MercuryDemo

Results 3 issues of MercuryDemo

During the process of submitting a pull request to solve issue 332, I found that pong_game.py has many basic logical errors that make it almost impossible to run properly, and...

Solve the Issue #332 Corrected logical errors and coding errors in Pon_Game.py 1. Fix variable name error 2. Fix paddle position errors, boundary errors, up and down movement errors 3....

https://github.com/huggingface/trl/blob/a57e75967c2b787f42f4e402ed7ca23cd9bad9a9/examples/notebooks/gpt2-sentiment.ipynb#L383 The intention of this line of code is to separate the new tokens generated by the model(i.e., response) from the prmopt+response. But since gen_len is used as max_new_tokens,len(response)