TFLite(LiteRT) - Generative AI on MT8875

This page lists the generative AI models and representative performance data for MT8875 platforms. For background information about generative workloads and usage notes, refer to TFLite(LiteRT) - Generative AI.

Model Support and Performance

Note

For OS-level GAI support across all platforms, see AI Supporting Scope.

The following symbols are used in the tables below:

  • -- : To be released.

  • By Request : Available by customer request only.

  • Q3/E : Support planned for Q3 end, 2026.

  • Q4/E : Support planned for Q4 end, 2026.

  • X : Platform does not support this model.

Large Language Models (LLMs)

Performance Comparison (Prompt/Generative) (Unit: tok/s)

Model

MT8875

Qwen3-0.6B

750.47 / 29.48

Qwen3-1.7B

238.31 / 6.63

Qwen3-4B

173.79 / 6.84

Qwen3-8B

88.49 / 3.97

Qwen2.5-1.5B-Instruct

394.42 / 17.20

Qwen2.5-3B-Instruct

220.04 / 9.55

Qwen2.5-7B-Instruct

80.92 / 4.36

gemma3-1B (Text Only)

–

gemma3-4B (Text-Only)

199.32 / 6.14

llama3.2-1B-Instruct

514.98 / 23.17

llama3.2-3B-Instruct

202.58 / 9.29

Phi-3-mini-4k-instruct

Q4/E

Phi-3.5-mini-instruct

Q4/E

Vision-Language Models (VLMs)

Performance Comparison (ViT Inference Time / Prompt / Generative)

Model

ViT Inference Time (s)

Prompt Mode (tok/s)

Generative Mode (tok/s)

Qwen3VL-2B

0.466408

284.238

14.4577

Speech Recognition Models (ASR)

Performance Data

Model

MT8875

Whisper-base (8w16a)

By Request

Qwen3-ASR-0.6B

Q4/E

Moonshine-Tiny

Q4/E

Measured latency for these models is reported in asr_cmdline_tool, together with the transcripts each package’s sample clip is expected to produce.

Android-only Models

The following model categories are supported on Android only and are not available on Yocto.

Note

Yocto support for these model categories is planned for 2026/E (estimated).

Stable Diffusion and Image Generation

Performance Comparison (Main Time/Inference Time) (Unit: ms)

Model

MT8875

Stable Diffusion v2.1 base model with controlnet

By Request

Stable Diffusion v.1.5 controlnet

Q4/E

CLIP and Embedding Models

Performance Comparison (Main Time/Inference Time) (Unit: ms)

Model

MT8875

img_encoder_proj_clip_vit_large_dynamic

Q4/E

img_encoder_proj_openclip_vit_h_dynamic

By Request

text_encoder_clip_vit_large

Q4/E

text_encoder_openclip_vit_h

By Request