---
license: apache-2.0
---

This model provides [HuggingFaceTB/SmolVLM-256M-Instruct](https://huggingface.co/HuggingFaceTB/SmolVLM-256M-Instruct) model in tflite format.    
You can use this model with [AI Edge Cpp Example](https://github.com/google-ai-edge/ai-edge-torch/tree/main/ai_edge_torch/generative/examples/cpp). 
You need to slightly modify this pipeline to support image as input (see COLAB example below).  
Currently, [AI Edge Torch](https://github.com/google-ai-edge/ai-edge-torch/tree/main/ai_edge_torch/generative/examples) vlms not supported
on [MediaPipe LLM Inference API](https://ai.google.dev/edge/mediapipe/solutions/genai/llm_inference),
for example [qwen_vl model](https://github.com/google-ai-edge/ai-edge-torch/tree/main/ai_edge_torch/generative/examples/qwen_vl),  
that was used as reference to write SmolVLM-256M-Instruct convertation scripts (coming soon).


## Use the models

### Colab




## Details

The model was converted with:

```shell
python convert_to_tflite.py --quantize="dynamic_int8"\
 --checkpoint_path='./models/SmolVLM-256M-Instruct' --output_path="./models/SmolVLM-256M-Instruct-tflite"\
 --mask_as_input=True --prefill_seq_lens=256 --kv_cache_max_len=2048
```