sij/khoj
1
0
Fork 0
mirror of https://github.com/khoj-ai/khoj.git synced 2024-12-20 19:37:45 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3
1446 commits 10 branches 114 tags 136 MiB
Commit graph

1446 commits

This branch
This branch
All branches
Author SHA1 Message Date
Debanjum Singh Solanky
c35c6fb0b3 Reuse asymmetric.setup & input validation from asymmetric & org_to_jsonl 
Create asymmetric.setup method to
  - initialize model
  - generate compressed jsonl
  - compute embeddings

put input_files, input_file_filter validation in org_to_jsonl for
reuse in main.py, asymmetic.py
 2021-08-17 00:45:40 -07:00
Debanjum Singh Solanky
02a84df37a Update state vars after regeneration. Minimize time app in inconsistent state 2021-08-16 23:47:33 -07:00
Debanjum Singh Solanky
0509854e14 Replace README.md with README.org. Can be used as notes for testing 2021-08-16 20:00:05 -07:00
Debanjum Singh Solanky
79aff85fcb Update Readme. No separate SETUP step required. Simpler RUN step 
- Setup now happens on first run of application
- Embeddings can now be regenerated without killing app by calling API
 2021-08-16 19:24:04 -07:00
Debanjum Singh Solanky
95bf26a7f2 Set verbosity commandline parameters default value to 0 2021-08-16 19:16:29 -07:00
Debanjum Singh Solanky
04a9a6d62f Expose API endpoint to (re-)generate embeddings from latest notes 
- Provides mechanism to update notes from within application
  - Instead of having to pass the same arguments multiple times
    Pass it once (or rely on defaults when possible) and let app keep
    state and location of intermediary files

- Allows user to not have to deal with the internals of the application
  - E.g user doesn't have to specify the jsonl.gz or embeddings file path
    The app will still put those files in a default location
  - The user doesn't have to run the generation from the commandline
    as a separate step
 2021-08-16 18:52:38 -07:00
Debanjum Singh Solanky
1c00c33e73 Improve debug output from org_to_jsonl.py script 2021-08-16 18:50:29 -07:00
Debanjum Singh Solanky
2a57156428 Fix org_to_jsonl. Use passed args not global variables in methods. Fix orgnode import 2021-08-16 17:37:44 -07:00
Debanjum Singh Solanky
66238004d8 Use verbosity level instead of bool across application 
For consistent, more granular verbosity controls across app
Allows user to increase verbosity by passing -vvv flags passed to main.py
 2021-08-16 17:15:41 -07:00
Debanjum Singh Solanky
adbf157deb Remove usage of the closure to search_notes as it's not required 2021-08-16 16:52:48 -07:00
Debanjum Singh Solanky
649e5d1327 Allow reuse of get_absolute_path, is_none_or_empty methods 
- Move them to utils.helper.py for reuse
- Import those modules where required
- Delete duplicate methods defined in org_to_jsonl.py, asymmetric.py
 2021-08-16 16:33:43 -07:00
Debanjum Singh Solanky
9703afb814 Rename search_types to search_type to standardize to singular naming 
Using singular names for other directories in application already
 - processor instead of processors
 - interface instead of interfaces
 2021-08-16 16:31:30 -07:00
Debanjum Singh Solanky
19d6678eb1 Allow importing org-to-jsonl as module for reuse 
To allow importing org-to-jsonl as module
  - Wrap code in __main__ into a org-to-jsonl method
  - Rename processor/org-mode to processor/org_mode
  - Add __init__.py to processor directory
 2021-08-16 16:31:30 -07:00
Debanjum Singh Solanky
5f8221f77e Remove unused verbose argument to collate_results method 2021-08-16 13:54:41 -07:00
Debanjum Singh Solanky
85bf15628d Use better cmdline argument names. Drop unneeded no-compress argument 
Can infer to compress or not via the output_file suffix
 2021-08-16 13:49:39 -07:00
Debanjum Singh Solanky
d9f60c00bf Warn if any input files to org-to-json are potentially non org-mode files 
That is, if the file paths in the input set don't end with .org
 2021-08-16 13:49:39 -07:00
Debanjum Singh Solanky
3aa0c30fee Use absolute file path to open files in org-to-jsonl.py, asymmetric.py 
Exit script if neither org_files, org_file_filter is present
 2021-08-16 13:49:39 -07:00
Debanjum Singh Solanky
e773611558 Remove unused jsonl_file argument from convert_org_entries_to_jsonl 2021-08-16 13:49:35 -07:00
Debanjum Singh Solanky
8b29e272d3 Standardize interface, better default args for org-to-json.py script 
- Remove non-standard, unnecessary argument for org-directory
  Pass path each file in org-files and org-files-filter argument directly
- Allow shorthand -i, -o for input files, output files
- Default to compress, unless user explicitly specifies not to
 2021-08-16 11:29:08 -07:00
Debanjum Singh Solanky
7547e90745 Minor doc updates after merging emacs package with main repository 2021-08-16 02:02:26 -07:00
Debanjum Singh Solanky
ec157ea0ff Add Emacs interface to semantic-search directly to main repository 
Too much overhead to maintain multiple repositories, especially when
the Emacs library for semantic-search is a single file.

