Views
No views yet
1import torch
2from peft import PeftModel, PeftConfig
3from transformers import AutoModelForCausalLM, AutoTokenizer
4
5peft_model_id = "hackathon-somos-nlp-2023/bertin-gpt-j-6B-es-finetuned-salpaca"
6config = PeftConfig.from_pretrained(peft_model_id)
7model = AutoModelForCausalLM.from_pretrained(config.base_model_name_or_path, return_dict=True, load_in_8bit=True, device_map='auto')
8# tokenizer = AutoTokenizer.from_pretrained(config.base_model_name_or_path)
9tokenizer = AutoTokenizer.from_pretrained(peft_model_id)
10
11# Load the Lora model
12model = PeftModel.from_pretrained(model, peft_model_id)
13
14def gen_conversation(text):
15 text = "<SC>instruction: " + text + "\n "
16 batch = tokenizer(text, return_tensors='pt')
17 with torch.cuda.amp.autocast():
18 output_tokens = model.generate(**batch, max_new_tokens=256, eos_token_id=50258, early_stopping = True, temperature=.9)
19
20 print('\n\n', tokenizer.decode(output_tokens[0], skip_special_tokens=False))
21
22text = "hola"
23
24gen_conversation(text)
@misc {hackathon-somos-nlp-2023,
author = { {Edison Bejarano, Leonardo Bolaños, Alberto Ceballos, Santiago Pineda, Nicolay Potes} },
title = { SAlpaca },
year = 2023,
url = { https://huggingface.co/hackathon-somos-nlp-2023/bertin-gpt-j-6B-es-finetuned-salpaca }
publisher = { Hugging Face }
}