---
license: apache-2.0
base_model: XHToken/Spark-X2.5-1.7B
pipeline_tag: text-generation
tags:
- rkllm
- rockchip
- rk3576
- rk3588
- quantized
- spark2_5
library_name: rkllm
---

# Spark-X2.5-1.7B · RKLLM

* Max context length: 4096

Text generation on Rockchip RK3576 and RK3588, converted from [XHToken/Spark-X2.5-1.7B](https://huggingface.co/XHToken/Spark-X2.5-1.7B).

- **Conversion toolkit:** RKLLM Toolkit 1.3.1.
- **Inputs / output:** text → text.
- **License:** `apache-2.0`; the [upstream model card](https://huggingface.co/XHToken/Spark-X2.5-1.7B) describes its terms and intended use.

## Choose your files

Use the files for your board’s exact SoC. `w4a16` uses 4-bit weights and 16-bit activations; `w4a16_g128` adds a weight group size of 128. `w8a8` uses 8-bit weights and activations. File size below is download size, not total runtime RAM.

| Board | Quantization | Size | Language model |
| --- | --- | --- | --- |
| RK3576 | `w4a16_g128` | 1.56 GiB | [Spark-X2.5-1.7B_RK3576_w4a16_g128.rkllm](RK3576/Spark-X2.5-1.7B_RK3576_w4a16_g128.rkllm) |
| RK3576 | `w8a8` | 2.14 GiB | [Spark-X2.5-1.7B_RK3576_w8a8.rkllm](RK3576/Spark-X2.5-1.7B_RK3576_w8a8.rkllm) |
| RK3588 | `w8a8` | 2.12 GiB | [Spark-X2.5-1.7B_RK3588_w8a8.rkllm](RK3588/Spark-X2.5-1.7B_RK3588_w8a8.rkllm) |

## Run on your board

With Docker installed on a matching Rockchip board, this example starts the **RK3576 w4a16_g128** model and its API on port 8001:

```bash
sudo docker run --rm -d --name rkllm \
  --privileged -v /dev:/dev -p 8001:8001 \
  ghcr.io/hanzo-huang/rkllm-docker/llm/spark-x2.5-1.7b:rk3576-w4a16-g128 \
  serve --max_context_len 4096
```

The image includes its model file. See [rkllm-docker](https://github.com/Hanzo-Huang/rkllm-docker#readme) for API requests, other board variants, and configuration.

To download the files separately instead:

```bash
hf download HanzoHuang/Spark-X2.5-1.7B-RKLLM \
  RK3576/Spark-X2.5-1.7B_RK3576_w4a16_g128.rkllm \
  --local-dir ./models
```

These are native RKLLM/RKNN artifacts. Use a compatible RKLLM runtime and the model’s correct chat template when integrating them in another application.

## Release notes

All three variants passed functional checks on their matching RK3576/RK3588 boards using RKLLM 1.3.1 and the packaged Docker images: arithmetic, a factual question, streamed output, and conversation recall. These are smoke tests, not a quality benchmark.

The exported context length is 4,096 tokens. Conversion used ModelScope source weights and SDK-generated calibration samples. Spark calibration required an isolated Transformers 4.57.1 environment; other conversion environments were left unchanged.

<details>
<summary>SHA-256 checksums</summary>

```text
725bed8bbd4e6c682b542920dd43081a7f1c29ce5e38b69eceb3e9e8c710e5b4  RK3576/Spark-X2.5-1.7B_RK3576_w4a16_g128.rkllm
fe47ea39dd7968f0cfd4673e9ed189dbec768600dd35f068ca3f33c38a9b8267  RK3576/Spark-X2.5-1.7B_RK3576_w8a8.rkllm
4c87d982c70424ce493a96f2c06bfb49ca9f5e7fd9f805c70f5cf7eb14f64577  RK3588/Spark-X2.5-1.7B_RK3588_w8a8.rkllm
```

</details>

A downloadable checksum list is also available in [SHA256SUMS](SHA256SUMS).

## Credits

Original model: [XHToken/Spark-X2.5-1.7B](https://huggingface.co/XHToken/Spark-X2.5-1.7B). Conversion tools: [Rockchip RKLLM](https://github.com/airockchip/rknn-llm).