Import Readme from the emacs-semantic-search repository too
 2021-08-16 01:27:46 -07:00
Debanjum Singh Solanky
dcf7b2d04f Remove requirements.txt for now as virtualenv setup doesn't work 
Haven't gotten it to work on Mac or Ubuntu. Remove to avoid confusion
for now. Application depends on miniconda for now
 2021-08-16 00:15:10 -07:00
Debanjum Singh Solanky
3b81fafa3e Use updated path to MiniLM bi-encoder model on hugging-face 2021-08-15 23:57:22 -07:00
Debanjum Singh Solanky
4839153086 Acknowledge ML models used for search. Simplify path used in commands 2021-08-15 23:56:18 -07:00
Debanjum Singh Solanky
c58c1d96aa Change default install directory to current, fix open file code 2021-08-15 23:01:55 -07:00
Debanjum Singh Solanky
ae15e429b5 Reduce indentation from 4 to 2 in Readme.md. 
Prevent everything looking like code blocks due to 4 space indentations
 2021-08-15 22:56:36 -07:00
Debanjum Singh Solanky
636b6195cc Add Readme, License. Update .gitignore 2021-08-15 22:52:37 -07:00
Debanjum Singh Solanky
354c541b62 Add org processor to generate compressed jsonl from org-mode files 
The corpus embeddings are generated from this compressed JSONL
using the specified transformer ML model
 2021-08-15 22:52:31 -07:00
Debanjum Singh Solanky
b74cb9a104 Move install.py to new utils dir as it's for cmdline ease of use only 2021-08-15 19:10:30 -07:00
Debanjum Singh Solanky
ec92f3e146 Move different search types into search_types directory 2021-08-15 19:09:50 -07:00
Debanjum Singh Solanky
4d681c86ec Update requirements.txt for users wanting to use pip install 2021-08-15 18:45:37 -07:00
Debanjum Singh Solanky
d75df54385 Create API interface for Semantic Search 
Use FastAPI, Uvicorn to create app with API endpoint at /search
Example Query: http://localhost:8000/?q="why sleep?"&t="notes'&n=5
 2021-08-15 18:11:48 -07:00
Debanjum Singh Solanky
e3088c8cf8 Create environment.yml to install prerequisites for app via conda 2021-08-15 17:48:38 -07:00
Debanjum Singh Solanky
660e6c3937 Add explicit filters to asymmetric search 
User can filter results to ones which include, exclude specified words
To show entries which include, exclude specific words, user should prepend
a '+', '-' before the word. E.g "+hello -bye"
 2021-08-15 17:48:38 -07:00
Debanjum Singh Solanky
91a2c598fe Resolve paths to absolute paths once. Use pathlib glob directly 2021-08-09 00:39:33 -07:00
Debanjum Singh Solanky
ca0a22f4dd Search for images similar to query image provided by the user 
Example user passes path to an image in query. e.g ~/Pictures/photo.jpg
The script should return images in images_embedding most similar to
the query image
 2021-08-09 00:21:02 -07:00
Debanjum Singh Solanky
00d0065c5b Allow user to search images via text queries 2021-08-08 23:02:30 -07:00
Debanjum Singh Solanky
181cab89d2 Ignore Title Notes i.e notes with just headings from compute 2021-08-04 21:30:09 -07:00
Debanjum Singh Solanky
d6d7b9d6a8 Make installed script executable. Minor clean-up of duplicate code 2021-08-04 18:29:20 -07:00
Debanjum Singh Solanky
2eb029a7b0 Create script to install semantic-search as a program 2021-08-02 00:29:09 -07:00
Debanjum Singh Solanky
13d5100ce6 Rename script similarity to symmetric 2021-07-31 20:37:07 -07:00
Debanjum Singh Solanky
ad7e90bec3 Modularize script, provide cmdline control, improve results rendering 2021-07-31 17:13:39 -07:00
Debanjum Singh Solanky
eb03f57917 Save, Load Embeddings to/from file to speed up script load time 2021-07-31 10:13:41 -07:00
Debanjum Singh Solanky
0914f284bb Re-rank using cross encoder to get even more relevant results 
The cross encoder re-ranked results are much better for more distant queries.
It does take more time with the cross-encoder re-ranking but it seems
worth it to get more relevant results
 2021-07-31 03:10:44 -07:00
Debanjum Singh Solanky
9864a2b551 Retrieve most relevant entries for a query using MSMarco based bi-encoder 
Returns best 3 results ranked by MSMarco based biencoder score of
query match to entries from org-mode notes
 2021-07-31 00:20:37 -07:00
debanjum
0ef5495701 Use Sentence Transformers to Encode, Query Schedule.org Headings 2021-04-04 04:53:03 -07:00