---
license: apache-2.0
tags:
  - gguf
  - lars
---

# lars packs

Private backup of the lars-sdk dev packs. They were rebuilt on 2026-10-08 and 09, after the
machine that held the originals died. Each pack is in `packs/<name>/<version>/`:

- `model.gguf`: the fused LoRA model, imatrix-quantized.
- `calibration.json`: the calibrator, fitted on 4,120 jev-bench validation rows.
- `pack.json`: the manifest, with the base, the llama.cpp commit, sha256 sums and per-config scores.

Scores are on the jev-bench test split (22 configs, 200 rows each), through the pack:

| Pack | Base, checkpoint | Quant | Macro accuracy | Macro Brier | Lost original |
|---|---|---|---|---|---|
| `lars-0.6b-dev` 0.0.2 | Qwen/Qwen3-0.6B, final step 4,449 | Q4_K_M | 0.637 | 0.418 | 0.650 / 0.419 |
| `lars-0.6b-dev-step2500` 0.0.2 | Qwen/Qwen3-0.6B, step 2,500 (selected) | Q4_K_M | 0.625 | 0.433 | |
| `lars-2b-dev` 0.0.1 | Qwen/Qwen3.5-2B, final step 5,370 | Q4_K_M | 0.713 | 0.331 | 0.716 / 0.333 |

- **Training:** LoRA on corpus E. The 2B also trained on the typed-decisions train split; it
  scores 0.757 on that benchmark's test split, against 0.759 for the lost pack.
- **The step-2,500 pack:** kept for comparison. The packs build from the final step.
- **`lars-8b-dev` 0.0.2** (Qwen/Qwen3-8B, Q3_K_M, was 0.748) is not rebuilt yet. It needs a
  machine with more than 64 GB; lars-sdk's `HANDOFF.md` has the steps.

Apache-2.0, like the base models.