le-harnais-ft-smoke-regular

le-harnais / ft-smoke-regular

⚠️ retrain-only — this is an smoke checkpoint, not useful for inference. It ships only to reproduce / continue the training study. For real use see the hero models: le-harnais-ft-agentworld-{1b,3b,8b}, le-harnais-ft-counsel.

Pipeline smoke test (128 samples). Validates the training path; not a measured result.

• Base model: meta-llama/Llama-3.2-1B-Instruct — Built with Llama; Llama Community License applies.
• Class: smoke
• Training data: 128-sample smoke split
• Headline: NOT A RESULT — 128-sample/1-epoch pipeline validation only

Reproduce

cd refs/llm-jepa && .venv/bin/python ../../tools/run_jepa_seeds.py --dataset synth --eval-cap 64

Full recipe, datasets, and eval commands: see docs/REPRODUCE.md in the [le-harnais distribution]. Provenance & license: docs/PROVENANCE.md.

Formats in this repo

• *.safetensors — bf16 inference weights (serve with transformers or le-harnais lh-serve/candle).
• *.Q4_K_M.gguf — portable 4-bit quant (run via ollama / llama.cpp; Mac-friendly).

This is a pipeline smoke test, not a measured result.