Blog Share what we know.
Deep dives from the field: Qualcomm toolchains, edge optimization, on-device voice, and robotics. We teach, document, and explain in the open.
/ 01
Featured
Straight from the workbench: technical write-ups from our own builds.
Edge AI · Voice TTS On LinkedIn · Devansh Shukla
SuperTonic: a TTS model, quantized for Qualcomm silicon
A 66M-parameter latent-diffusion TTS model: ConvNeXt backbone, flow-matching diffusion, cross-attention alignment, optimised to run on Qualcomm chipsets through QAIRT, AIMET and QAT quantization pipelines.
Read on LinkedIn
LLM · ML LLM On LinkedIn · Sakthivel Senthilkumar
PebbleLM-117M: training a language model from scratch
A decoder-only transformer built end-to-end: 117M parameters, 8 layers, a custom 16K BPE tokenizer, and base + chat-finetuned variants trained on Wikipedia, OpenWebText and TinyStories.
Read on LinkedIn / 02
More writing
Field notes and technical deep dives, published in the open.
Request a Demo
Want the deep dives first?
Email info@raemox.tech
Ask us anything Toolchains, chips, models, we answer
Principle We explain in the open