marella
Please review: change chatdocs numpy version requirements to <1.23.0 for ROCM experimentation
Overview I'm attempting to use Radeon GPUs with chatdocs. What I've done below seems to work up to the point I ask a question. I get a segmentation fault now regardless if the chatdocs.yml is configured for "cuda" or not. Thanks.
Request: Please Change chatdocs numpy version requirements to <1.23.0 for ROCM experimentation
Why: To create a ROCM environment using docker for use with chatdocs
How: Using PyTorch Installation for ROCm Option 1, for a docker environment to install chatdocs
Behavior: The follow errors were noticed:
root@docker-container:/workspace# pip install chatdocs
<stuff>
ERROR: onnxruntime 1.15.1 has requirement numpy>=1.21.6, but you'll have numpy 1.18.5 which is incompatible.
ERROR: filelock 3.12.3 has requirement typing-extensions>=4.7.1; python_version < "3.11", but you'll have typing-extensions 4.7.0 which is incompatible.
ERROR: pandas 2.0.3 has requirement numpy>=1.20.3; python_version < "3.10", but you'll have numpy 1.18.5 which is incompatible.
ERROR: chromadb 0.3.29 has requirement numpy>=1.21.6, but you'll have numpy 1.18.5 which is incompatible.
ERROR: transformers 4.33.1 has requirement tokenizers!=0.11.3,<0.14,>=0.11.1, but you'll have tokenizers 0.14.0 which is incompatible.
The 'change':
root@docker-container:/workspace# pip install numpy==1.22.0 && \
pip install typing-extensions>=4.7.1 && \
pip install "tokenizers<0.14,>=0.11.1" --force-reinstall && \
pip install "pandas>=1.0.0,<2.0.0" --force-reinstall && \
pip install chatdocs && \
chatdocs download
root@docker-container:/app# python3 -c 'import torch; print(torch.cuda.is_available())'
True
root@docker-container:/app# cat chatdocs.yml
embeddings:
model_kwargs:
device: cuda
root@docker-container:/app# chatdocs add /app/documents/
<normal output until>
Type your query below and press Enter.
Type 'exit' or 'quit' or 'q' to exit the application.
Q: Configure the network interfaces and hostname on a rhel server
A:Segmentation fault
