Capability matrix
- Compute and format FP32 compute from BF16 weights · CPU; see model report.
- Local entry point WAV generation.
- Server Unverified for this artifact. Text to speech
- Harness No recorded coding task. Check listening quality and a matching service deployment.
When to use it
Experimental FP32 text-to-WAV, including a sentence-chunked 100-word trial. Listening quality, speed and publisher BF16 agreement remain open.
Configuration and tested scope · Publisher model card
Hardware and tuning
No comparable x86/ARM prefill and decode benchmark pair is recorded here. Full-checkpoint training, LoRA and other PEFT methods: no verified CKE recipe for this exact checkpoint in the reviewed evidence. For a trainable starter, use the FP32 and adapter recipes.
Write your first WAV
Download the checkpoint and the publisher source revision listed in the linked Ming3 guide. Replace the three paths with your local paths. Use an empty output directory.
python3 version/v8/tts/run_ming3_native_v8.py \
--checkpoint /path/to/checkpoint --publisher-root /path/to/pinned/publisher \
--output /path/to/new-run --text "Hello. This is Ming speech." \
--capacity 256 --max-frames 100 --threads 2Read the completion status, then listen to the saved audio. The native run uses FP32 computation with weights originating in BF16.