paLLM: On-Device Language Model for a Palm PDA

paLLM is a constrained-computing experiment: a tiny transformer language model running locally on a Palm Tungsten E2. I built a custom Palm OS app in C that embeds the model in Palm resources and accepts prompts through an editable input field.

The implementation uses a 512-token tokenizer, a TinyStories/llama2.c-style 260K-parameter checkpoint quantized row-wise to INT8, fixed-point inference, and decoder guardrails. Once the .prc is installed, generation runs entirely on the device with no server in the loop.

On the tested Tungsten E2, generation runs at roughly one token every seven seconds. This is an experimental demonstration of local inference under severe hardware constraints.

Source: https://github.com/maddiedreese/paLLM