Home

Awesome

🤗 Transformers CFG

Python 3.8+ License

💭 Latest News

We are thrilled to announce that transformers-cfg has been integrated into the Text-Generation-WebUI project, enabling users to utilize our CFG capabilities within this popular web interface for text generation. For more details, see the relevant pull request.

🚀 Introduction

transformers-cfg is an extension library for the popular Transformers library by Hugging Face, tailored for working with context-free grammars (CFG). This package provides additional tools and functionalities to enhance your experience with natural language processing tasks involving CFGs.

Initially developed as a pull request to the Hugging Face Transformers library, you can find the relevant discussion here.

💻 Installation

🔧 Quick Start: Force LLM to Generate a Valid JSON Object

Command-Line Interface

transformers-cfg-cli is a command-line tool that allows you to generate text using a model and a grammar. You can specify the model, grammar, prompts, and other parameters to generate text that conforms to the specified grammar.

transformers-cfg-cli generate \
    -m "microsoft/Phi-3-mini-4k-instruct" \
    -g "examples/grammars/json.ebnf" \
    -p "This is a valid json string for http request:" \
    --use_4bit \
    --max_new_tokens 60 \
    --repetition_penalty 1.1
# {"name":"John","age":30,"car":null}

We support also Unicode characters in the grammar:

transformers-cfg-cli generate \
    -m "microsoft/Phi-3-mini-4k-instruct" \
    -g "examples/grammars/chinese.ebnf" \
    -p "Translate the following sentence into Chinese: My neighbor is a very nice person. -> " \
    --use_4bit \
    --max_new_tokens 60 \
    --repetition_penalty 1.1

transformers-cfg-cli generate --help provides a list of available options and arguments.

<details> <summary>Click here to see an example of generating a JSON object with minimal changes to HF code, or check it out in <code>examples/generate_json.py</code></summary> 
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from transformers_cfg.grammar_utils import IncrementalGrammarConstraint
from transformers_cfg.generation.logits_process import GrammarConstrainedLogitsProcessor

if __name__ == "__main__":
    # Detect if GPU is available, otherwise use CPU
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print(f"Using device: {device}")

    model_id = "mistralai/Mistral-7B-v0.1"

    # Load model and tokenizer
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.pad_token = tokenizer.eos_token

    model = AutoModelForCausalLM.from_pretrained(model_id).to(device)
    model.generation_config.pad_token_id = model.generation_config.eos_token_id

    # Load JSON grammar
    with open("examples/grammars/json.ebnf", "r") as file:
        grammar_str = file.read()
    grammar = IncrementalGrammarConstraint(grammar_str, "root", tokenizer)
    grammar_processor = GrammarConstrainedLogitsProcessor(grammar)

    # Generate
    prompts = ["This is a valid json string for http request:", "This is a valid json string for shopping cart:"]
    input_ids = tokenizer(prompts, add_special_tokens=False, return_tensors="pt", padding=True)["input_ids"]

    output = model.generate(
        input_ids,
        max_length=50,
        logits_processor=[grammar_processor],
        repetition_penalty=1.1,
        num_return_sequences=1,
    )
    # Decode output
    generations = tokenizer.batch_decode(output, skip_special_tokens=True)
    print(generations)

    """
    'This is a valid json string for http request:{ "request": { "method": "GET", "headers": [], "content": "Content","type": "application" }}'
    'This is a valid json string for shopping cart:{ "name": "MyCart", "price": 0, "value": 1 }'
    """
</details> <details> <summary>Click here to see an example with HF pipeline API, or check it out in <code>examples/pipeline_json.py</code></summary> 
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from transformers_cfg.grammar_utils import IncrementalGrammarConstraint
from transformers_cfg.generation.logits_process import GrammarConstrainedLogitsProcessor

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token
model = AutoModelForCausalLM.from_pretrained(model_id).to(device)

# Load grammar
with open(f"examples/grammars/json.ebnf", "r") as file:
    grammar_str = file.read()
grammar = IncrementalGrammarConstraint(grammar_str, "root", tokenizer)
grammar_processor = GrammarConstrainedLogitsProcessor(grammar)

# Initialize pipeline
pipe = pipeline(
    "text-generation",
    model=model,
    tokenizer=tokenizer,
    device_map="auto",
    max_length=50,
    batch_size=2,
)

generations = pipe(
    [
        "This is a valid json string for http request: ",
        "This is a valid json string for shopping cart: ",
    ],
    do_sample=False,
    logits_processor=[grammar_processor],
)
</details>

💡 Why Should I Use transformers-cfg?

🤔 What Is a Grammar?

TL;DR: Think of it as an enhanced version of regular expressions.

<details> <summary>Here is a simple example of a JSON grammar:</summary> 
# A JSON object is the root of the grammar
root ::= object

# An object starts with "{" and ends with "}" and contains pairs separated by ","
object ::= "{" pair ("," pair)* "}"

# A pair is a string followed by a ":" and a value
pair ::= string ":" value

# A string is a sequence of alphanumeric characters enclosed in double quotes
string ::= '"' [a-zA-Z0-9]* '"'

# A value can be a string, another object, or a boolean value
value ::= string | object | "true" | "false" | "null"

This grammar describes the structure of a JSON object. It specifies that a JSON object consists of key-value pairs, where the key is a string, and the value can be a string, another object, or a boolean value.

You can use grammars to describe simple but useful constructs, such as valid email addresses, URLs, or phone numbers:

phone_number ::= "+" [0-9]+
</details>

For advanced grammar debugging, check out our debugging guide.

Automatic JSON Schema Grammar Conversion[Experimental]

Learn how to automatically create custom grammars for complex JSON objects in our documentation on JSON schema to grammar conversion.

Grammar Collection

We provide a collection of grammars in the examples/grammars folder, which are mostly identical to the grammars in the llama-cpp project. We try to keep these grammars up-to-date with the original project, though we cannot yet guarantee that all grammars from llama-cpp can be directly used in transformers-cfg.

Available grammars include:

Supported Models

See supported_models.yaml for the full list of supported models.

As a rule of thumb, all models with the same tokenizer should be naturally supported.

If you find any model that is not supported, please open an issue or submit a pull request.

Citation

Please consider citing our work if you find the provided resources useful:

@inproceedings{geng-etal-2023-grammar,
	title        = {Grammar-Constrained Decoding for Structured {NLP} Tasks without Finetuning},
	author       = {Geng, Saibo  and Josifoski, Martin  and Peyrard, Maxime  and West, Robert},
	year         = 2023,
	month        = dec,
	booktitle    = {Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing},
	publisher    = {Association for Computational Linguistics},
	address      = {Singapore},
	url          = {https://aclanthology.org/2023.emnlp-main.674},
	editor       = {Bouamor, Houda  and Pino, Juan  and Bali, Kalika}
}

License

This project is licensed under the MIT License.

Acknowledgements

This project is derived from the torch-grammars project, which was itself derived from the llama-cpp project.