mistral-hackaton-2026/ecotopia-citizens-8b-merged

fix: PreTrainedTokenizerFast

fix: mistral arch for HF compat

revert: restore LlamaTokenizerFast for A10G

revert: restore ministral arch for A10G endpoint

fix: remove token_type_ids from model inputs

fix: use mistral arch with fixed tokenizer config

fix: remove pip install, rely on config fixes instead

fix: use PreTrainedTokenizerFast (LlamaTokenizerFast not in old transformers)

fix: also upgrade tokenizers to match transformers

fix: pip install transformers>=4.48 at startup for Ministral support

revert: restore ministral arch (mistral rename broke weights)

fix: use mistral arch (compatible with older transformers)

fix: require transformers>=4.48 for Ministral support

fix: set tokenizer_class to LlamaTokenizerFast

Upload requirements.txt with huggingface_hub

Upload handler.py with huggingface_hub

Upload tokenizer

Upload MinistralForCausalLM

initial commit