File size: 1,735 Bytes
4f45291
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b602e5c
 
 
5726fe6
 
b602e5c
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
---
license: mit
tags:
- LiteRT
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
---

# litert-community/DeepSeek-R1-Distill-Qwen-1.5B

This model was converted to LiteRT (aka TFLite) format from [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) using [Google AI Edge Torch](https://github.com/google-ai-edge/ai-edge-torch).

## Run the model in colab

[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/#fileId=https://huggingface.co/litert-community/DeepSeek-R1-Distill-Qwen-1.5B/blob/main/deepseek%20tflite.ipynb)

## Run the model on Android

Please follow the [instructions](https://github.com/google-ai-edge/mediapipe-samples/blob/main/examples/llm_inference/android/README.md).

## Benchmarking results

Note that all benchmark stats are from a Samsung S24 Ultra.

| Model                                          | Params |                    | GGML tk/s (CPU, 4 threads) | GGML tk/s (CPU, 8 threads) | LiteRT tk/s (XNNPACK, 4 threads) | LiteRT tk/s (XNNPACK, 8 threads) |
| ---------------------------------------------- | ------ | ------------------ | -------------------------- | -------------------------- | -------------------------------- | -------------------------------- |
| DeepSeek-R1-Distill-Qwen-1.5B (Int8 quantized) | 1.78 B | Prefill 512 tokens | 64.66                      | 87.18                      | 260.95                           | 299.15                           |
|                                                |        | Decode 128 tokens  | 23.85                      | 15.37                      | 23.126                           | 10.486                           |