llm-grep
LLM plugin for matching text using semantic regular expressions. The high-level matching technique is loosely based on this paper. Matching is done in two passes: one using traditional regular expressions to narrow down candidate matches, and a second pass using an LLM to filter those candidates based on any semantic tags.
The pattern syntax is similar to traditional regular expressions (enclosed in {{ and }}), but adds semantic tags (enclosed in < and >) to indicate the type of concept being matched.
Installation
Install this plugin in the same environment as LLM.
llm install llm-grep
Usage
The plugin adds a new command, llm grep.
This command has an interface similar to the GNU grep command, but extends it with semantic matching capabilities, using an LLM as a matching oracle.
Input can be a standard file or stdin. Simple examples you can try:
# Match lines from a file
llm grep -e '^{{.*}}<about birds>$' notes.txt
# Read from standard input
cat recipes.txt | llm grep -e '^{{.*}}<baking related>$'
# Enable color, only output matched content, and use a custom model and prompt:
llm grep --color always -o -e '{{[A-Za-z0-9]+}}<outdoor activities>' --model gpt-4\
--prompt 'Answer yes or no. Query: {query} Text: {span}.' headlines.txt
# Slightly more specific capture (will not match names like 'sparrow')
llm grep -e '\\b{{[A-Z][a-z]+(?:\\s+[A-Z][a-z]+)?}}<bird species>\\b' bird_log.txt
The default prompt used is:
Does the following text satisfy the semantic concept described by the query?
Query: {query}
Text: {span}
Your answer should include "yes" or "no".
Development
To set up this plugin locally, first checkout the code. Then create a new virtual environment:
cd llm-grep
python3 -m venv venv
source venv/bin/activate
Now install the dependencies and test dependencies:
pip install -e '.[test]'
To run the tests:
python -m pytest
Future plans
- Currently, some of
grep's functionality is not implemented (e.g.-rfor recursive searching). Re-implementing these features is a losing battle, and will be deprioritized in favor of somehow hooking into (or wrapping around) existinggrepimplementations.
Metadata
Release files for llm-grep 0.1
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| llm_grep-0.1.tar.gz | 6.8 kB | Without provenance | Details for llm_grep-0.1.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| llm_grep-0.1-py3-none-any.whl | Python 3 | none | any | Details for llm_grep-0.1-py3-none-any.whl |
Total release size: 13.0 kB
Release files / llm_grep-0.1.tar.gz
| Download URL | llm_grep-0.1.tar.gz |
|---|---|
| Size | 6.8 kB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
3dd7d4ab1edb282cc0fc7af38cbeb33c42d9be0bcef72aeed401bbaf85b06091
|
|
BLAKE2b-256 checksum How to use checksums |
c142435f44768e9c1fe6c788a4f9427323b2482009dea2224adf5085c83f79aa
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.11.14
|
Release files / llm_grep-0.1-py3-none-any.whl
| Download URL | llm_grep-0.1-py3-none-any.whl |
|---|---|
| Size | 6.2 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
b97f338b875afaf0bac82acee3e08a71d6dccc1e873c54f9296a5e25abf0d5e2
|
|
BLAKE2b-256 checksum How to use checksums |
9f71e150af99bac74fddb6ad6b77dafbaabfd901632852dabb882076a6e19b16
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
No |
| Uploaded via |
twine/6.2.0 CPython/3.11.14
|
Security for llm-grep 0.1
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
No provenance found
None of the files in this release were uploaded with provenance metadata.
A note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.