llama-fridman

meta-llama/Llama-2-7b trained on ~350 episodes of the Lex Friedman podcast (Lex=Assistant), QLoRA, ChatML

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_path="g-ronimo/llama-fridman"

model = AutoModelForCausalLM.from_pretrained(
    model_path, 
    torch_dtype=torch.bfloat16,
    )
tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=True, legacy=False)    # fast tokenizer 

# sampling parameters: llama-precise
gen_config = {
    "temperature": 0.7,
    "top_p": 0.1,
    "repetition_penalty": 1.18,
    "top_k": 40,
    "do_sample": True,
    "max_new_tokens": 300,
}

messages = [
    {"role": "user", "content": "Good morning, I am Mark Zuckerberg"},
    {"role": "assistant", "content": "The founder of Meta"},
    {"role": "user", "content": "Yes exactly! And the future of AI"}
]

prompt=tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
prompt_tokenized=tokenizer(prompt,  return_tensors="pt", add_special_tokens=False).to("cuda")

output_ids = model.generate(**prompt_tokenized, **gen_config)
response=tokenizer.decode(output_ids[0])

print(response)

|im_start|>user
Good morning, I am Mark Zuckerberg<|im_end|> 
<|im_start|>assistant
The founder of Meta<|im_end|> 
<|im_start|>user
Yes exactly! And the future of AI<|im_end|> 
<|im_start|>assistant
Today we are here to talk about the metaverse. What is it? How do you see it evolving in the next decades? Let's start with some basics. What is the metaverse?<|im_end|>