-
Notifications
You must be signed in to change notification settings - Fork 0
Faiss cuVS details
The use_cuvs=True
flag in GPU index configurations can be used to enable cuVS imeplementations for an index. Note that this is automatically set to True for supported index types if -DFAISS_ENABLE_CUVS=ON
is set while building Faiss from source or with the faiss-gpu-cuvs
conda package.
Building a GPU index:
>>> import faiss
>>> res = faiss.StandardGpuResources()
>>> co = faiss.GpuIndexIVFFlatConfig()
>>> co.use_cuvs = True
>>> index_gpu = faiss.GpuIndexIVFFlat(res, ncols, nlist, faiss.METRIC_L2, co)
Once the index has been initialized with this config, all subsequent operations on the index, such as add
and search
can be done using the function calls as regular GPU Faiss.
Cloning a CPU index:
>>> co = faiss.GpuClonerOptions()
>>> co.use_cuvs = True
>>> index_gpu = faiss.index_cpu_to_gpu(res, 0, index_cpu, co)
- TODO
- TODO
Faiss building blocks: clustering, PCA, quantization
Index IO, cloning and hyper parameter tuning
GPU Faiss + NVIDIA cuVS - Overview
GPU Faiss + NVIDIA cuVS - Usage and Performance
Threads and asynchronous calls
Inverted list objects and scanners
Indexes that do not fit in RAM
Brute force search without an index
Fast accumulation of PQ and AQ codes (FastScan)
Setting search parameters for one query
Binary hashing index benchmark