tinyllama-20M

---

Model trained on the TinyStories Dataset, replicating https://arxiv.org/abs/2305.07759, based on LLaMA architecture.

---

Hyperparams used to train this model:

        "batch_size": 32,
        "block_size": 256,
        "lr": 5e-4,
        "num_hidden_layers": 8,
        "num_attention_heads": 8,
        "hidden_size": 160,
        "dropout": 0.1,
        "weight_decay": 0.01,
        "epochs": 1,
        "eval_interval": 200,
        "eval_steps": 50,
        "vocab_size": 50257, 
        "warmup_tokens": 10000,
        "gradient_accumulation_steps": 16, 

---

EXAMPLE USAGE

  !pip install --quiet transformers 
  from transformers import AutoModelForCausalLM, AutoTokenizer
  from huggingface_hub import notebook_login, login
  import os

  #login to hf to check for llama access
  hf_token = os.getenv('HF_TOKEN')
  login(token=hf_token)

  model = AutoModelForCausalLM.from_pretrained('AnirudhRajagopalan1201/tinyllama-20M')
  tokenizer = AutoTokenizer.from_pretrained("meta-llama/Llama-2-7b-hf")
  prompt = "Lily likes cats and dogs. She asked her mom for a dog and her mom said no, so instead she asked"
  input_ids = tokenizer.encode(prompt, return_tensors="pt")
  output = model.generate(input_ids, temperature=0.1, max_length = 100, do_sample=True)
  output_text = tokenizer.decode(output[0], skip_special_tokens=True)
  print(output_text)