NeuronUz/NeuronAI-4B

Benchmarks: replace alloma-3B/1B with Llama-3.1-8B-Instruct-Uz and Mistral-7B-Instruct-Uz

Fix default generation and EOS configuration

Improve inference defaults and examples

Audit and correct training hyperparameters

Correct training loss description

Remove Cyrillic comparison from tokenizer chart

Remove Cyrillic tokenizer comparison

Remove model lineage from card

Remove checkpoint selection section

Add alloma-8B benchmark comparison

Clarify held-out-source fertility methodology

Replace in-domain fertility claim with large out-of-domain evaluation

Release NeuronAI-4B card, benchmark runner, visuals, and non-commercial license

Add files using upload-large-folder tool

initial commit