Closed Goluck-Konuko closed 11 months ago
Hello, I'm curious about the results of your testing using the model lacking the TS block provided by the author. When I tried it out, my results were pretty close to those mentioned in the article. Not sure if I'm getting it wrong though.
Thank you, @Goluck-Konuko and @MrX0404 , for highlighting the issues. Indeed, we have included the TS-prompts on the decoder side. The embeddings can be found within the weights at g_s0.residual_group.prompt_embeddings, g_s2.residual_group.prompt_embeddings, g_s4.residual_group.prompt_embeddings, and g_s6.residual_group.prompt_embeddings, representing the TS-prompts.
Hi, Is there a reason for not including the TS-prompts on the decoder side for the task-specific inference models?