search

yuntaeyang / TelME

16 stars 3 forks source link

Running the source code cannot reach the score #4

Open chenyuyao1127 opened 4 months ago

chenyuyao1127 commented 4 months ago

I ran the weight model you provided and indeed achieved this score, but I don't know why. According to this source code, both the teacher model and the student model reached the score in the paper, but the final fusion score did not reach it. Can you tell me if it is due to the experimental settings or other reasons?

yuntaeyang commented 4 months ago

I'm not sure what your experimental setting is like. However, I guess the performance may be different due to various problems such as hardware type, torch version, etc. However, we checked the performance for fusion result in Colab setting in the link below and got the same result in MELD. Please check the Closed issue link below. https://github.com/yuntaeyang/TelME/issues/3