Refactor and improve space and leaderboard initialization

- Added [`ignore/`]
- Refactored space initialization in [`app.py`]
- Updated labels and filters in the [`init_leaderboard`]
- Removed evaluation script in `about.py`.
- Updated labels in [`src/display/utils.py`]
- Modified queue and results repositories in [`src/envs.py`]

.gitignore

@@ -1,3 +1,4 @@
+ignore/
 auto_evals/
 venv/
 __pycache__/

app.py

@@ -36,24 +36,37 @@ from src.submission.submit import add_new_eval
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
-### Space initialisation
+# Space initialisation
 try:
-    print(EVAL_REQUESTS_PATH)
     snapshot_download(
-        repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=QUEUE_REPO,
+        local_dir=EVAL_REQUESTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
 try:
-    print(EVAL_RESULTS_PATH)
     snapshot_download(
-        repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+        repo_id=RESULTS_REPO,
+        local_dir=EVAL_RESULTS_PATH,
+        repo_type="dataset",
+        tqdm_class=None,
+        etag_timeout=30,
+        token=TOKEN,
     )
 except Exception:
     restart_space()
 
 
-LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
+LEADERBOARD_DF = get_leaderboard_df(
+    EVAL_RESULTS_PATH,
+    EVAL_REQUESTS_PATH,
+    COLS,
+    BENCHMARK_COLS,
+)
 
 (
     finished_eval_queue_df,
@@ -70,25 +83,21 @@ def init_leaderboard(dataframe):
         select_columns=SelectColumns(
             default_selection=[c.name for c in fields(AutoEvalColumn) if c.displayed_by_default],
             cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
-            label="Select Columns to Display:",
+            label="Columns",
         ),
         search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
         hide_columns=[c.name for c in fields(AutoEvalColumn) if c.hidden],
         filter_columns=[
-            ColumnFilter(AutoEvalColumn.model_type.name, type="checkboxgroup", label="Model types"),
-            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Precision"),
+            ColumnFilter(AutoEvalColumn.precision.name, type="checkboxgroup", label="Floating-point format"),
             ColumnFilter(
                 AutoEvalColumn.params.name,
                 type="slider",
-                min=0.01,
-                max=150,
-                label="Select the number of parameters (B)",
-            ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=False
+                min=1,
+                max=500,
+                step=1,
+                label="Number of parameters (billions)",
             ),
         ],
-        bool_checkboxgroup_label="Hide models",
         interactive=False,
     )
 

src/about.py

@@ -76,238 +76,6 @@ 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)
-```
-
-'''
+EVALUATION_SCRIPT = ''
 CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
-CITATION_BUTTON_TEXT = r"""
-"""
+CITATION_BUTTON_TEXT = ''

src/display/utils.py

@@ -42,9 +42,9 @@ auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type",
 auto_eval_column_dict.append(["architecture", ColumnContent, ColumnContent("Architecture", "str", False)])
 auto_eval_column_dict.append(["weight_type", ColumnContent, ColumnContent("Weight type", "str", False, True)])
 auto_eval_column_dict.append(["precision", ColumnContent, ColumnContent("Precision", "str", False)])
-auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("Hub License", "str", False)])
-auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("#Params (B)", "number", False)])
-auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Hub ❤️", "number", False)])
+auto_eval_column_dict.append(["license", ColumnContent, ColumnContent("License", "str", False)])
+auto_eval_column_dict.append(["params", ColumnContent, ColumnContent("Parameters ⚙️", "number", False)])
+auto_eval_column_dict.append(["likes", ColumnContent, ColumnContent("Likes ❤️", "number", False)])
 auto_eval_column_dict.append(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
 auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
 

src/envs.py

@@ -14,8 +14,8 @@ OWNER = "braindao"  # Change to your org - don't forget to create a results and
 # ----------------------------------
 
 REPO_ID = f"{OWNER}/solidity-leaderboard"
-QUEUE_REPO = f"{OWNER}/requests"
-RESULTS_REPO = f"{OWNER}/results"
+QUEUE_REPO = f"{OWNER}/solbench-leaderboard-queue"
+RESULTS_REPO = f"{OWNER}/solbench-leaderboard-results"
 
 # If you setup a cache later, just change HF_HOME
 CACHE_PATH = os.getenv("HF_HOME", ".")