---
base_model: IFM/K2-Horizon-3.7B
base_model_relation: quantized
license: apache-2.0
language:
  - en
pipeline_tag: text-generation
library_name: mlx
tags:
  - mlx
  - mlx-lm
  - 8-bit
  - k2-horizon
  - long-context
  - 512k-context
  - dense
---

# K2-Horizon-3.7B MLX — 8-bit

MLX 8-bit conversion of [IFM/K2-Horizon-3.7B](https://huggingface.co/IFM/K2-Horizon-3.7B), a 3.7B dense decoder-only model for reasoning, coding, long-context work, and tool use. The source checkpoint supports a native context length of **524,288 tokens (512K)**.

## Benchmarks

![IFM/K2-Horizon-3.7B benchmark results](assets/k2-horizon-3.7b-benchmarks.png)

*Benchmark results reported by IFM for the original IFM/K2-Horizon-3.7B checkpoint.*

## Release

| Format | Quantization | Size |
| --- | --- | ---: |
| MLX safetensors | Affine 8-bit, group size 64 | 5.40 GB |

The included `k2_horizon_mlx.py` adapter preserves K2 Horizon's grouped RMSNorm. Use it with MLX-LM and `--trust-remote-code`. The model is text-only; no vision projector or MTP files are included.

## Usage

```bash
pip install -U mlx-lm
mlx_lm.generate \
  --model . \
  --trust-remote-code \
  --prompt "Explain why reproducible builds matter." \
  --max-tokens 512
```

## Source

- Model: [IFM/K2-Horizon-3.7B](https://huggingface.co/IFM/K2-Horizon-3.7B)
- Source revision: [`633f52a`](https://huggingface.co/IFM/K2-Horizon-3.7B/tree/633f52ad28b17edeabd82afc61d2d13b4c59a561)
- License: [Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)
- Checksums: [SHA256SUMS.txt](SHA256SUMS.txt)