Gemma3-1B-IT

An-Upfeat/Gemma3-1B-IT

Gemma3-1B-IT is a text generation model from An-Upfeat. Use it when you need the model to write or continue text. The card lists the license as gemma.

This model provides a few variants of google/Gemma-3-1B-IT that are ready for deployment on Android using the LiteRT (fka TFLite) stack and MediaPipe LLM Inference API.

Downloads · 30 days

Access

Public

Updated Jul 30, 2025

Repo size

555 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.task555 MB · 100%

At a glance

Task: Text Generation
License: gemma
Access: Public
Created: Jul 30, 2025
Updated: Jul 30, 2025
SHA: 0b3d9b32

Try a prompt

Base models

google/Gemma-3-1B-IT

Model card

From the Hugging Face model README

litert-community/Gemma3-1B-IT

This model provides a few variants of google/Gemma-3-1B-IT that are ready for deployment on Android using the LiteRT (fka TFLite) stack and MediaPipe LLM Inference API.

Use the models

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.

Customize

Fine tune Gemma 3 1B and deploy with either LiteRT or Mediapipe LLM Inference API:

Android via Google AI Edge Gallery and MediaPipe

Download and install the apk.
Follow the instructions in the app.

To build the demo app from source, please follow the instructions from the GitHub repository.

Android or Desktop via LiteRT LM

Follow the LitRT LM instructions to build our Open Source LiteRT LM runtime to run LiteRT models.

iOS via MediaPipe

Clone the MediaPipe samples repository and follow the instructions to build the LLM Inference iOS Sample App using XCode.
Run the app via the iOS simulator or deploy to an iOS device.

Performance

Android via Google AI Edge Gallery and MediaPipe

Note that all benchmark stats are from a Samsung S24 Ultra and use models with multiple prefill signatures.

<table border="1"> <tr> <th style="text-align: left">Backend</th> <th style="text-align: left">Quantization scheme</th> <th style="text-align: left">Context length</th> <th style="text-align: left">Prefill (tokens/sec)</th> <th style="text-align: left">Decode (tokens/sec)</th> <th style="text-align: left">Time-to-first-token (sec)</th> <th style="text-align: left">CPU Memory (RSS in MB)</th> <th style="text-align: left">GPU Memory (RSS in MB)</th> <th style="text-align: left">Model size (MB)</th> <th></th> </tr> <tr> <td rowspan="8">CPU</td> <td>fp32 (baseline)</td> <td>1280</td> <td>49 tk/s</td> <td>10 tk/s</td> <td>5.59 s</td> <td>4,123 MB</td> <td></td> <td>3,824 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_f32_ekv1280.task">🔗</a></td> </tr> <tr> <td rowspan="2">dynamic_int4 (block size 128)</td> <td>1280</td> <td>138 tk/s</td> <td>50 tk/s</td> <td>2.33 s</td> <td>982 MB</td> <td></td> <td>657 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q4_block128_ekv1280.task">🔗</a></td> </tr> <tr> <td>4096</td> <td>87 tk/s</td> <td>37 tk/s</td> <td>3.40 s</td> <td>1,145 MB</td> <td></td> <td>657 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q4_block128_ekv4096.task">🔗</a></td> </tr> <tr> <td rowspan="2">dynamic_int4 (block size 32)</td> <td>1280</td> <td>107 tk/s</td> <td>48 tk/s</td> <td>3.49 s</td> <td>1,045 MB</td> <td></td> <td>688 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q4_block32_ekv1280.task">🔗</a></td> </tr> <tr> <td>4096</td> <td>79 tk/s</td> <td>36 tk/s</td> <td>4.40 s</td> <td>1,210 MB</td> <td></td> <td>688 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q4_block32_ekv4096.task">🔗</a></td> </tr> <tr> <td>dynamic_int4 QAT</td> <td>2048</td> <td>322 tk/s</td> <td>47 tk/s</td> <td>3.10 s</td> <td>1,138 MB</td> <td></td> <td>529 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int4.task">🔗</a></td> </tr> <tr> <td rowspan="2">dynamic_int8</td> <td>1280</td> <td>177 tk/s</td> <td>33 tk/s</td> <td>1.69 s</td> <td>1,341 MB</td> <td></td> <td>1,005 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q8_ekv1280.task">🔗</a></td> </tr> <tr> <td>4096</td> <td>123 tk/s</td> <td>29 tk/s</td> <td>2.34 s</td> <td>1,504 MB</td> <td></td> <td>1,005 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q8_ekv4096.task">🔗</a></td> </tr> <tr> <td rowspan="3">GPU</td> <td>dynamic_int4 QAT</td> <td>2048</td> <td>2585 tk/s</td> <td>56 tk/s</td> <td>4.50 s</td> <td>1,205 MB</td> <td></td> <td>529 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int4.task">🔗</a></td> </tr> <tr> <td rowspan="2">dynamic_int8</td> <td>1280</td> <td>1191 tk/s</td> <td>24 tk/s</td> <td>4.68 s</td> <td>2,164 MB</td> <td>1,059 MB</td> <td>1,005 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q8_ekv1280.task">🔗</a></td> </tr> <tr> <td>4096</td> <td>814 tk/s</td> <td>24 tk/s</td> <td>4.99 s</td> <td>2,167 MB</td> <td>1,181 MB</td> <td>1,005 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/Gemma3-1B-IT_multi-prefill-seq_q8_ekv4096.task">🔗</a></td> </tr> </table>

