Guys this model its my respect for meme about Le Chaton Fat model from Mistral AI, surely benchmarks not real like not real this model or any Le Chaton Fat models
1from transformers import AutoModelForCausalLM, AutoTokenizer
2import torch
3tok = AutoTokenizer.from_pretrained("squ11z1/LeChatonFat")
4model = AutoModelForCausalLM.from_pretrained("squ11z1/LeChatonFat", dtype=torch.bfloat16, device_map="cuda")
5msgs = [{"role": "user", "content": "Who are you?"}]
6ids = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt").to("cuda")
7print(tok.decode(model.generate(ids, max_new_tokens=128)[0], skip_special_tokens=True))
Inherits the base model's license (Apache-2.0).