Metadata-Version: 2.1
Name: grammar-guide
Version: 0.0.11
Summary: Speculative grammar backtracking algorithm for LLM decoding conforming to some lark context-free grammar (CFG)
Author-email: Parker Glenn <parkervg5@gmail.com>
License:                                  Apache License
                                   Version 2.0, January 2004
                                http://www.apache.org/licenses/
        
           TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
        
           1. Definitions.
        
              "License" shall mean the terms and conditions for use, reproduction,
              and distribution as defined by Sections 1 through 9 of this document.
        
              "Licensor" shall mean the copyright owner or entity authorized by
              the copyright owner that is granting the License.
        
              "Legal Entity" shall mean the union of the acting entity and all
              other entities that control, are controlled by, or are under common
              control with that entity. For the purposes of this definition,
              "control" means (i) the power, direct or indirect, to cause the
              direction or management of such entity, whether by contract or
              otherwise, or (ii) ownership of fifty percent (50%) or more of the
              outstanding shares, or (iii) beneficial ownership of such entity.
        
              "You" (or "Your") shall mean an individual or Legal Entity
              exercising permissions granted by this License.
        
              "Source" form shall mean the preferred form for making modifications,
              including but not limited to software source code, documentation
              source, and configuration files.
        
              "Object" form shall mean any form resulting from mechanical
              transformation or translation of a Source form, including but
              not limited to compiled object code, generated documentation,
              and conversions to other media types.
        
              "Work" shall mean the work of authorship, whether in Source or
              Object form, made available under the License, as indicated by a
              copyright notice that is included in or attached to the work
              (an example is provided in the Appendix below).
        
              "Derivative Works" shall mean any work, whether in Source or Object
              form, that is based on (or derived from) the Work and for which the
              editorial revisions, annotations, elaborations, or other modifications
              represent, as a whole, an original work of authorship. For the purposes
              of this License, Derivative Works shall not include works that remain
              separable from, or merely link (or bind by name) to the interfaces of,
              the Work and Derivative Works thereof.
        
              "Contribution" shall mean any work of authorship, including
              the original version of the Work and any modifications or additions
              to that Work or Derivative Works thereof, that is intentionally
              submitted to Licensor for inclusion in the Work by the copyright owner
              or by an individual or Legal Entity authorized to submit on behalf of
              the copyright owner. For the purposes of this definition, "submitted"
              means any form of electronic, verbal, or written communication sent
              to the Licensor or its representatives, including but not limited to
              communication on electronic mailing lists, source code control systems,
              and issue tracking systems that are managed by, or on behalf of, the
              Licensor for the purpose of discussing and improving the Work, but
              excluding communication that is conspicuously marked or otherwise
              designated in writing by the copyright owner as "Not a Contribution."
        
              "Contributor" shall mean Licensor and any individual or Legal Entity
              on behalf of whom a Contribution has been received by Licensor and
              subsequently incorporated within the Work.
        
           2. Grant of Copyright License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              copyright license to reproduce, prepare Derivative Works of,
              publicly display, publicly perform, sublicense, and distribute the
              Work and such Derivative Works in Source or Object form.
        
           3. Grant of Patent License. Subject to the terms and conditions of
              this License, each Contributor hereby grants to You a perpetual,
              worldwide, non-exclusive, no-charge, royalty-free, irrevocable
              (except as stated in this section) patent license to make, have made,
              use, offer to sell, sell, import, and otherwise transfer the Work,
              where such license applies only to those patent claims licensable
              by such Contributor that are necessarily infringed by their
              Contribution(s) alone or by combination of their Contribution(s)
              with the Work to which such Contribution(s) was submitted. If You
              institute patent litigation against any entity (including a
              cross-claim or counterclaim in a lawsuit) alleging that the Work
              or a Contribution incorporated within the Work constitutes direct
              or contributory patent infringement, then any patent licenses
              granted to You under this License for that Work shall terminate
              as of the date such litigation is filed.
        
