Fix: vocab_size=947 (BPE base from model.vocab); top-level weights; all compat fixes

Fix: vocab_size=947 (BPE base from model.vocab); top-level weights; all compat fixes

Fix: vocab_size=947 (BPE base from model.vocab); top-level weights; all compat fixes

Fix: vocab_size=32005 from step-branch tokenizer; top-level weights; ZeroDivisionError guard; all_tied_weights_keys->dict

Fix: vocab_size=32768 from tokenizer; top-level weights; ZeroDivisionError guard; all_tied_weights_keys->dict

Fix pico_decoder.py: __init__ defaults, ZeroDivisionError, all_tied_weights_keys

Fix: add PicoDecoderHFConfig.__init__ with MODEL_BASE defaults; config.json from generate_configs.py MODEL_BASE

Add pico_decoder.py + auto_map config (main)

Fix config.json — inferred from step-30000 (vocab=947, hidden=768, layers=14)

initial commit