| Name | rellm JSON |
| Version |
0.0.5
JSON |
| download |
| home_page | |
| Summary | Get exact structure out of any language models with regular expressions. |
| upload_time | 2023-05-06 21:39:41 |
| maintainer | Matt Rickard |
| docs_url | None |
| author | Matt Rickard |
| requires_python | >=3.8,<4.0 |
| license | MIT |
| keywords |
llm
ai
prompt
large language models
gpt-3
chatgpt
|
| VCS |
 |
| bugtrack_url |
|
| requirements |
No requirements were recorded.
|
| Travis-CI |
No Travis.
|
| coveralls test coverage |
No coveralls.
|
# ReLLM
Regular Expressions for Language Model Completions.
> *Some people, when confronted with a problem, think
“I know, I'll use regular expressions.” Now they have two problems.*
Exact structure out of any language model completion with regular expressions.
Return specific syntactic structure (e.g. JSON or XML), or specific semantic structure (e.g. a date or a number), or even complete templates (e.g. a sentence with a blank to fill in).
How does it work? ReLLM filters non-matching tokens pre-generation. For each token, ReLLM tests every possible completion against a partial regex. For the potential completions that do not match the pattern, ReLLM masks the logits so that the language model does not generate them.
### Installation
```
pip install rellm
```
The preliminary results are interesting -- even for small models, constraining the token space with ReLLM can improve the quality of the completions. Not to mention the ability to more easily parse the output programmatically. Take a look at some of the [examples](examples).
```python
import regex
from transformers import AutoModelForCausalLM, AutoTokenizer
from rellm import complete_re
model = AutoModelForCausalLM.from_pretrained("gpt2")
tokenizer = AutoTokenizer.from_pretrained("gpt2")
prompt = "ReLLM, the best way to get structured data out of LLMs, is an acronym for "
pattern = regex.compile(r'Re[a-z]+ L[a-z]+ L[a-z]+ M[a-z]+')
output = complete_re(tokenizer=tokenizer,
model=model,
prompt=prompt,
pattern=pattern,
do_sample=True,
max_new_tokens=80)
print(output)
```
```
> Realized Logistic Logistics Model
```
## Examples using GPT2 (124 million parameters)
#
Using GPT2 (124m)
**Prompt**: ReLLM, the best way to get structured data out of LLMs, is an acronym for
**Pattern**: Re[a-z]+ L[a-z]+ L[a-z]+ M[a-z]+
**ReLLM**: Realized Logistic Logistics Model
**Without ReLLM**: Largest Largest Address Space (MELSP), which has its roots in the Internet network, at least when compared
#
**Prompt**: Return the first three letters of the alphabet in a json array:
**Pattern** \[\"[a-z]\", \"[a-z]\", \"[a-z]\"\]
**ReLLM**: ["a", "b", "c"]
**Without ReLLM**: { "index": 0, "id":"1", "description":"", "text": "[{ "id": 0, "name":
#
**Prompt**: Fill in the sentence with an interesting story about the dentist:
**Pattern**: Today I\'m going to the [a-z]+ to [a-z]+ because ([a-z]+ )*\.
**ReLLM**: Today I'm going to the dentist to see because it is a very important day for me
**Without ReLLM**: 'My family bought me an appointment with a dentist when I was 15. The dentist gave me one a year and then I was told on
#
**Prompt**: Is this a good demo?
**Pattern**: (Yes|No)
**ReLLM**: No.
**Without ReLLM**: I don't know, but this is amazing! Even more amazing is how the design can take place on a small stage that uses LEDs.
As
#
**Prompt**: Convert the date May 4, 2023 to the format mm/dd/yyyy:
**Pattern**: [0-9]{2}/[0-9]{2}/[0-9]{4}
**ReLLM**: 00/00/0045
**Without ReLLM**: mm:ss
A-Z, Z-A, W-H (0-9:9:19)
Z-R
#
**Prompt**: Jeff Dean is a
**Pattern** (Programmer|Computer Scientist|AGI)
**ReLLM**: Computer Scientist
**Without ReLLM**: former national basketball champion and a former professional basketball player. He currently serves as general counsel for the NCAA Office of the Vice President for Academic Affairs.