           4. Redistribution. You may reproduce and distribute copies of the
              Work or Derivative Works thereof in any medium, with or without
              modifications, and in Source or Object form, provided that You
              meet the following conditions:
        
              (a) You must give any other recipients of the Work or
                  Derivative Works a copy of this License; and
        
              (b) You must cause any modified files to carry prominent notices
                  stating that You changed the files; and
        
              (c) You must retain, in the Source form of any Derivative Works
                  that You distribute, all copyright, patent, trademark, and
                  attribution notices from the Source form of the Work,
                  excluding those notices that do not pertain to any part of
                  the Derivative Works; and
        
              (d) If the Work includes a "NOTICE" text file as part of its
                  distribution, then any Derivative Works that You distribute must
                  include a readable copy of the attribution notices contained
                  within such NOTICE file, excluding those notices that do not
                  pertain to any part of the Derivative Works, in at least one
                  of the following places: within a NOTICE text file distributed
                  as part of the Derivative Works; within the Source form or
                  documentation, if provided along with the Derivative Works; or,
                  within a display generated by the Derivative Works, if and
                  wherever such third-party notices normally appear. The contents
                  of the NOTICE file are for informational purposes only and
                  do not modify the License. You may add Your own attribution
                  notices within Derivative Works that You distribute, alongside
                  or as an addendum to the NOTICE text from the Work, provided
                  that such additional attribution notices cannot be construed
                  as modifying the License.
        
              You may add Your own copyright statement to Your modifications and
              may provide additional or different license terms and conditions
              for use, reproduction, or distribution of Your modifications, or
              for any such Derivative Works as a whole, provided Your use,
              reproduction, and distribution of the Work otherwise complies with
              the conditions stated in this License.
        
           5. Submission of Contributions. Unless You explicitly state otherwise,
              any Contribution intentionally submitted for inclusion in the Work
              by You to the Licensor shall be under the terms and conditions of
              this License, without any additional terms or conditions.
              Notwithstanding the above, nothing herein shall supersede or modify
              the terms of any separate license agreement you may have executed
              with Licensor regarding such Contributions.
        
           6. Trademarks. This License does not grant permission to use the trade
              names, trademarks, service marks, or product names of the Licensor,
              except as required for reasonable and customary use in describing the
              origin of the Work and reproducing the content of the NOTICE file.
        
           7. Disclaimer of Warranty. Unless required by applicable law or
              agreed to in writing, Licensor provides the Work (and each
              Contributor provides its Contributions) on an "AS IS" BASIS,
              WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
              implied, including, without limitation, any warranties or conditions
              of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
              PARTICULAR PURPOSE. You are solely responsible for determining the
              appropriateness of using or redistributing the Work and assume any
              risks associated with Your exercise of permissions under this License.
        
           8. Limitation of Liability. In no event and under no legal theory,
              whether in tort (including negligence), contract, or otherwise,
              unless required by applicable law (such as deliberate and grossly
              negligent acts) or agreed to in writing, shall any Contributor be
              liable to You for damages, including any direct, indirect, special,
              incidental, or consequential damages of any character arising as a
              result of this License or out of the use or inability to use the
              Work (including but not limited to damages for loss of goodwill,
              work stoppage, computer failure or malfunction, or any and all
              other commercial damages or losses), even if such Contributor
              has been advised of the possibility of such damages.
        
           9. Accepting Warranty or Additional Liability. While redistributing
              the Work or Derivative Works thereof, You may choose to offer,
              and charge a fee for, acceptance of support, warranty, indemnity,
              or other liability obligations and/or rights consistent with this
              License. However, in accepting such obligations, You may act only
              on Your own behalf and on Your sole responsibility, not on behalf
              of any other Contributor, and only if You agree to indemnify,
              defend, and hold each Contributor harmless for any liability
              incurred by, or claims asserted against, such Contributor by reason
              of your accepting any such warranty or additional liability.
        
           END OF TERMS AND CONDITIONS
        
           APPENDIX: How to apply the Apache License to your work.
        
              To apply the Apache License to your work, attach the following
              boilerplate notice, with the fields enclosed by brackets "[]"
              replaced with your own identifying information. (Don't include
              the brackets!)  The text should be enclosed in the appropriate
              comment syntax for the file format. We also recommend that a
              file or class name and description of purpose be included on the
              same "printed page" as the copyright notice for easier
              identification within third-party archives.
        
           Copyright [yyyy] [name of copyright owner]
        
           Licensed under the Apache License, Version 2.0 (the "License");
           you may not use this file except in compliance with the License.
           You may obtain a copy of the License at
        
               http://www.apache.org/licenses/LICENSE-2.0
        
           Unless required by applicable law or agreed to in writing, software
           distributed under the License is distributed on an "AS IS" BASIS,
           WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
           See the License for the specific language governing permissions and
           limitations under the License.
        
