File size: 1,032 Bytes
2de81a2 d0dda7a 597d3b2 8a7481d 14d6f0a 8a7481d f267949 14d6f0a 4da900a eefa152 4da900a 388d851 4da900a 7fc3159 4da900a 7fc3159 388d851 4da900a f267949 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 |
Update (4/1): Added ggml for Cuda model
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.
Cuda info (use this one):
Command:
CUDA_VISIBLE_DEVICES=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
Prev. info
Quantized on GPTQ-for-LLaMa commit 5955e9c67d9bfe8a8144ffbe853c2769f1e87cdd
GPTQ 4bit quantization of: https://huggingface.co/chavinlo/gpt4-x-alpaca
Note: This was quantized with this branch of GPTQ-for-LLaMA: https://github.com/qwopqwop200/GPTQ-for-LLaMa/tree/triton
Because of this, it appears to be incompatible with Oobabooga at the moment. Stay tuned?
Command:
CUDA_VISIBLE_DEVICES=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
|