| # How to use me? | |
| ```python | |
| from transformers import AutoTokenizer, AutoModelForCausalLM | |
| import transformers | |
| import torch | |
| tokenizer = AutoTokenizer.from_pretrained("tiiuae/falcon_tokenizer") | |
| model = AutoModelForCausalLM.from_pretrained( | |
| "tiiuae/falcon-micro-self-instruct", | |
| trust_remote_code=True, | |
| torch_dtype=torch.bfloat16, | |
| use_auth_token="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", | |
| ) | |
| model.bfloat16() | |
| model.cuda() | |
| pipeline = transformers.pipeline("text-generation", model=model, tokenizer=tokenizer, device="cuda:0") | |
| sequences = pipeline( | |
| "What is your favourite dad joke?", | |
| max_length=200, | |
| do_sample=True, | |
| top_k=10, | |
| repetition_penalty=1.2, | |
| num_return_sequences=2, | |
| eos_token_id=tokenizer.eos_token_id, | |
| ) | |
| for seq in sequences: | |
| print(f"Result: {seq['generated_text']}") | |
| ``` | |
| There will be a warning that the model is not supported for generation, it can safely be ignore. | |