ChaiVerse is the community's one-stop repository for large language model training, deployment and LIVE USERS evaluation package. Train models and win prizes!
To train a LLaMA7b model and push it to huggingface it's just 5 lines of code ๐ฅณ
import chaiverse as cv
dataset = cv.load_dataset('ChaiML/davinci_150_examples', 'chatml')
model = cv.LLaMA7b()
model.fit(dataset, output_dir='./my_first_llama', num_epochs=1)
model_url = 'ChaiML/llama7b_dummy' # your huggingface URL
model.push_to_hub(model_url, private=True)
chai-guanaco login
Now submit the model you have just trained!
import chai_guanaco as chai
model_url = "ChaiML/llama7b_dummy"
generation_params = {
'temperature': 1.0,
'repetition_penalty': 1.13,
'top_p': 0.2,
"top_k": 40,
"stopping_words": ['\n'],
"presence_penalty": 1.5,
"frequency_penalty": 1.5
}
submission_parameters = {'model_repo': model_url, 'generation_params': generation_params, 'model_name': 'my-awesome-llama'}
submitter = chai.ModelSubmitter()
submission_id = submitter.submit(submission_parameters)