Downloads · 30 days
10
5% of all-time downloads
Dayara13/llm-course-hw1
llm-course-hw1 is a text generation model from Dayara13. Use it when you need the model to write or continue text.
The task was to train a LLM to make up funny jokes.
Downloads · 30 days
10
5% of all-time downloads
All-time downloads
208
Public
Parameters
29.5M
693 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors119 MB · 100%
How the weights are stored.
F3229.4M · 100%
From the Hugging Face model README
The task was to train a LLM to make up funny jokes.
During the task Byte-Level BPE tokenizer and GPT like model were trained.
Vocabulary of BPE tokenizer consists of 1024 tokens.
This model has been pushed to the Hub using the [PytorchModelHubMixin] with trained BPE tokenizer.
LLM consists of next blocks and parts:
During the training:
Also warm up with following lambda lr scheduler were used during the training of LLM.
<img src="lr.png" alt="" width="700" height="150">Here are below some examples of generated russian jokes, to be honest, there are pretty strange :)