Project-URL: Documentation, https://github.com/parkervg/grammar-guide
Keywords: language models,structured generation,constrained decoding,formal grammars
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Developers
Classifier: Intended Audience :: Science/Research
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Text Processing
Classifier: Programming Language :: Python :: 3
Requires-Python: >=3.9
Description-Content-Type: text/markdown
License-File: LICENSE
Requires-Dist: guidance>=0.1.16
Requires-Dist: numpy<2
Requires-Dist: exrex
Requires-Dist: lark
Requires-Dist: colorama
Requires-Dist: pygtrie
Requires-Dist: ipython

<div align="center"><h1> Grammar Guide </h1>
    Speculative grammar backtracking algorithm to perform grammar-constrained decoding with <i>any</i> text generation function (OpenAI, Anthropic, etc.)
</div>

---

This repo is a slightly modified implementation of the decoding mechanism described in Section 3.2 of [Grammar Prompting for Domain-Specific Language
Generation with Large Language Models](https://arxiv.org/pdf/2305.19234) by [@berlino](https://github.com/berlino). I refer to the general algorithm as **speculative grammar backtracking**.

It is a form of constrained decoding, and can be used to guide even proprietary, black-box LLM APIs according to some context-free grammar. It's rooted in the idea that as LLMs get better, not *all* steps of the decoding process need to apply a strict logit mask - like a good teacher, we let the student give an answer, and only intervene and correct when necessary.

When using local Transformer models, we can efficiently backtrack the KV cache according to a given Lark CFG. Below is a benchmark showing tokens/sec when generating a JSON object with [10, 20, 30, 40] keys using [HuggingFaceTB/SmolLM-135M](https://huggingface.co/HuggingFaceTB/SmolLM-135M) on my Macbook M1.
![naive-vs-grammar-guide](img/naive-vs-gg.png)

### Features
- Compatible with *any* text generation function
  - OpenAI, Anthropic etc. - as long as you can provide some `generate(prompt: str) -> str` function!
- Efficient re-use of KV cache for all CausalLM Transformer models
  - Optimistic, speculative decoding = no need to manually update to support new tokenizers
- Visualization and logging of grammar corrections
- Token healing to ensure high probability continuations

```
pip install grammar-guide
```
## Examples
- [With Anthropic](./examples/anthropic-gg.ipynb)
- [With OpenAI](./examples/openai-gg.ipynb)
- [With Transformers](./examples/transformers-gg.ipynb)

### With Transformer Models
When using HuggingFace Transformer models, we get an extra speed boost by leveraging efficient caching and backtracking of the KV cache. When a grammar correction is made, we backtrack to the state of the KV cache aligned to the longest prefix that is valid under our Lark context-free grammar.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import guidance

import grammar_guide as gg

model_name_or_path = "HuggingFaceTB/SmolLM-135M"
model = AutoModelForCausalLM.from_pretrained(model_name_or_path)
tokenizer = AutoTokenizer.from_pretrained(model_name_or_path)
parser = gg.load_parser("../grammars/json_five_values_string_only.lark")

res = gg.guide(
  draft_model=model,
  tokenizer=tokenizer,
  parser=parser,
  prompt="Here's a JSON object with only string values:",
  target_model=guidance.models.Transformers(
      model_name_or_path, echo=False
  ),
  stop_at=['```'],
  token_lookahead=20,
  max_grammar_corrections=20,
  temperature=0.0
)
```
<img src="./img/grammar-guide-recording.gif" width="100%" height="100%" />

In the visualization below:
- <span style="background-color: lightgreen;">Green highlights</span> = text generated by the draft model
- <span style="background-color: lightblue;">Blue highlights</span> = candidates selected by our target model
- <span style="color: red;">Red text</span> = backtracked text that violated the context-free grammar
- <span style="color: orange;">Orange text</span> = tokens that were fed through the token healing logits processor


![jupyer-visualization](img/jupyter-example.png)

### With General API-based Providers
```python
import os
from openai import OpenAI
import guidance

import grammar_guide as gg

client = OpenAI(api_key=os.environ.get("OPENAI_API_KEY"))
parser = gg.load_parser("../grammars/json_five_values_string_only.lark")

# Define our core completion predict function
# This just needs to follow the `fn(s: str) -> str` contract
#   so we can use any black-box API provider.
def openai_generate(prefix: str, prompt: str, max_new_tokens: int) -> str:
    messages = [
        {
            "role": "system",
            "content": prompt
        }
    ]
    if prefix:
        messages += [
            {
                "role": "assistant",
                "content": prefix
            }
        ]
    chat_completion = client.chat.completions.create(
        messages=messages,
        model="gpt-4o-mini",
        max_tokens=max_new_tokens,
        temperature=0.0
    )
    return chat_completion.choices[0].message.content

