pullrequests Search Results · language:Dune language:Python language:JavaScript language:PHP language:JavaScript
Filter by
162M results
Motivation
- Reduce the runtime surface of core/pipeline_orchestrator.py by moving deterministic packaging of already-computed
retrieval-stop and ordinary-continuation telemetry into a bounded helper ...
codex
notes/ held working scratch — superseded spike code (the FLUX.2 unpack helper + its synthetic-shape test, integrated
into kernels/flux.py + integrations/mflux.py in v0.3.0) and two research notes. None ...
fix(ship): validate MR description first line client-side (#1367)
What
validate_mr_metadata (the single deterministic chokepoint behind OverlayMetadata.validate_pr, the t3 tool validate-mr
CLI, and the ...
Intent
Closes 007 s last deferred Done-When (e): machine-checkable computed gates that honor CORE.md:57-62 s machine-vs-human
gate split.
What shipped
- music.pregen_check(lifecycle_id, concept_ready, ...
!-- Thanks for contributing a pull request, we appreciate you!
If this PR fixes an issue, please reference it, e.g., Fixes #1234 .
You can delete this comment.
Summary
- Replaces peer-facing import draft CSV language with import-file guidance
- Keeps validation, preview-first, and copy-only guardrails intact
- Adds dashboard helper assertions to prevent ...
Summary
When Studio loads a GGUF vision model it starts llama-server with --mmproj projector . If the installed llama.cpp
prebuilt is older than the model s vision projector format, llama-server aborts ...
Refresh the hops-app skill with the current app model: Streamlit and custom apps, HopsFS or Git sources, git-backed
examples for Streamlit, FastAPI, Flask, and Gradio, and the current monitoring defaults. ...
Summary
- ElevenLabs returns multiple previews per design call but the bill is per call, not per preview. We were already
taking previews[0] and discarding the rest. This PR keeps the discarded ...