Skip to content

Cache: WIP (5/6) Browse content-addressed inference caches - #125

Closed
ErlisLushtaku wants to merge 4 commits into
cache-on-118/04-runnersfrom
cache-on-118/05-browser
Closed

Cache: WIP (5/6) Browse content-addressed inference caches#125
ErlisLushtaku wants to merge 4 commits into
cache-on-118/04-runnersfrom
cache-on-118/05-browser

Conversation

@ErlisLushtaku

Copy link
Copy Markdown
Collaborator

Description

Note

WIP: This is a replacement cache stack on top of #118. It is not final and may still change before review.

Ports the cache browser from #71 onto the content-addressed store.

  • Discovers completion and judgement cells under store_root.
  • Filters by task, model, judge, and language, and joins context with completions for rendering.
  • Adds the terminal browser and notebook widgets under the optional browser extra.

This is stacked on #124.

Look up and write cache rows at the shared inference boundary so full hits skip backend materialization.
Resolve provider descriptors and input modes so hosted and local backends share the same content-addressed cache path.
Wire store_root through generation and judging paths and drop the legacy dataframe cache helpers.
Keep task, model, judge, and language filtering plus the terminal and notebook UIs on the new store layout.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant