umbra-scientia

We also observe this behavior in faiss-gpu 1.7.2. (using python 3.8 and cuda 11.3 installed via `conda -c pytorch`) ```python index = faiss.read_index(filename) index = faiss.index_cpu_to_gpu(faiss.StandardGpuResources(), gpu_id, index) index.add(...) ```...

Lets skip this PR and use this one instead: https://github.com/LAION-AI/Open-Assistant/pull/2829 The simple XOR format is too finicky and users are having problems with minor differences in HuggingFace and PyTorch versions....

This error appears to be the non-trivial one: ``` fast_marching_method.hpp: In constructor 'thinks::fast_marching_method::detail::VaryingSpeedEikonalSolverBase::VaryingSpeedEikonalSolverBase(const std::array&, const std::array&, const std::vector&)': fast_marching_method.hpp:2027:7: error: class 'thinks::fast_marching_method::detail::VaryingSpeedEikonalSolverBase' does not have any field named 'EikonalSolverBase' :...

If we could have a column for "URLs and references used" in each turn (and ideally a matching input field in the front-end), it would be useful for fine-tuning information...

**THIS IS A DANGEROUSLY BAD IDEA.** === Please, do not pursue this "synthetic CSAM" concept. Private companies quietly use datasets like these, thinking they are reasonable, without caring that basic...

I would take inspiration from earlier LAION success and lean more towards distributed pre-processing of training data. There is plenty of cost in things like the reinforcement learning rollouts of...

We can learn each of these: > - how do we decide that retrieval system should be activated > - how is the query generated > - how are the...