Document RKLLM artifacts and checksums
Browse files
README.md
CHANGED
|
@@ -1,3 +1,57 @@
|
|
| 1 |
---
|
| 2 |
license: llama3.2
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
---
|
| 2 |
license: llama3.2
|
| 3 |
+
base_model: meta-llama/Llama-3.2-3B-Instruct
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
+
library_name: rkllm
|
| 6 |
+
tags:
|
| 7 |
+
- rkllm
|
| 8 |
+
- rockchip
|
| 9 |
+
- rk3576
|
| 10 |
+
- rk3588
|
| 11 |
+
- llama
|
| 12 |
+
- llama-3.2
|
| 13 |
---
|
| 14 |
+
|
| 15 |
+
# Llama-3.2-3B-Instruct-RKLLM
|
| 16 |
+
|
| 17 |
+
RKLLM-converted Llama 3.2 3B Instruct language-model artifacts for Rockchip RK3576 and RK3588 NPUs.
|
| 18 |
+
|
| 19 |
+
These hardware-specific `.rkllm` files require a compatible Rockchip RKLLM runtime. They are not Transformers checkpoints and cannot be loaded directly with Transformers, llama.cpp, or Ollama.
|
| 20 |
+
|
| 21 |
+
## Base model
|
| 22 |
+
|
| 23 |
+
- Upstream model: [meta-llama/Llama-3.2-3B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct)
|
| 24 |
+
- License: Llama 3.2 Community License
|
| 25 |
+
- Model type: LLM (text only)
|
| 26 |
+
|
| 27 |
+
The upstream model is gated. Review Meta's license and acceptable-use policy before use or redistribution.
|
| 28 |
+
|
| 29 |
+
## Conversion and variants
|
| 30 |
+
|
| 31 |
+
Converted with RKLLM Toolkit 1.3.0. Use a file built for the exact target SoC.
|
| 32 |
+
|
| 33 |
+
| Target | Quantization | File | SHA256 |
|
| 34 |
+
| --- | --- | --- | --- |
|
| 35 |
+
| RK3576 | W4A16 (g128) | [Llama-3.2-3B-Instruct_RK3576_w4a16_g128.rkllm](RK3576/Llama-3.2-3B-Instruct_RK3576_w4a16_g128.rkllm) | `211ddffb27a2e429917639ce319a96fcc878b493433ebca3d0c28666a188750e` |
|
| 36 |
+
| RK3576 | W8A8 | [Llama-3.2-3B-Instruct_RK3576_w8a8.rkllm](RK3576/Llama-3.2-3B-Instruct_RK3576_w8a8.rkllm) | `117fa31e1f7d483d02513ddb073f1818b62ea48affc474f1e562e733179e50e9` |
|
| 37 |
+
| RK3588 | W8A8 | [Llama-3.2-3B-Instruct_RK3588_w8a8.rkllm](RK3588/Llama-3.2-3B-Instruct_RK3588_w8a8.rkllm) | `5373bb69394504fd7df55087a3431ccacb6af6df52781940ff11307ab409d535` |
|
| 38 |
+
|
| 39 |
+
The repository also includes `Llama-3.2-3B-Instruct_data_quant.json`, used as calibration data during conversion.
|
| 40 |
+
|
| 41 |
+
## Usage
|
| 42 |
+
|
| 43 |
+
```bash
|
| 44 |
+
hf download HanzoHuang/Llama-3.2-3B-Instruct-RKLLM \
|
| 45 |
+
RK3576/Llama-3.2-3B-Instruct_RK3576_w4a16_g128.rkllm \
|
| 46 |
+
--local-dir Llama-3.2-3B-Instruct-RKLLM
|
| 47 |
+
```
|
| 48 |
+
|
| 49 |
+
Use the Llama 3.2 Instruct chat template with the RKLLM runtime. For Docker deployment, see [Hanzo-Huang/rkllm-docker](https://github.com/Hanzo-Huang/rkllm-docker).
|
| 50 |
+
|
| 51 |
+
## Limitations
|
| 52 |
+
|
| 53 |
+
These are target-specific converted artifacts. Validate quality, memory use, and runtime compatibility on your own Rockchip device.
|
| 54 |
+
|
| 55 |
+
## Acknowledgements
|
| 56 |
+
|
| 57 |
+
Thanks to Meta, Rockchip, and the RKLLM community.
|