feat: Faster ML - #76
Merged
Merged
Conversation
The crate ran a model in process through candle and MKL. It also needed a target-cpu=native build flag, which tied the binary to the build host. A llama.cpp server takes the work in the commit that follows.
The workspace already carried the french feature of lingua, but the binary enabled neither it nor the other two. The language actor now reads them.
The reorder_arrays option also sorts the commit_parsers array of cliff.toml. git-cliff takes the first rule that matches, so the sort moves the group rules in front of the skip rules and changes which commits it skips.
Every consolidation example collapsed a modifier and a noun into that noun, as in "steel sword" to "swords". None collapsed different sibling nouns into a parent, so one mod listed machine guns, autocannons, cannons, missile weapons, electromagnetic weapons, laser weapons and tracking weapons as seven separate features. Three examples of the right shape plus one rule take that item from 23 features to 12. The examples alone made it worse, at 21. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01NJmogiWB2qU1XHsS8BtVmo
The actor posts each prompt to the chat completions endpoint and holds the answer to a JSON schema. The config gains ml_extraction.url next to the enabled flag, which was a plain bool before.
Contributor
Svelte Check ResultsFound 0 errors and 2 warnings (2 total) with the files in this PR. ui/src/routes/item/[item]/Property.svelteui/src/routes/item/[item]/Property.svelte:26:24Warn: This reference only captures the initial value of `property`. Did you mean to reference it inside a derived instead?
https://svelte.dev/e/state_referenced_locally
let voteState = $state(property.vote_state);ui/src/routes/item/[item]/Property.svelte:27:26Warn: This reference only captures the initial value of `property`. Did you mean to reference it inside a derived instead?
https://svelte.dev/e/state_referenced_locally
let upvoteCount = $state(property.upvote_count);Last Updated: 15th September at 22:39 (9550c84) |
The align and align-middle classes do nothing to a grid item.
The catch-all type collected items that the other types already named. The new type marks an item whose author used a model for part of the content.
Each task held its own copy of the request builder and its own schema structs. One table now names the two tasks, and each schema type carries a default, so a third task costs one entry. The actor returns LlamaError instead of Whatever, and the health check gives up after five minutes rather than looping on a dead server.
disconsented
force-pushed
the
feat/faster-ml
branch
from
September 15, 2026 22:38
28a0e15 to
ed8fbc8
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The existing candle backend is very slow w/ the mistral model running at about .7T/s.
Swapping over to a Gemma4 E4B model & a llama.cpp server brings us up to about 20T/s on the same hardware.