell reads `hetzner/Qwen3.8-27B +1 more` with both models in the hover; on the new `free_paused` fixture it reads `Paused until …` with `data-paused`; the top bar contains no `%` in either state. Unit tests in `free.test.ts` cover the change of model, the waiting and the empty cases. - Screenshots of the real console on the test server: `Free model / hetzner/Qwen3.8-27B +1 more`, and `Free quota / Paused until 16:29`. The name lags a switch by one answer (it is the model that last answered), as the issue's "Check first" expected; a step with no answer yet says so.
Sign in to join this conversation.
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
cmoriarty/braid#364
No description provided.