torch module missing in venv

[estolle]

Hi

I was just trying Fantasia-Lite but running the test, i encountered a similar error as the other person (the other issue about Numpy module missing in the venv). But this time torch is missing.

./scripts/minimal_pipeline.sh fasta_test/test.fa

...
Computed embeddings for 33 sequences with prot_t5
All embeddings saved to /scratch/progz/FANTASIA-Lite/outputs_20260603_160547/tmp/chunk_embeddings/prot_t5_test_embeddings.npz (NumPy .npz format)      
No sequences skipped. Wrote empty report to /scratch/progz/FANTASIA-Lite/outputs_20260603_160547/tmp/failures/prot_t5_test_failures.csv.
Successfully embedded 33 sequence(s) across 1 chunk(s).
Running annotation lookup once for merged model group prot_t5
Traceback (most recent call last):
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_no_db_gpu.py", line 30, in <module>
    import torch
ModuleNotFoundError: No module named 'torch'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_pipeline.py", line 1422, in <module>
    sys.exit(main())
             ^^^^^^
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_pipeline.py", line 1410, in main
    run_pipeline(config, argv=argv)
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_pipeline.py", line 1348, in run_pipeline
    stats = run_chunk_pipeline(config, venv_python, device)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_pipeline.py", line 1270, in run_chunk_pipeline
    run_lookup_in_process(lookup_conf, use_gpu_lookup=use_gpu_lookup)
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_pipeline.py", line 629, in run_lookup_in_process
    from fantasia_no_db_gpu import EmbeddingLookUpLocalGPU
  File "/scratch/progz/FANTASIA-Lite/src/fantasia_no_db_gpu.py", line 32, in <module>
    raise ImportError(
ImportError: PyTorch is required for fantasia_no_db_gpu.py. Install torch with CUDA support

[estolle]

I got it to work by switching on a conda environment which has pytorch installed. but ideally it should just be in the fantasia venv