For the list of supported quantization schemes see supported-schemes. For these models, we are using prefill signature lengths of 32, 128, 512 and 1280.
Model Size: measured by the size of the .tflite flatbuffer (serialization format for LiteRT models)
Memory: indicator of peak RAM usage
The inference on CPU is accelerated via the LiteRT XNNPACK delegate with 4 threads
Benchmark is run with cache enabled and initialized. During the first run, the time to first token may differ.

Android via LiteRT LM

Note that all benchmark stats are from a Samsung S24 Ultra and use models with multiple prefill signatures.

For the list of supported quantization schemes see supported-schemes. For these models, we are using prefill signature lengths of 32, 128, 512 and 1280.
Model Size: measured by the size of the .tflite flatbuffer (serialization format for LiteRT models)
Memory: indicator of peak RAM usage
The inference on CPU is accelerated via the LiteRT XNNPACK delegate with 4 threads
Benchmark is run with cache enabled and initialized. During the first run, the time to first token may differ.

Android via LiteRT LM with NPU

Note that the benchmark stats are from a Samsung S25 Ultra and use models with 128 token prefill chunks towards 1024 tokens.

Model Size: measured by the size of the .tflite flatbuffer (serialization format for LiteRT models)
Memory: indicator of peak RAM usage from malloc.

Web

Note that all benchmark stats are from a MacBook Pro 2024 (Apple M4 Max chip) running with 1280 KV cache size, 1024 tokens prefill, 256 tokens decode.

<table border="1"> <tr> <th style="text-align: left">Backend</th> <th style="text-align: left">Quantization scheme</th> <th style="text-align: left">Precision</th> <th style="text-align: left">Prefill (tokens/sec)</th> <th style="text-align: left">Decode (tokens/sec)</th> <th style="text-align: left">Time-to-first-token (sec)</th> <th style="text-align: left">CPU Memory</th> <th style="text-align: left">GPU Memory</th> <th style="text-align: left">Model size (MB)</th> <th></th> </tr> <tr> <td rowspan="5">GPU</td> <td rowspan="2">dynamic_int4</td> <td>F16</td> <td>4339 tk/s</td> <td>133 tk/s</td> <td>0.51 s</td> <td>460 MB</td> <td>1,331 MB</td> <td>700 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int4-web.task">🔗</a></td> </tr> <tr> <td>F32</td> <td>2837 tk/s</td> <td>134 tk/s</td> <td>0.49 s</td> <td>481 MB</td> <td>1,331 MB</td> <td>700 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int4-web.task">🔗</a></td> </tr> <tr> <td>dynamic_int4 QAT</td> <td>F16</td> <td>1702 tk/s</td> <td>77 tk/s</td> <td></td> <td></td> <td></td> <td>529 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int4.task">🔗</a></td> </tr> <tr> <td rowspan="2">dynamic_int8</td> <td>F16</td> <td>4321 tk/s</td> <td>126 tk/s</td> <td>0.6 s</td> <td>471 MB</td> <td>1,740 MB</td> <td>1,011 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int8-web.task">🔗</a></td> </tr> <tr> <td>F32</td> <td>2805 tk/s</td> <td>129 tk/s</td> <td>0.58 s</td> <td>474 MB</td> <td>1,740 MB</td> <td>1,011 MB</td> <td><a style="text-decoration: none" href="https://huggingface.co/litert-community/Gemma3-1B-IT/resolve/main/gemma3-1b-it-int8-web.task">🔗</a></td> </tr> </table>

Model size: measured by the size of the .tflite flatbuffer (serialization format for LiteRT models)
dynamic_int4: quantized model with int4 weights and float activations.
dynamic_int8: quantized model with int8 weights and float activations.
a16w4: quantized model with int4 weights and int16 activations.

Task: Text Generation
License: gemma
Created: Jul 30, 2025
Updated: Jul 30, 2025