#!/usr/bin/env python # -*- coding: utf-8 -*- # flake8: noqa E501 from dataclasses import dataclass from enum import Enum @dataclass class Task: benchmark: str metric: str col_name: str # Select your tasks here # --------------------------------------------------- class Tasks(Enum): # task_key in the json file, metric_key in the json file, name to display in the leaderboard task0 = Task("hf", "score", "LLM JUDGE") task1 = Task("rouge1", "score", "ROUGE-unigrams") task2 = Task("rouge2", "score", "ROUGE-bigrams") task3 = Task("rougeL", "score", "ROUGE-Longest Common Subsequence") task4 = Task("rougeLsum", "score", "ROUGE-Lsum") task5 = Task("bleu", "score", "Bleu") task6 = Task("brevity_penalty", "score", "Brevity Penalty") task7 = Task("hf", "score", "LLM JUDGE") # task = Task("precisions", "score", "Precision") NUM_FEWSHOT = 0 # Change with your few shot # --------------------------------------------------- # Your leaderboard name TITLE = """

Evaluating Solidity Code Generation

""" # What does your leaderboard evaluate? INTRODUCTION_TEXT = "" # Which evaluations are you running? how can people reproduce what you have? LLM_BENCHMARKS_TEXT = """ ## How it works ## Reproducibility To reproduce our results, here is the commands you can run: """ EVALUATION_QUEUE_TEXT = """ ## Some good practices before submitting a model ### 1) Make sure you can load your model and tokenizer using AutoClasses: ```python from transformers import AutoConfig, AutoModel, AutoTokenizer config = AutoConfig.from_pretrained("your model name", revision=revision) model = AutoModel.from_pretrained("your model name", revision=revision) tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision) ``` If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded. Note: make sure your model is public! Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted! ### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index) It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`! ### 3) Make sure your model has an open license! This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗 ### 4) Fill up your model card When we add extra information about models to the leaderboard, it will be automatically taken from the model card ## In case of model failure If your model is displayed in the `FAILED` category, its execution stopped. Make sure you have followed the above steps first. If everything is done, check you can launch the EleutherAIHarness on your model locally, using the above command without modifications (you can add `--limit` to limit the number of examples per task). """ EVALUATION_SCRIPT = ''' To evaluate the model you can access the colab notebook at [this link](https://colab.research.google.com/drive/145KAGvgdAb8BrkObUrxAVWBd9EGDqy8N?usp=sharing). ## First install the necessary libraries ``` pip install accelerate openai anthropic datasets ``` ## Setup your : * OPENAI_API_KEY * ANTHROPIC_API_KEY * HF_TOKEN ## Select a model ```python MODEL_ID = # model_id_here ``` ## Then run the following script ````python from transformers import pipeline import torch import os import json from openai import OpenAI import anthropic from huggingface_hub.utils._token import get_token from huggingface_hub import InferenceClient HF_TOKEN = get_token() from datasets import load_dataset ds = load_dataset("braindao/solbench-naive-judge-random-v1",split="test") pipe = pipeline("text-generation", model= MODEL_ID , torch_dtype=torch.bfloat16, device_map="auto") def generate(message): messages = [ {"role": "user", "content": message}, ] return pipe(messages,max_new_tokens=1024)[0]["generated_text"][1]["content"] def convert_to_int(text): value = 0 try : value = int(text) except : pass return value def anthropic_judge(code,baseline): prompt = f"""Analyze the provided Solidity code and assign a score from 0 to 10 based on these criteria: 1. Functionality (0-2 points) 2. Security (0-2 points) 3. Efficiency (0-2 points) 4. Readability and Style (0-2 points) 5. Similarity with the Expert Code (0-2 points) We Evaluate the code thoroughly, sum up the points, and return ONLY an integer value representing the final score. Your entire response should consist of a single integer between 0 and 10, inclusive. Solidity code to evaluate: ```solidity {code} ``` Expert Code: ```solidity {baseline} ``` OUTPUT FORMAT: [integer]""" sys = """You are a solidity code judge, You will only reply with an integer value between 0-10""" client = anthropic.Anthropic() message = client.messages.create( model="claude-3-5-sonnet-20240620", max_tokens=1000, temperature=0, system=sys, messages=[ { "role": "user", "content": [ { "type": "text", "text": prompt } ] } ] ) return convert_to_int(message.content[0].text) def openai_judge(code,baseline): prompt = f"""evaluate the following solidity code and return a score between 0 and 10 based how far the code achieves the following criteria: 1. Functionality (0-2 points) 2. Security (0-2 points) 3. Efficiency (0-2 points) 4. Readability and Style (0-2 points) 5. Similarity with the Expert Code (0-2 points) code to evaluate: {code} expert code: {baseline} return only an integer value and no additional comment, score should be either 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10. """ client = OpenAI() completion = client.chat.completions.create( model="gpt-4o", messages=[ {"role": "user", "content": prompt} ] ) return convert_to_int(completion.choices[0].message.content) def hf_judge(code,baseline): prompt = f"""evaluate the following solidity code and return a score between 0 and 10 based how far the code achieves the following criteria: 1. Functionality (0-2 points) 2. Security (0-2 points) 3. Efficiency (0-2 points) 4. Readability and Style (0-2 points) 5. Similarity with the Expert Code (0-2 points) code to evaluate: {code} expert code: {baseline} return only an integer value and no additional comment, score should be either 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 or 10. """ client = InferenceClient( "meta-llama/Meta-Llama-3.1-405B-Instruct", token=HF_TOKEN, ) out = "" try : for message in client.chat_completion( messages=[{"role":"system","content" : "you are a solidity code judge, you will only reply with an integer value between 0-10"}, {"role": "user", "content": prompt}], max_tokens=500, stream=True, ): out += message.choices[0].delta.content except : pass return convert_to_int(out) def LLM_JUDGE(code,baseline,judges=["openai","anthropic","hf"]) : out = {} if "openai" in judges : out["openai"] = openai_judge(code,baseline) if "anthropic" in judges : out["anthropic"] = anthropic_judge(code,baseline) if "hf" in judges : out["hf"] = hf_judge(code,baseline) return out # Judge model against data from tqdm import tqdm scores = {"openai":[],"anthropic":[],"hf":[]} for sample in tqdm(ds) : score = evaluate_sample(sample) for key in score.keys(): scores[key].append(score[key]) # normalize scores for key in scores.keys(): scores[key] = sum(scores[key])/(10*len(scores[key])) d = { "config": { "model_dtype": "torch.bfloat16", "model_name": MODEL_ID, "model_sha": "main" }, "results": { "openai": { "score": 0 }, "anthropic": { "score": 0 }, "hf": { "score": 0 } } } for key in scores.keys() : d["results"][key]["score"] = scores[key] # Serializing json json_object = json.dumps(d, indent=4) # Writing to sample.json file_name = MODEL_ID.split("/")[1] + ".json" with open(file_name, "w") as outfile: outfile.write(json_object) ```` ## if you are not part of braindao set `create_pr` to **True** ```python from huggingface_hub import upload_file upload_file(path_or_fileobj = file_name, path_in_repo=f"{MODEL_ID}.json", repo_id="braindao/results", repo_type="dataset", create_pr=False) ``` ''' CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results" CITATION_BUTTON_TEXT = r""" """