README.md

---
license: other
license_name: lfm-open-license-v1.0
license_link: https://huggingface.co/LiquidAI/LFM2.5-Encoder-350M/blob/main/LICENSE
library_name: minima-lfm
base_model: LiquidAI/LFM2.5-Encoder-350M
tags:
  - lfm2
  - encoder
  - ternary
  - bitnet
  - quantization
---

# Minima

A strict W1.58A8 adaptation of [LiquidAI/LFM2.5-Encoder-350M](https://huggingface.co/LiquidAI/LFM2.5-Encoder-350M),
built with [SSHDotCodes/minima](https://github.com/SSHDotCodes/minima).

- Logical matrix values: `{-1, 0, +1}` (1.585 bits)
- Physical artifact format: I2_S, four trits per byte
- Dynamic int8 activations
- Group size 128, **no recovery adapters**
- Full 8,192-token encoder context
- Packed weight file: **89.9 MB** (94,298,568 bytes)
- Release status: **release candidate** (six-task gated retention 85.45% vs FP32)

## Use

```bash
pip install "minima-lfm @ git+https://github.com/SSHDotCodes/minima.git"
```

```python
from minima import MinimaModel

model = MinimaModel.from_pretrained("ProCreations/minima", device="cpu")
outputs = model(input_ids=input_ids, attention_mask=attention_mask)
```

CPU inference defaults to a one-time FBGEMM dynamic-int8 packing of each
ternary matrix. Set `MINIMA_CPU_BACKEND=i2s` for the direct packed 2-bit
AVX2/ARM NEON kernel.

## Encoder quality

The six-task downstream gate compares matched 800-step fine-tunes. Packed
ternary matrices stay frozen; only the task head and non-matrix parameters
adapt. Required relative mean: **>= 96%**.

| Task | FP32 | Minima | Capped retention |
| --- | ---: | ---: | ---: |
| SST2 | 0.79817 | 0.75459 | 94.54% |
| QNLI | 0.60608 | 0.59491 | 98.16% |
| MNLI | 0.40621 | 0.38074 | 93.73% |
| MRPC | 0.81694 | 0.81873 | 100.00% |
| STSB | 0.54812 | 0.40660 | 74.18% |
| COLA | 0.11382 | 0.05927 | 52.07% |
| **Mean** | | | **85.45%** |

## License

The weights remain subject to the LFM Open License v1.0 shipped in this
repository. The Minima runtime code is MIT licensed.