---
base_model:
- XHToken/Spark-X2.5-4B
license: apache-2.0

library_name: mlx
pipeline_tag: text-generation
tags:
- mlx
- spark-x2.5
- long-context
- 1m-context
---

> [!IMPORTANT]
> **Compatibility:**
> This model uses the `spark2_5` architecture. It is currently supported with [Spark-MLX-LLM](https://github.com/XHToken/Spark-MLX-LLM), but not yet with oMLX or standard MLX-LM.

# Spark-X2.5-4B MLX 8-bit

MLX 8-bit quantization of [XHToken/Spark-X2.5-4B](https://huggingface.co/XHToken/Spark-X2.5-4B), a 4B general-purpose language model for reasoning, coding, tool use, and agentic workflows. Native context: **1,048,576 tokens (1M)**.

## Benchmarks

![Upstream Spark-X2.5-4B benchmark comparison](assets/benchmark.png)

*Benchmark results reported by XHToken for Spark-X2.5-4B in thinking mode.*

## Files

| Format | Weights | Size |
| --- | --- | ---: |
| MLX 8-bit | [model.safetensors](model.safetensors) | 4.37 GB |

Includes the upstream `chat_template.jinja`. Checksums: [SHA256SUMS.txt](SHA256SUMS.txt).

## Source

- Source model: [XHToken/Spark-X2.5-4B](https://huggingface.co/XHToken/Spark-X2.5-4B)
- Source revision: [`ea14618d20e76b5b093d3ee20a5b9d733bb12410`](https://huggingface.co/XHToken/Spark-X2.5-4B/tree/ea14618d20e76b5b093d3ee20a5b9d733bb12410)
- Source license: [Apache-2.0](https://huggingface.co/XHToken/Spark-X2.5-4B/blob/main/LICENSE)