search

MAGICS-LAB / DNABERT_2

[ICLR 2024] DNABERT-2: Efficient Foundation Model and Benchmark for Multi-Species Genome
Apache License 2.0
212 stars 49 forks source link

Random issues still come up in use #95

Closed josecar24 closed 2 weeks ago

josecar24 commented 2 weeks ago

Hello Dr. ZHOU(@Zhihan1996),

For the random problem, I found the problem still exists for same DNA sequence if I reload the model by running the first code cell. For instance, If I run the second code cell twice in a row directly after the model is loaded. The output tensor will be consistent due to 'model = model.eval()' as you suggested. But if I run the first code cell again between running the second and third code cells, and the random problem will reappear (and not be solved with my addition of model.eval after model loaded in code cell one). Is that designed to be different outcome for operations between batches? Is there any possible solution to keep the output tensor from the same DNA sequence always the same please? Appreciate again!

截屏2024-06-17 下午12 51 41
GUMI-QXP commented 4 days ago

Hello, I have also encountered this problem. Have you resolved it?