Steveluo
Steveluo
> Hey,bro.How did you solve it?
where is the document of metrics
for utterance in dialog["utterances"]: history = utterance["history"][-(2*args.max_history+1):] for j, candidate in enumerate(utterance["candidates"][-num_candidates:]): lm_labels = bool(j == num_candidates-1) instance = build_input_from_segments(persona, history, candidate, tokenizer, lm_labels) **Hello, can you teach me how...
This error occurred when I run 'run_evaluate_ppl_fused.py'. The possiable reason is that the avg_nll_list is null. In this picture, the list ' lm_labels_shifted ' I printed doesnot contain chitchat_token_id.
And when I run this code, the loop stop, maybe all tensor_dataset[0][batch_idx][1] donnot contain chitchat_token_id.
Another thing I want to ask Mr.Tom for is that how to improve the fused model. In my opinion, GPT2DoubleHeadsModel can't be replaced. And the loss function cannot also be...
> GPT2 is no longer the state of the art backbone. I would suggest instruction tuning + large language models Thank you for your advice. And why did the error...