27B model with skills layer handles regulated data workflows locally
Ubunta · x · 2026-08-29
- Experiment: Tested if adding skills to a small 27B model could handle regulated data workflows instead of moving to a larger model.
- Setup: 27B dense model (Q6), 2 consumer GPUs, fully local, no fine-tuning.
- Implementation: 5 skill MD files (cohort specs, value sets, index/washout logic, bias review, ledger classification), constrained JSON output. All codes, counts, and numbers come from tool calls, not the model.
- Findings: Most improvements came from the skill and context layer. Main issue is loading the wrong skill. Adding skills may be sufficient for many tasks before scaling up.
Related event: 27B Model with Skill Layer Handles Regulated Healthcare Workflows Locally(2 posts)→
More from coding & agent
- Same prompt, three tools: Claude vs ChatGPT vs Figma Make building a trips app UI — Tegadesigns · 2026-10-16
- TRL adds day-0 support for fine-tuning Muse Spark Open's 816B MoE with LoRA on 64 H100s — QGallouedec · 2026-10-16
- TRL publishes example code and SLURM script for fine-tuning Muse Spark Open — QGallouedec · 2026-10-16
- Muse Spark Open: Meta's 816B multimodal MoE built for agentic coding hits Hugging Face — mervenoyann · 2026-10-16
- Databricks engineers land impactful PRs on day 2 with AI dev tools, vs week 3 before — Yuchenj_UW · 2026-10-16
- Perplexity open-sources SPACE, the sandbox platform powering its Computer agents — AravSrinivas · 2026-10-16