Skip to content

Integrate desktop training fixes on the API completion branch - #265

Draft
ProfSynapse wants to merge 7 commits into
mainfrom
release/api-v1-completion
Draft

ProfSynapse wants to merge 7 commits into
mainfrom
release/api-v1-completion

Conversation

@ProfSynapse

@ProfSynapse ProfSynapse commented Oct 20, 2026 •

Copy link
Copy Markdown
Owner

Collect the reviewed desktop-training fixes on a dedicated API completion branch: model SDK validation before consuming preparation claims, closed offline SFT worker diagnostics, and the internal MLX prepared-token SFT boundary. This gives qualification a single pushed source while keeping main unchanged.

Validation on exact head 0d515ae: all 18 CI checks passed, including Linux/Windows/macOS installed-base compatibility and provider-free coordinator conformance. The integrated local suite passed 265 CPU tests; all seven closure/lock checks are current and canonical skill mirrors match. Independent reviews accepted the diagnostics and internal MLX scope.

Live qualification remains incomplete. The fresh Windows SFT attempt failed at the closed Torch-import boundary; the fresh Modal GRPO attempt passed its CPU package check but failed at trainer construction. Neither failure establishes its underlying cause. Both consumed attempts are preserved. Apple execution still needs the native runtime/provider joins and real hardware qualification. Syntunia's private planning and known-run read bridge is merged, while product training custody and the engine pointer remain separate gates. This draft is not a qualified release.

Includes the changes reviewed in #263 and #264 plus the independently reviewed MLX increment f072d97. Further reviewed fixes target this branch. HF Jobs qualification remains paused.

…api-v1-completion

# Conflicts:
#	.agents/skills/fine-tuning/reference/sft-training.md
#	.claude/skills/fine-tuning/reference/sft-training.md
#	.skills/fine-tuning/reference/sft-training.md
…api-v1-completion

# Conflicts:
#	.agents/skills/fine-tuning/SKILL.md
#	.claude/skills/fine-tuning/SKILL.md
#	.skills/fine-tuning/SKILL.md
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