graylab / iglm Goto Github PK

Dear IgLM Team,

I'm excited to try out the model, but ran into the following error when running the README examples for iglm_infill, iglm_generate, and iglm_evaluate, as well as in the Python code example.

CUDA error: CUBLAS_STATUS_INVALID_VALUE when calling cublasSgemmStridedBatched( handle, opa, opb, m, n, k, &alpha, a, lda, stridea, b, ldb, strideb, &beta, c, ldc, stridec, num_batches)

Environment information:
NVIDIA-SMI 470.94 Driver Version: 470.94 CUDA Version: 11.4
Cuda compilation tools, release 11.4, V11.4.152
Build cuda_11.4.r11.4/compiler.30521435_0
Python 3.8
cudatoolkit 11.3.1
nvidia-cudnn-cu11 8.5.0.96

Thanks for taking a look!

Best,
Chloe

Will the models and vocab.txt be made available, as required for installing IgLM from github?

Hello authors, it's an excellent work. I would like to use the same dataset to train GPT. I believe that the data in your article can become the standard data in this field in the future. Could you please provide the data used in the article? It can be done through email or Google Cloud Drive, etc.

If you are worried that the data is too large and inconvenient to upload, it is also great to provide scripts for data processing.

Thank you very much.

The example code of the full sequence log-likelihood calculation sets the infill_range but the value is not assigned.

log_likelihood = iglm.log_likelihood( sequence, chain_token, species_token, infill_range=infill_range, )

Is infill_range=infill_range can be safely deleted for the full sequence calculation?

I tried running the CDR redesign example you provide by downloading the fasta file for 1JPT from PDB however the file format does not appear to be compatible with the expectations of IgLM. Could you provide an explicit example of what the input files should look like please? This would help with accessibility.

Thanks for providing the code & in general the interface is very usable.

Hello,

The log-likelihood calculation for Heavy and Light chains is working great. Thank you for that! On that note, I was wondering if it would also be possible to calculate the log-likelihood values for the Kappa and Lambda chains.

Thanks!

how to set mutliple infill_range