Dear Developer Team,

I hope this message finds you well. I am reaching out to report an issue I encountered while working with your software, specifically when attempting to load a model that I trained using the training.py file. Upon executing the test code to utilize the trained model, I encountered an UnpicklingError that halted the process.

To provide a detailed context, here is the error message that was generated:

UnpicklingError: A load persistent id instruction was encountered, but no persistent_load function was specified.

This error occurred during the following operation in the code:

model = CaLM(weights_file='/home/hugeng/39-CaLM/production-run/latest-0.ckpt')

Thank you for your time and assistance. I look forward to your response and any suggestions you may offer to resolve this issue.

Best regards,

Geng Hu

Hello! I noticed that in the data you provided, some sequences do not begin with "ATG", for example, 'TTGAAAAGAAAAGCCAGTATCATGTTTGTCCATCAAGACAAGTACGAAGAATACAAACAGCGGCATGATGACATTTGGCCTGAGATGGCAGAAGCACTCAAAGCTCATGGAGCACACCATTATTCCATTTTTCTAGACGAGGAAACAGGCAGGCTTTTTGCATATTTAGAAATAGAGGATGAAGAGAAATGGAGAAAGATGGCGGACACGGAAGTTTGCCAAAGATGGTGGAAATCGATGGCGCCATTAATGAAAACAAATTCGGATTTCAGTCCTGTTGCGATAGATCTAAAGGAAGTTTTTTATTTGGATTGA'.
When tokenizing, should I discard the part before ATG and start from ATG, or should I just use the entire sequence as it is?
Similarly, when translating it into an amino acid sequence, should I translate the entire sequence directly or start translating from ATG?

Hi,

Thank you for your work.

I noticed this is trained on cDNA data, while the tokeniser seems to use RNA vocab (https://github.com/oxpig/CaLM/blob/main/calm/alphabet.py)

Can you please clarify the data preprocessing pipeline?

Great resource, however the repository does not contain a license, which makes it difficult to use/reuse.

Dear Developer Team,

I am writing to seek your assistance regarding an issue I encountered while attempting to run the code associated with your paper titled "Codon language embeddings provide strong signals for use in protein engineering". When running the 'training.py' file, I encountered the following traceback error:

Traceback (most recent call last):
  File "training.py", line 141, in <module>
    ckpt_path='production-run/latest-56000.ckpt')
  ...
  FileNotFoundError: [Errno 2] No such file or directory: 'training_data.fasta'

It seems that the 'training_data.fasta' file is not found, leading to this error. I would greatly appreciate it if you could provide some guidance on how to address this issue.

Thank you very much for your time and consideration. I look forward to your valuable guidance.

Sincerely,
Geng Hu

Can you please share your pytorch lightning model snapshot, so I can fine-tune a model on top of yours?

Currently you only share a weights file (calm_weights.pkl). I full pytorch lightning model snapshot allows me to run Trainer() from your location.

Hi and thanks again for the great work,

I will fix that in my local copy of the repo but as far as I see, the CaLM class in pretrained.py which is used for inference doesn't support a device argument and setting e.g. model.model.cuda() results in a conflict between the model device and the device of the tensors put in the forward method.

Hi @couteiral and thanks for sharing your work,

I tried to run python setup.py install but when I try to do from calm import CaLM the pytorch init triggers an error:
OSError: libnvJitLink.so.12: cannot open shared object file: No such file or directory

So I tried to build the env myself starting from a fresh conda environment, such that pytorch/cuda can be imported without issues. Then I install your repository within this env, all seems to go through and imports work well, however when I try to init the model, it downloads the weights but after that I get another cuda error:
ModuleNotFoundError: No module named 'fused_layer_norm_cuda'

Do you have any hints on how to finalise the installation please?