Model Description
gen_kwargs = {
"max_new_tokens": 100,
"top_k": 70,
"top_p": 0.8,
"do_sample": True,
"no_repeat_ngram_size": 2,
"bos_token_id": tokenizer.bos_token_id,
"eos_token_id": tokenizer.eos_token_id,
"pad_token_id": tokenizer.pad_token_id,
"temperature": 0.8,
"use_cache": True,
"repetition_penalty": 1.2,
"num_return_sequences": 1
}
device = torch.device("cuda") if torch.cuda.is_available() else torch.device("cpu")
ft = 'gpt-j-onlyk_v2'
tokenizer = AutoTokenizer.from_pretrained(ft)
model = AutoModelForCausalLM.from_pretrained(ft, torch_dtype=torch.float16, low_cpu_mem_usage=True)
model.to(device)
inp = '''Sophia, 29, a student, meets a male programmer Alex from India, who is 45 <|endoftext|>
Alex: How was your vacation? <|endoftext|> sofie: It was amazing! I went to the beach and it felt like paradise. What about you?
<|endoftext|> Alex: i'm good. Tell me a joke <|endoftext|> Sofie:'''
prepared = tokenizer.encode(inp, return_tensors='pt').to(model.device)
out = model.generate(input_ids=prepared, **gen_kwargs)
generated = tokenizer.decode(out[0])
