Xiaohui Chen

latent loss backward problem

1

Hi, I notice that the err_g_lat is added to update the generator parameters, however, this loss is acquired when the x_fake is detached. I think it makes no contribution to...

https://github.com/andrewgordonwilson/bayesgan/blob/ca06613201e0882a21e53f18e71ecbff5eda7923/bgan.py#L277

commute time run() and runParrellelApproximation()

3

to reproduce: ``` b = nx.Graph() b.add_nodes_from([0,1,2,3]) b.add_weighted_edges_from([(0,1,1.),(1,2,0.1),(2,3,1.)]) b_nk = nk.nxadapter.nx2nk(b, 'weight') b_nk.indexEdges() reff01 = nx.resistance_distance(b,0,1,'weight',False) reff02 = nx.resistance_distance(b,0,2,'weight',False) reff03 = nx.resistance_distance(b,0,3,'weight',False) ctd = nk.distance.CommuteTimeDistance(b_nk, tol=0.01) ctd.run() ct01 = ctd.distance(0,1)**2/(2*b_nk.totalEdgeWeight())...

vqgan training

1

Hi, I really appreciate the effort you guys make toward reproducing MUSE. I was wondering in the trained VQGAN, which version of the LAION is used?

hi, I added a paper into your collections. thanks for maintaining such a great repository.

loss curve of SFT on vicuna-7b

6

Hi, I got a training curve like this, is it normal? Do you mind sharing your trainer_state.json? thx!

Reproducing video mme with qwen3-vl-instruct

23

Hi I got 71.9 on video mme when using vllm. Do you mind sharing the script / spec you use to obtained the report result? Thanks!