Selyam/gpt4-x-alpaca-13b-native-4bit-128g

Dataset is here (instruct): https://github.com/teknium1/GPTeacher Okay… Two different models now. One generated in the Triton branch, one generated in Cuda. Use the Cuda one for now unless the Triton branch becomes widely used. CUDAVISIBLEDEVICES=0 python llama.py ./models/chavinlo-gpt4-x-alpaca —wbits 4 —true-sequential —groupsize 128 —save gpt-x-alpaca-13b-native-4bit-128g-cuda.pt Quantized on GPTQ-for-LLaMa commit 5955e9c67d9bfe8a8144ffbe853c2769f1e87cdd GPTQ 4bit quantization of: https://huggingface.co/chavinlo/gpt4-x-alpaca Because of this, it appears to be incompatible with Oobabooga at the moment. Stay tuned? CUDAVISIBLEDEVICES=0 python llama.py ./models/chavinlo-gpt4-x-alpaca —wbits 4 —true-sequential —act-order —groupsize 128 —save gpt-x-alpaca-13b-native-4bit-128g.pt

Модальности:
Генерация текста

Задача: Генерация текста
Автор: Selyam
Теги: llama, text-generation-inference, endpoints_compatible
Лайков: 3  |  Загрузок: 40

Открыть на HuggingFace →

Описание основано на материалах HuggingFace. Перевод выполнен автоматически.