Hugging Face's logo

xaskasdf
/
llm-compiler-13b-ftd-gguf

GGUF
English
Inference Endpoints
Model card Files Community
llm-compiler-13b-ftd-gguf / README.md
xaskasdf's picture
xaskasdf
Update README.md
12d8564 verified
preview code
|
raw
history blame contribute delete
4.24 kB
metadata
language:
  - en

Model Card: LLM Compiler 13B FTD GGUF

This is a GGUF conversion and quantization of Meta's LLM Compiler 13B FTD model, optimized for code and compiler tasks.

Model Details

  • Developed by: Meta
  • Model type: LLM Compiler 13B FTD (Fine-tuned for code size and disassembly)
  • Language(s): English, relevant programming languages, LLVM IR, x86_64 assembly, and ARM assembly
  • License: Meta Large Language Model Compiler (LLM Compiler) License Agreement
  • Model Architecture: Auto-regressive language model using an optimized transformer architecture
  • Conversion: GGUF format conversion and quantization performed using llama.cpp build 3263 (26a39bbd) on Windows 11
  • Compiler used: MSVC 19.37.32824.0 for x64

Model Description

LLM Compiler 13B FTD is part of the LLM Compiler family developed by Meta, designed specifically for code optimization tasks. This GGUF version is a converted and quantized variant of the original model, optimized for use with llama.cpp.

The model is fine-tuned for:

  1. Predicting optimal optimization passes for LLVM's opt to minimize code size
  2. Generating LLVM IR from x86_64 or ARM assembly code

Intended Use

This model is intended for commercial and research use in tasks related to compiler optimization, code analysis, and transformation. It excels in tasks such as:

  • Emulating compiler transformations
  • Optimizing code for size
  • Disassembling and decompiling code

Limitations and Ethical Considerations

  • The model's performance may vary from the original due to quantization.
  • It should not be used for generating or manipulating code in ways that could lead to harmful or malicious software.
  • The model may produce unexpected or inaccurate results in some cases.
  • Users should perform their own safety testing and tuning for specific applications.

How to Use

To use this model, you'll need to use llama.cpp or compatible software that can work with GGUF format models. Refer to the llama.cpp documentation for specific usage instructions.

Example prompt formats can be found in the llm_compiler_demo.py file from the original repository.

License and Use Restrictions

This model is subject to multiple licenses:

  1. The original model is licensed under the Meta Large Language Model Compiler (LLM Compiler) License Agreement. Key points of this license include:

    • Non-exclusive, worldwide, non-transferable and royalty-free limited license to use, reproduce, distribute, copy, create derivative works of, and make modifications to the LLM Compiler Materials.
    • Any distribution must include a copy of the license agreement and display "Built with LLM Compiler" prominently.
    • Compliance with applicable laws and the Acceptable Use Policy for Llama Materials is required.
    • The model cannot be used to improve other large language models.
    • Special licensing is required for products or services with over 700 million monthly active users.

  2. The GGUF conversion is subject to the MIT License, as it uses llama.cpp for the conversion process.

  3. Usage of the model must comply with the Acceptable Use Policy for Llama Materials.

Commercial use may require additional licensing from Meta. Users are responsible for ensuring their use complies with all applicable licenses and policies.

For full license details and use restrictions, please refer to:

It is strongly recommended to review the full license text and consult with legal counsel if you have any questions about your specific use case.

Additional Information

For more details on the original model's performance, training process, and ethical considerations, please refer to the research paper: "Meta Large Language Model Compiler: Foundation Models of Compiler Optimization"

This GGUF version was created to enable efficient use with llama.cpp and related projects. Performance may differ from the original model due to quantization and format conversion.