My-MiniGPT 254.5M FP16 Weights
Weights-only FP16 inference artifact for the ground-up PyTorch My-MiniGPT model.
The model is primarily trained for English and Simplified-Chinese literary continuation, including narrative prose, scene description, dialogue, and style imitation. The repository contains the model configuration, FP16 weights, and the matching 50,304-token byte-level BPE tokenizer.
This repository intentionally does not contain optimizer state, scheduler state, AMP state, or random-number-generator state. The full training checkpoint remains separate and is required only for resuming training.
The model architecture and inference code are maintained at envymoon/My-MiniGPT.