Setup ORT for Text and Images

[HAKSOAT]

Main Changes

This pull requests made the following changes:

• Removed dependence on FastEmbed to use ORT solely
• Added support for ClipViT Text model
• Implements Preprocessors and Postprocessors for Image inputs
• Implements Preprocessors and Postprocessors for Text inputs
• Brushed up the Python demo file

On removing dependence on FastEmbed

This is because ORT was used in the implementation for ClipViT, meaning other models could as well just use ORT. The reason for using ORT is because FastEmbed takes away control on how preprocessing works from us, taking raw texts and file names as inputs. Fastembed uses ORT under the hood and a lot of their utility methods are private, so can't be called directly.

[github-actions]

Test Results

153 tests   153 ✅  2m 39s ⏱️
  8 suites    0 💤
  2 files      0 ❌

Results for commit 4f7d123.

♻️ This comment has been updated with latest results.

[github-actions]

Benchmark Results

group                                                        main                                   pr
-----                                                        ----                                   --
store_batch_insertion_without_predicates/size_100            1.00    823.5±5.88µs        ? ?/sec    1.00    825.4±4.85µs        ? ?/sec
store_batch_insertion_without_predicates/size_1000           1.02      9.4±0.13ms        ? ?/sec    1.00      9.2±0.08ms        ? ?/sec
store_batch_insertion_without_predicates/size_10000          1.10     94.7±0.66ms        ? ?/sec    1.00     85.9±3.67ms        ? ?/sec
store_batch_insertion_without_predicates/size_100000         1.02   855.9±72.67ms        ? ?/sec    1.00   838.0±76.24ms        ? ?/sec
store_retrieval_no_condition/size_100                        1.00      2.6±0.05ms        ? ?/sec    1.04      2.7±0.06ms        ? ?/sec
store_retrieval_no_condition/size_1000                       1.00     18.7±0.13ms        ? ?/sec    1.04     19.4±0.32ms        ? ?/sec
store_retrieval_no_condition/size_10000                      1.00    186.0±1.60ms        ? ?/sec    1.03    192.5±3.27ms        ? ?/sec
store_retrieval_no_condition/size_100000                     1.00  1893.7±30.97ms        ? ?/sec    1.01  1918.0±20.76ms        ? ?/sec
store_retrieval_non_linear_kdtree/size_100                   1.00      2.8±0.02ms        ? ?/sec    1.03      2.8±0.03ms        ? ?/sec
store_retrieval_non_linear_kdtree/size_1000                  1.00     20.7±0.25ms        ? ?/sec    1.02     21.1±0.19ms        ? ?/sec
store_retrieval_non_linear_kdtree/size_10000                 1.00    206.3±3.98ms        ? ?/sec    1.01    208.9±2.44ms        ? ?/sec
store_retrieval_non_linear_kdtree/size_100000                1.00       2.1±0.03s        ? ?/sec    1.00       2.1±0.03s        ? ?/sec
store_sequential_insertion_without_predicates/size_100       1.01  1715.6±32.70µs        ? ?/sec    1.00  1698.1±35.04µs        ? ?/sec
store_sequential_insertion_without_predicates/size_1000      1.00     17.5±0.35ms        ? ?/sec    1.04     18.2±0.42ms        ? ?/sec
store_sequential_insertion_without_predicates/size_10000     1.00    179.4±2.77ms        ? ?/sec    1.01    181.8±2.92ms        ? ?/sec
store_sequential_insertion_without_predicates/size_100000    1.00  1693.0±37.25ms        ? ?/sec    1.10  1858.7±28.70ms        ? ?/sec

[deven96]

Switching to using ModelProcessing and SkipProcessing