I am running into a weird output involving the generative text AI models. I expect that the huggingface models that state to be similar to GPT-3, GPT-4 or llama would perform similar to the webUI counterparts that can be found on openAI or poe. However when I use them I end up receiving very strange and nonsensical output. Is there something that I am missing that is resulting in the output?

code

model_name = "EleutherAI/gpt-j-6b" # chatgpt-3 model
tokenizer = GPT2Tokenizer.from_pretrained(model_name)
gpt_model = GPTNeoForCausalLM.from_pretrained(model_name)

max_length = 150
input_str = "Provide a list of five words that are similar to the phrase: 'Knights who say Nie'"
input_ids = tokenizer.encode(input_str, return_tensors='pt')
attention_mask = torch.ones(input_ids.shape, dtype=torch.long)
pad_token_id = tokenizer.eos_token_id

with torch.no_grad():
    outputs = gpt_model.generate(input_ids, max_length=max_length, attention_mask=attention_mask, pad_token_id=pad_token_id, do_sample=True)
generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(generated_text)

attempt 1 output

(Nie = 'no').
A:
I think the answer is:
Nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie, nie

attempt 2 output

Nie can be found in the name of some of the most famous kings in history. The name of another king, Nieun-neng, was given by one of the best-known Chinese historians, the historian Tang Sanzang. The word 'Nie-neng' literally means 'the son of an uncle.' That means that Nieun-neng was the only son of King Nieun's uncle, Nio-neng.
Some kings began the use of the royal name from the moment they turned seven years of age and took the title of a 'Prince.' As a young child

here is the output from Poe that claims to be the ChatGPT-3.5 Turbo model

1. Warriors who declare Nay
2. Paladins who shout Nay
3. Guardians who utter Nay
4. Defenders who proclaim Nay
5. Sentinels who cry Nay