#
**Prompt**: I can eat
**Pattern**: [0-9]{1,10} [a-z]* of [a-z]*
**ReLLM**: 800 calories of coffee
**Without ReLLM**: iced coffee here on the west side and do this, so can you?"
"Why, I don't understand. What did you mean by
Raw data
{
"_id": null,
"home_page": "",
"name": "rellm",
"maintainer": "Matt Rickard",
"docs_url": null,
"requires_python": ">=3.8,<4.0",
"maintainer_email": "pypi@matt-rickard.com",
"keywords": "llm,ai,prompt,large language models,gpt-3,chatgpt",
"author": "Matt Rickard",
"author_email": "pypi@matt-rickard.com",
"download_url": "https://files.pythonhosted.org/packages/dc/fb/96db913168c2c770cc54b80781fc02d2c5c82a7a23e2089446a4a909fd6c/rellm-0.0.5.tar.gz",
"platform": null,
"description": "# ReLLM\nRegular Expressions for Language Model Completions.\n\n> *Some people, when confronted with a problem, think\n\u201cI know, I'll use regular expressions.\u201d Now they have two problems.*\n\nExact structure out of any language model completion with regular expressions.\n\nReturn specific syntactic structure (e.g. JSON or XML), or specific semantic structure (e.g. a date or a number), or even complete templates (e.g. a sentence with a blank to fill in).\n\nHow does it work? ReLLM filters non-matching tokens pre-generation. For each token, ReLLM tests every possible completion against a partial regex. For the potential completions that do not match the pattern, ReLLM masks the logits so that the language model does not generate them.\n\n### Installation\n```\npip install rellm\n```\n\nThe preliminary results are interesting -- even for small models, constraining the token space with ReLLM can improve the quality of the completions. Not to mention the ability to more easily parse the output programmatically. Take a look at some of the [examples](examples).\n\n```python\nimport regex\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\n\nfrom rellm import complete_re\n\nmodel = AutoModelForCausalLM.from_pretrained(\"gpt2\")\ntokenizer = AutoTokenizer.from_pretrained(\"gpt2\")\n\nprompt = \"ReLLM, the best way to get structured data out of LLMs, is an acronym for \"\npattern = regex.compile(r'Re[a-z]+ L[a-z]+ L[a-z]+ M[a-z]+')\noutput = complete_re(tokenizer=tokenizer, \n model=model, \n prompt=prompt,\n pattern=pattern,\n do_sample=True,\n max_new_tokens=80)\nprint(output)\n```\n\n```\n> Realized Logistic Logistics Model\n```\n\n\n## Examples using GPT2 (124 million parameters)\n\n#\n\nUsing GPT2 (124m)\n\n**Prompt**: ReLLM, the best way to get structured data out of LLMs, is an acronym for\n\n**Pattern**: Re[a-z]+ L[a-z]+ L[a-z]+ M[a-z]+\n\n**ReLLM**: Realized Logistic Logistics Model\n\n**Without ReLLM**: Largest Largest Address Space (MELSP), which has its roots in the Internet network, at least when compared\n#\n\n**Prompt**: Return the first three letters of the alphabet in a json array:\n\n**Pattern** \\[\\\"[a-z]\\\", \\\"[a-z]\\\", \\\"[a-z]\\\"\\]\n\n**ReLLM**: [\"a\", \"b\", \"c\"]\n\n**Without ReLLM**: { \"index\": 0, \"id\":\"1\", \"description\":\"\", \"text\": \"[{ \"id\": 0, \"name\":\n#\n**Prompt**: Fill in the sentence with an interesting story about the dentist:\n\n**Pattern**: Today I\\'m going to the [a-z]+ to [a-z]+ because ([a-z]+ )*\\.