cuVS docs updates for release

[cjnolet]

No description provided.

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 70.21%. Comparing base (1912355) to head (4c7c175).
Report is 3 commits behind head on branch-24.06.

Additional details and impacted files

@@              Coverage Diff              @@
##           branch-24.06     #161   +/-   ##
=============================================
  Coverage         70.21%   70.21%           
=============================================
  Files                11       11           
  Lines                47       47           
=============================================
  Hits                 33       33           
  Misses               14       14           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[cjnolet]

/merge

[atbraoy]

Hi Corey (@cjnolet), first thank you all for the great effort.
We are trying to use RAFT to take advantages of its acceleration. We just started and are trying to execute the notebook "VectorSearch_QuestionRetrieval.ipynb" on Azure instant (running the notebook on Databricks, 4 GPUs 'Tesla V100-PCIE-16GB' with enough memory). We are getting the error below when executing the function 'search_cuvs_pq(query="Who was Grace Hopper?")'. Note, we getting same error with 'pylibraft'

TypeError: The like argument must be an array-like that implements the __array_function__ protocol.
File , line 9
5 print("Original dataset: {:.1f} MB".format(original_mem / 2**20))
7 print("Memory saved: {:.1f}%".format(100 * (1 - pq_index_mem / original_mem)))
----> 9 search_raft_pq(query="Who was Grace Hopper?")
File /local_disk0/.ephemeral_nfs/envs/pythonEnv-aae71dcc-9f46-4c89-8451-4501a1d9016d/lib/python3.11/site-packages/pylibraft/common/device_ndarray.py:174, in device_ndarray.copy_to_host(self)
169 def copy_to_host(self):
170 """
171 Returns a new numpy.ndarray object on host with the current contents of
172 this device_ndarray
173 """
--> 174 ret = np.frombuffer(
175 self.device_buffer_.tobytes(),
176 dtype=self.dtype,
177 like=self.ndarray_,
178 ).astype(self.dtype)
179 ret = np.lib.stride_tricks.as_strided(ret, self.shape, self.strides)
180 return ret

And below is the terminal error, no Databricks notebook used in this case, just executing directly from a python file:

Traceback (most recent call last):
File "/Workspace/Users/cagra_test.py", line 78, in
search_raft_pq(query="Who was Grace Hopper?")
File "/Workspace/Users/cagra_test.py", line 61, in search_raft_pq
hits = ivf_pq.search(search_params, pq_index, question_embedding[None], top_k)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "handle.pyx", line 225, in pylibraft.common.handle.auto_sync_handle.wrapper
File "/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.11/site-packages/pylibraft/common/outputs.py", line 88, in wrapper
return tuple(conv(ret_value))
^^^^^^^^^^^^^^^^^^^^^^
File "/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.11/site-packages/pylibraft/common/outputs.py", line 71, in conv
yield convert_to_cai_type(i)
^^^^^^^^^^^^^^^^^^^^^^
File "/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.11/site-packages/pylibraft/common/outputs.py", line 57, in convert_to_cai_type
return output_as_(device_ndarray)
^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/Workspace/Users/cagra_test.py", line 9, in
pylibraft.config.set_output_as(lambda device_ndarray: device_ndarray.copy_to_host())
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/local_disk0/.ephemeral_nfs/cluster_libraries/python/lib/python3.11/site-packages/pylibraft/common/device_ndarray.py", line 173, in copy_to_host
ret = np.frombuffer(
^^^^^^^^^^^^^^
TypeError: The like argument must be an array-like that implements the __array_function__ protocol.

I will appreciate your feedback!

Note: @cjnolet, our org has a direct connection with nvidia, if you would like to connect from there please dm me!