this is a tiny 1.3M parameter model based on the GPT2 architecture that it has been trained only on the Bible
This model needs to be fine-tuned, because it lacks the ability to answer questions
Training:
dataset size: 4.3Mb, 4.3M tokens
training hardware: Intel i7-5500u, AVX2 CPU only, Fedora linux server
training time: ~28h
current loss: 1.43
current training rate: 1.6*10^-4
requirements:
GPU required: no
RAM requirements: negligeable
Sample output(from an empty prompt):
there said, and he
dilinest of the mest of the beasts of miney of side, the
between shall not be the things shall days, which he we neig the
stander.
22:14 And his shall be against the priest of the LORD, saying, What them
was the with would the sanctument, and shall before him.
32:15 For therefore the slaid, and shall by from the hand of the
land of the priest.
22:8 And in the priest whereince and was breach the hand of the
cornament of his house against them: and to the hand of the bodin
of the people
- Downloads last month
- 12
4-bit
8-bit
16-bit
32-bit