\n\n**ReLLM**: Today I'm going to the dentist to see because it is a very important day for me\n\n**Without ReLLM**: 'My family bought me an appointment with a dentist when I was 15. The dentist gave me one a year and then I was told on\n#\n\n**Prompt**: Is this a good demo?\n\n**Pattern**: (Yes|No)\n\n**ReLLM**: No.\n\n**Without ReLLM**: I don't know, but this is amazing! Even more amazing is how the design can take place on a small stage that uses LEDs.\nAs\n\n#\n\n**Prompt**: Convert the date May 4, 2023 to the format mm/dd/yyyy:\n\n**Pattern**: [0-9]{2}/[0-9]{2}/[0-9]{4}\n\n**ReLLM**: 00/00/0045\n\n**Without ReLLM**: mm:ss\n\nA-Z, Z-A, W-H (0-9:9:19)\n\nZ-R\n\n#\n\n**Prompt**: Jeff Dean is a\n\n**Pattern** (Programmer|Computer Scientist|AGI)\n\n**ReLLM**: Computer Scientist\n\n**Without ReLLM**: former national basketball champion and a former professional basketball player. He currently serves as general counsel for the NCAA Office of the Vice President for Academic Affairs.\n\n#\n\n**Prompt**: I can eat \n\n**Pattern**: [0-9]{1,10} [a-z]* of [a-z]*\n\n**ReLLM**: 800 calories of coffee\n\n**Without ReLLM**: iced coffee here on the west side and do this, so can you?\"\n\n\"Why, I don't understand. What did you mean by",
"bugtrack_url": null,
"license": "MIT",
"summary": "Get exact structure out of any language models with regular expressions.",
"version": "0.0.5",
"project_urls": {
"repository": "https://github.com/r2d4/rellm"
},
"split_keywords": [
"llm",
"ai",
"prompt",
"large language models",
"gpt-3",
"chatgpt"
],
"urls": [
{
"comment_text": "",
"digests": {
"blake2b_256": "975531d0a7e22748eb8c4c7beff947ba030508915e053593e1672d60a8483dc7",
"md5": "e4f0e343caf52a5f7919ccebd35e5d73",
"sha256": "e95e1c3587d443251f3a1153174a5dc4310197c07e7b3242a1b15f9c669870fe"
},
"downloads": -1,
"filename": "rellm-0.0.5-py3-none-any.whl",
"has_sig": false,
"md5_digest": "e4f0e343caf52a5f7919ccebd35e5d73",
"packagetype": "bdist_wheel",
"python_version": "py3",
"requires_python": ">=3.8,<4.0",
"size": 5778,
"upload_time": "2023-05-06T21:39:39",
"upload_time_iso_8601": "2023-05-06T21:39:39.820855Z",
"url": "https://files.pythonhosted.org/packages/97/55/31d0a7e22748eb8c4c7beff947ba030508915e053593e1672d60a8483dc7/rellm-0.0.5-py3-none-any.whl",
"yanked": false,
"yanked_reason": null
},
{
"comment_text": "",
"digests": {
"blake2b_256": "dcfb96db913168c2c770cc54b80781fc02d2c5c82a7a23e2089446a4a909fd6c",
"md5": "76ee8b73a64bf8e2c7b5077cee9e9a78",
"sha256": "c357eea1cdab0e0d2d1fc77988d1b7ae0556e625f944f8b65ad87c9712fbc1a4"
},
"downloads": -1,
"filename": "rellm-0.0.5.tar.gz",
"has_sig": false,
"md5_digest": "76ee8b73a64bf8e2c7b5077cee9e9a78",
"packagetype": "sdist",
"python_version": "source",
"requires_python": ">=3.8,<4.0",
"size": 5572,
"upload_time": "2023-05-06T21:39:41",
"upload_time_iso_8601": "2023-05-06T21:39:41.464499Z",
"url": "https://files.pythonhosted.org/packages/dc/fb/96db913168c2c770cc54b80781fc02d2c5c82a7a23e2089446a4a909fd6c/rellm-0.0.5.tar.gz",
"yanked": false,
"yanked_reason": null
}
],
"upload_time": "2023-05-06 21:39:41",
"github": true,
"gitlab": false,
"bitbucket": false,
"codeberg": false,
"github_user": "r2d4",
"github_project": "rellm",
"travis_ci": false,
"coveralls": false,
"github_actions": false,
"lcname": "rellm"
}
