from transformers import AutoModelForCausalLM, AutoTokenizer

model_path = "mrcuddle/Tiny-DarkLlama-Chat"

tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(
    model_path,
    device_map="auto",
    torch_dtype='auto'
).eval()

# Prompt content: "hi"
messages = [
    {"role": "user", "content": "hi"}
]

input_ids = tokenizer.apply_chat_template(conversation=messages, tokenize=True, add_generation_prompt=True, return_tensors='pt')
output_ids = model.generate(input_ids.to('cuda'))
response = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)

# Model response: "Hello! How can I assist you today?"
print(response)

Datasets used in training:

ChaoticNeutrals/Synthetic-Dark-RP
ChaoticNeutrals/Synthetic-RP
ChaoticNeutrals/Luminous_Opus
NobodyExistsOnTheInternet/ToxicQAFinal

Eval

huggingface (pretrained=mrcuddle/tiny-darkllama-chat), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: 16

Tasks	Version	Filter	Metric		Value		Stderr
hellaswag	1	none	acc	↑	0.4659	±	0.0050
		none	acc_norm	↑	0.6044	±	0.0049
lambada_openai	1	none	acc	↑	0.6101	±	0.0068
		none	perplexity	↓	5.9720	±	0.1591