cross-posted from: https://lemmy.world/post/3879861

Beating GPT-4 on HumanEval with a Fine-Tuned CodeLlama-34B

Hello everyone! This post marks an exciting moment for [email protected] and everyone in the open-source large language model and AI community.

We appear to have a new contender on the block, a model apparently capable of surpassing OpenAI’s state of the art ChatGPT-4 in coding evals (evaluations).

This is huge. Not too long ago I made an offhand comment on us catching up to GPT-4 within a year. I did not expect that prediction to end up being reality in half the time. Let’s hope this isn’t a one-off scenario and that we see a new wave of open-source models that begin to challenge OpenAI.

Buckle up, it’s going to get interesting!

Here’s some notes from the blog, which you should visit and read in its entirety:


Blog Post

We have fine-tuned CodeLlama-34B and CodeLlama-34B-Python on an internal Phind dataset that achieved 67.6% and 69.5% pass@1 on HumanEval, respectively. GPT-4 achieved 67% according to their official technical report in March. To ensure result validity, we applied OpenAI’s decontamination methodology to our dataset.

The CodeLlama models released yesterday demonstrate impressive performance on HumanEval.

  • CodeLlama-34B achieved 48.8% pass@1 on HumanEval
  • CodeLlama-34B-Python achieved 53.7% pass@1 on HumanEval

We have fine-tuned both models on a proprietary dataset of ~80k high-quality programming problems and solutions. Instead of code completion examples, this dataset features instruction-answer pairs, setting it apart structurally from HumanEval. We trained the Phind models over two epochs, for a total of ~160k examples. LoRA was not used — both models underwent a native fine-tuning. We employed DeepSpeed ZeRO 3 and Flash Attention 2 to train these models in three hours using 32 A100-80GB GPUs, with a sequence length of 4096 tokens.

Furthermore, we applied OpenAI’s decontamination methodology to our dataset to ensure valid results, and found no contaminated examples. 

The methodology is:

  • For each evaluation example, we randomly sampled three substrings of 50 characters or used the entire example if it was fewer than 50 characters.
  • A match was identified if any sampled substring was a substring of the processed training example.

For further insights on the decontamination methodology, please refer to Appendix C of OpenAI’s technical report. Presented below are the pass@1 scores we achieved with our fine-tuned models:

  • Phind-CodeLlama-34B-v1 achieved 67.6% pass@1 on HumanEval
  • Phind-CodeLlama-34B-Python-v1 achieved 69.5% pass@1 on HumanEval

Download

We are releasing both models on Huggingface for verifiability and to bolster the open-source community. We welcome independent verification of results.


If you get a chance to try either of these models out, let us know how it goes in the comments below!

If you found anything about this post interesting, consider subscribing to [email protected].

Cheers to the power of open-source! May we continue the fight for optimization, efficiency, and performance.

  • @babysharknanana
    link
    English
    11 year ago

    Exciting, but as far as I know, we can’t use LLaMA commercially. So I ask myself how to use it in a non-commercial context? Isn’t it expensive to embedd such a model in free/open-source software?

      • @babysharknanana
        link
        English
        11 year ago

        I know, but the text is only talking of Llama. So this is using Llama 2?

        • @abhibeckert
          link
          English
          2
          edit-2
          1 year ago

          LLama2 and Llama are basically excatly the same model, except the “2” version has a more permissive license and was trained with a larger source data set. Nobody should use the old one ever, and I expect the noncommercial license is part of a contract Meta signed with someone who provided source material.

          This is “CodeLlama” which was built on Llama2 and allows commercial use.

    • @backgroundcow
      link
      English
      0
      edit-2
      1 year ago

      I understand LLaMA and some other models come with instructions that say that they cannot be used commercially. But, unless the creators can show that you have formally accepted a license agreement to that effect, on what legal grounds can that be enforceable?

      If we look at the direction US law is moving, it seems the current legal theory is that AI generated works fall in the public domain. That means restricting their use commercially should be impossible regardless of other circumstances - public domain means that anyone can use them for anything. (But it also means that your commercial use isn’t protected from others likewise using the exact same output).

      If we instead look at what possible legal grounds restrictions on the output of these models could be based on if you didn’t agree to a license agreement to access the model. Copyright don’t restrict use, it restricts redistribution. The creators of LLMs cannot reasonably take the position that output created from their models is a derivative work of the model, when their model itself is created from copyrighted works, many of which they have no right to redistribute. The whole basis of LLMs rest on that “training data” -> “model” produces a model that isn’t encumbered by the copyright of the training data. How can one take that position and simultaneously belive “model” -> “inferred output” produces copyright encumbered output? That would be a fundamentally inconsistent view.

      (Note: the above is not legal advice, only free-form discussion.)