poedator/opt-125m-bnb-4bit

facebook/OPT-125m quantized using bitsandbytes 4-bit NF4 quantization. All license matters are set based on the underlying facebook/OPT-125m model.

This model is work in progress, use it for testing https://github.com/huggingface/transformers/pull/26037 pull request only.

known issue: safetensors deleted linked "quant_map" / "nested_quant_map" items.