Update README.md

README.md

@@ -8,41 +8,51 @@ base_model:
 
 # 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).
+This model provides a few variants of [deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B) that are ready for deployment on Android using the [LiteRT (fka TFLite) stack](https://ai.google.dev/edge/litert) and [MediaPipe LLM Inference API](https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inference).
 
-## Run the model in colab
+## Use the models
 
-*Disclaimer: Model performance demonstrated with the Python API in this notebook is not representative of performance on a local device.*
+### Colab
+
+*Disclaimer: The target deployment surface for the LiteRT models is Android/iOS/Web and the stack has been optimized for performance on these targets. Trying out the system in Colab is an easier way to familiarize yourself with the LiteRT stack, with the caveat that the performance (memory and latency) on Colab could be much worse than on a local device.*
 
 [![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
+### Android
+
+To build the demo app from source, please follow the [instructions](https://github.com/google-ai-edge/mediapipe-samples/blob/main/examples/llm_inference/android/README.md) from the GitHub repository.
 
-Please follow the [instructions](https://github.com/google-ai-edge/mediapipe-samples/blob/main/examples/llm_inference/android/README.md).
+## Performance
 
-## Benchmarking results
+### Android
 
-Note that all benchmark stats are from a Samsung S24 Ultra.
+Note that all benchmark stats are from a Samsung S24 Ultra with 1280 KV cache size, 65 tokens prefill, 128 tokens decode.
 
 <table border="1">
   <tr>
-    <th>Model</th>
-      <td colspan="2">DeepSeek-R1-Distill-Qwen-1.5B (Int8 quantized)</td>
-  </tr>
-  <tr>
-    <th>Params</th>
-    <td colspan="2">1.78 B</td>
-  </tr>
-  <tr>
-    <th></th>
-    <td><b>Prefill 512 tokens</b></td><td><b>Decode 128 tokens</b></td>
+   <th></th>
+   <th>Backend</th>
+   <th>Prefill (tokens/sec)</th>
+   <th>Decode (tokens/sec)</th>
+   <th>Time-to-first-token (sec)</th>
+   <th>Memory (RSS in MB)</th>
+   <th>Model size (MB)</th>
   </tr>
   <tr>
-    <th>LiteRT tk/s (XNNPACK, 4 threads)</th>
-    <td>260.95</td><td>23.126</td>
+   <td>fp32 (baseline)</td>
+   <td rowspan="2">CPU</td>
+   <td><p style="text-align: right">45</p></td>
+   <td><p style="text-align: right">6</p></td>
+   <td><p style="text-align: right">1.58</p></td>
+   <td><p style="text-align: right">6,144</p></td>
+   <td><p style="text-align: right">7,124</p></td>
   </tr>
   <tr>
-    <th>GGML tk/s (CPU, 4 threads)</th>
-      <td>64.66</td><td>23.85</td>
+   <td>dynamic_int8</td>
+   <td><p style="text-align: right">271</p></td>
+   <td><p style="text-align: right">23</p></td>
+   <td><p style="text-align: right">0.54 </p></td>
+   <td><p style="text-align: right">1,869 </p></td>
+   <td><p style="text-align: right">1,861</p></td>
   </tr>
-</table>             
+</table>