SmolVLM Benchmark Demo
Model Options (Note: Benchmarking all three SmolVLM models by default)
Select Model ID:
hf-internal-testing/tiny-random-Idefics3ForConditionalGeneration
HuggingFaceTB/SmolVLM-256M-Instruct
HuggingFaceTB/SmolVLM-500M-Instruct
HuggingFaceTB/SmolVLM-Instruct
Decoder (decoder_model_merged) dtype:
fp32
fp16
q8
q4
q4f16
Embed Tokens dtype:
fp32
fp16
q8
q4
q4f16
Vision Encoder dtype:
fp32
fp16
q8
q4
q4f16
Hardware Options
Select Device:
wasm
webgpu
Benchmark Options
Image URL:
Do Image Splitting (do_image_splitting)
Number of Tokens to Generate:
Number of Runs:
Start Benchmark