---
base_model: LiquidAI/LFM2.5-1.2B-Instruct
base_model_relation: adapter
library_name: mlx
license: other
license_name: lfm1.0
license_link: https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/main/LICENSE
language:
  - en
pipeline_tag: text-generation
tags:
  - system-one
  - lora
  - mlx
  - jev
  - peft
  - text-generation
---

# jevons-lfm25-1.2b-systemone

Seed **LoRA** for [jevons](https://github.com/gopalanj/jevons): a local System One server that scores allowed outcomes from logits and assembles `{choice, probabilities, confidence, noul, score, legend}` in code.

**This is not official Jev.** Official Jev cannot be cloned. These weights only bias LFM2.5 toward the same option-key / yes-no / score-digit tokens jevons reads at serve time.

## Base model

| Role | Hub id |
| --- | --- |
| Official checkpoint | [`LiquidAI/LFM2.5-1.2B-Instruct`](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct) |
| Trained and served 8-bit MLX | [`mlx-community/LFM2.5-1.2B-Instruct-8bit`](https://huggingface.co/mlx-community/LFM2.5-1.2B-Instruct-8bit) |

QLoRA on the MLX 8-bit checkpoint (rank 16, scale 2.0, 16 layers, attn q/k/v/o + MLP w1/w2/w3). Prompt tokens were masked. Completions are teacher-forced aliases that match serve-time scoring: option keys, `yes`/`no`, level digits.

## Serve with jevons

Install jevons from [github.com/gopalanj/jevons](https://github.com/gopalanj/jevons), then download this adapter and the 8-bit base:

```sh
cd jevons
uv sync --extra mlx --extra dev

uv run hf download mlx-community/LFM2.5-1.2B-Instruct-8bit \
    --local-dir models/LFM2.5-1.2B-Instruct-8bit

uv run hf download gopalanj/jevons-lfm25-1.2b-systemone \
    --local-dir adapters/lfm25-1.2b-systemone
```

Serve at **T=1** with calibration off (do not apply a leftover `calibration.json` from the base model):

```sh
JEVONS_ADAPTER=adapters/lfm25-1.2b-systemone \
JEVONS_TEMPERATURE=1 \
JEVONS_CALIBRATION=off \
JEVONS_MODEL=models/LFM2.5-1.2B-Instruct-8bit \
uv run jevons serve --host 127.0.0.1 --port 8000
```

Or:

```sh
uv run jevons serve \
  --adapter adapters/lfm25-1.2b-systemone \
  --temperature 1 \
  --calibration off
```

`POST /v1/systemone` matches the [TypeSafe HTTP contract](https://docs.typesafe.ai/api.md). The adapter autoloads from `adapters/lfm25-1.2b-systemone` when `adapters.safetensors` is present. Disable with `--adapter off`.

## Honest metrics (T=1, no calibration)

Seed-only official-Jev teacher aliases (78 train items / 126 examples; frozen holdout 17 items / 29 examples). Schema validity is 100% because jevons never asks the model to write JSON.

| split | run | modal | choice / noul / score | ECE | Brier | acc@≥0.8 | schema |
| --- | --- | --- | --- | --- | --- | --- | --- |
| full n=155 | base | 69.7% | 78.6 / 68.2 / 56.1 | 0.107 | 0.387 | 92.1% (n=63) | 100% |
| full n=155 | LoRA | **83.9%** | 85.7 / 95.5 / 68.3 | 0.077 | 0.193 | 95.9% (n=97) | **100%** |
| holdout n=29 | base | 72.4% | 76.9 / 87.5 / 50.0 | 0.122 | 0.390 | 100% (n=11) | 100% |
| holdout n=29 | LoRA | **72.4%** | 69.2 / 75.0 / 75.0 | 0.248 | 0.303 | 100% (n=14) | **100%** |

Full-set modal **83.9% includes the 78 train items**. The honest ship metric is **holdout modal 72.4%**, unchanged vs base and below a 90% bar. Holdout ECE got worse (0.122 → 0.248). This is a ranking adapter, not calibrated System One / RLCD.

## Training

- Data: seed teacher aliases only (not grown templates)
- Iters: 64 (2 epochs), batch 4, AdamW 5e-5, ~131s on Apple Silicon
- Failed grown run is **not** these weights: repetitive `grow.py` templates + LoRA scale 20 collapsed choice (holdout modal 38%). Do not serve that run.

See `hyperparams.json` / `hyperparams.md` in this repo.

## License caveat

- **Adapter + LFM weights:** [LFM 1.0](https://huggingface.co/LiquidAI/LFM2.5-1.2B-Instruct/blob/main/LICENSE) (`license: other` / `lfm1.0`). Obtain and accept a license from Liquid AI before downloading or using the base model or this derivative adapter. This repo does **not** ship LFM base weights.
- **jevons server code:** MIT — [github.com/gopalanj/jevons](https://github.com/gopalanj/jevons)

## Files

- `adapters.safetensors` — MLX LoRA weights
- `adapter_config.json` — mlx-lm LoRA config
- `hyperparams.json` / `hyperparams.md` — training report and eval numbers