TFLite(LiteRT) - Generative AI on Genio 520
This page lists the generative AI models and representative performance data for Genio 520 platforms. For background information about generative workloads and usage notes, refer to TFlite(LiteRT) - Generative AI.
Model Support and Performance
Note
For VLM, Qwen3-vl is to be released by the end of June,2026.
The following tables summarize the supported generative models and measured performance on Genio 520.
Model |
Prompt Mode (tok/s) |
Generative Mode (tok/s) |
DeepSeek-R1-Distill-Qwen-1.5B |
276.82 |
9.18 |
DeepSeek-R1-Distill-Qwen-7B |
Not Support |
Not Support |
DeepSeek-R1-Distill-Llama-8B |
Not Support |
Not Support |
Qwen3-1.7B |
229.63 |
10.53 |
Qwen2.5-1.5B-Instruct |
269.87 |
15.11 |
Qwen2.5-3B-Instruct |
Not Support |
Not Support |
Qwen2.5-7B-Instruct |
Not Support |
Not Support |
gemma3-1B (Text Only) |
491.53 |
22.13 |
gemma2-2b-it |
Not Support |
Not Support |
llama3.2-1B-Instruct |
335.92 |
20.59 |
llama3.2-3B-Instruct |
Not Support |
Not Support |
llama3-8b |
Not Support |
Not Support |
MiniCPM-2B-sft-bf16-llama-format |
153.14 |
6.48 |
llava1.5-7b-speculative-decoding |
Not Support |
Not Support |
medusa_v1_0_vicuna_7b_v1.5 |
Not Support |
Not Support |
vicuna1.5-7b-tree-speculative-decoding-plus |
Not Support |
Not Support |
baichuan-7b-int8-cache |
Not Support |
Not Support |
Model |
Main Time (ms) |
Inference Time (ms) |
Stable Diffusion v.1.5 controlnet |
47923 |
37581 |
Stable Diffusion v2.1 base model with controlnet |
Not Support |
Not Support |
Model |
Main Time (ms) |
Inference Time (ms) |
CLIP Image Encoder |
||
img_encoder_proj_clip_vit_large_dynamic |
662.22 |
296.84 |
img_encoder_proj_openclip_vit_big_g_dynamic |
Not Support |
Not Support |
img_encoder_proj_openclip_vit_h_dynamic |
Not Support |
Not Support |
CLIP Text Encoder |
||
text_encoder_clip_vit_large |
748.45 |
45.56 |
text_encoder_openclip_vit_h |
Not Support |
Not Support |