res = gg.guide(
    draft_model=openai_generate,
    parser=parser,
    prompt="Here's a JSON object with only string values:",
    target_model=guidance.models.Transformers(
        "HuggingFaceTB/SmolLM-135M", echo=False
    ),
    max_grammar_corrections=20,
    verbose=True,
)
```

## Documentation

All calls to `gg.guide` take the following arguments. When `draft_model` is of type `AutoModelForCausalLM`, we have a bit of extra control, hence the 'Transformers only' arguments.

| Argument | Type | Description |
|----------|------|-------------|
| `draft_model` | `Union[AutoModelForCausalLM, Callable[[str], str]]` | A transformer model or callable to use for text generation. |
| `tokenizer` | `AutoTokenizer` | Transformers only, the tokenizer associated with the model. |
| `parser` | `EarleyParser` | The parser used for grammar checking. |
| `prompt` | `str` | The initial prompt for text generation. |
| `target_model` | `guidance.models.Model` | The guidance model to use for constrained grammar correction. See [guidance-ai/guidance](https://github.com/guidance-ai/guidance) |
| `seed_str` | `Optional[str]` | An optional seed string to start the generation. |
| `max_grammar_corrections` | `int` | Maximum number of grammar corrections to attempt. |
| `stop_at` | `Collection[str]` | Collection of strings to stop generation at. |
| `token_healing` | `Optional[bool]` | Transformers only, whether to use token healing during generation. |
| `top_p` | `float` | Transformers only, the cumulative probability for top-p sampling. |
| `temperature` | `float` | Transformers only, the temperature for controlling randomness in generation. |
| `token_lookahead` | `int` | Maximum number of new tokens to generate using draft model. Essentially the $K$ parameter in speculative decoding. |
| `save_html` | `bool` | Whether to save the generation process as HTML. |
| `verbose` | `bool` | Whether to print verbose output. |
| `debug` | `bool` | Whether to run in debug mode with additional checks. |

As described in [the paper](https://arxiv.org/pdf/2305.19234), one way many existing libraries achieve constrained decoding is by enforcing some constraint at each decoding timestep. For local models, it is possible to pre-process the logit masks such that this is relatively efficient. However, for closed models (think OpenAI, Anthropic, etc.), this can be 'prohitively expensive', since it would require calling the API at each timestep with the full prompt and valid continuation tokens.

Instead, this library takes an optimistic approach to constrained decoding. Autoregressive language models are only going to get better, and often times the overhead of strict, mask-driven constrained decoding isn't worth it.

For example, if we want gpt-4o to generate some SQLite query, chances are, it'll generate a valid query without any constraints.

If there is a mistake, though, we use our grammar to parse the longest prefix that abides by our grammar definition.

```python
prediction = "SELECT * FROM students WHERE name SIMILAR TO 'Dan%';"
# Oops! `SIMILAR TO` works in PostgreSQL, but not SQLite
prefix, candidates = obtain_correction_pairs(prediction, parser)
print(prefix)
# SELECT * FROM students WHERE name
print(candidates)
# ['IN', '>', '=', 'NOT', 'BETWEEN', 'LIKE', ...]
```
Once we have a list of candidates, we can use our target model to select a valid continuation. In the above example, our candidates are fairly simple strings. However, our grammar may define regular expression continuations as well (e.g. `(?:(?:[A-Z]|[a-z])|_)(?:(?:(?:[A-Z]|[a-z])|[0-9]|_))*`).
This is powered by the library [guidance](https://github.com/guidance-ai/guidance).

Once the target model has selected a valid continuation, we are free to pass the new prefix back to the draft lanugage model to complete the prediction.

```python
selected_candidate = choose_candidate(candidates, prefix, target_model)
print(selected_candidate)
# 'LIKE'
# Now, pass back to the main model to continue its prediction from this new breakpoint
draft_model.predict(prefix + selected_candidate)
```

> [!TIP]
> We borrow "Draft" and "Target" terminology from one of the original speculative decoding papers ([1](https://arxiv.org/pdf/2302.01318)). However, in our case, we consider the model constrained by the grammar which generates very small bits of text to be the 'target' model, since these generations will always be accepted. The draft model, then, is the often larger model that generates unconstrained up until we tell it to stop (governed by the `token_lookahead` parameter)

### Benchmarks
The below benchmarks are done on my Macbook M1, with the command `python -m examples.benchmarks.run`.

They measure the tokens/sec for the respective methods to generate a JSON with exactly *n* string key-value pairs, using [HuggingFaceTB/SmolLM-135M](https://huggingface.co/HuggingFaceTB/SmolLM-135M) and the below prompt.
> Here is a JSON object, with {n} keys, using only string values:\n\n```json\n

For most general usecases when using local Transformers models, I highly recommend the library [transformers-CFG](https://github.com/epfl-dlab/transformers-CFG)!

![runtime-benchmark](img/runtime-lineplot.png)
