fix-table

app.py

@@ -1,39 +1,34 @@
+import html
+import json
+from pathlib import Path
+
 import gradio as gr
-from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
 import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
-import html
-from pathlib import Path
+from gradio_leaderboard import ColumnFilter, Leaderboard, SelectColumns
 from huggingface_hub import snapshot_download
 
 from src.about import (
-    CITATION_BUTTON_LABEL,
-    CITATION_BUTTON_TEXT,
-    EVALUATION_QUEUE_TEXT,
     INTRODUCTION_TEXT,
     LLM_BENCHMARKS_TEXT,
     TITLE,
 )
-from src.display.css_html_js import custom_css
+from src.display.css_html_js import custom_css, custom_js
 from src.display.utils import (
     BENCHMARK_COLS,
     COLS,
     EVAL_COLS,
-    EVAL_TYPES,
     AutoEvalColumn,
-    ModelType,
     fields,
-    WeightType,
-    Precision
 )
 from src.envs import API, EVAL_REQUESTS_PATH, EVAL_RESULTS_PATH, QUEUE_REPO, REPO_ID, RESULTS_REPO, TOKEN
 from src.populate import get_evaluation_queue_df, get_leaderboard_df
-from src.submission.submit import add_new_eval
 
 
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
+
 ENABLE_REMOTE_DATA_SYNC = False
 
 if ENABLE_REMOTE_DATA_SYNC:
@@ -70,6 +65,7 @@ if ENABLE_REMOTE_DATA_SYNC:
         pending_eval_queue_df,
     ) = get_evaluation_queue_df(EVAL_REQUESTS_PATH, EVAL_COLS)
 
+
 def init_leaderboard(dataframe):
     if dataframe is None or dataframe.empty:
         raise ValueError("Leaderboard DataFrame is empty or None.")
@@ -93,9 +89,7 @@ def init_leaderboard(dataframe):
                 max=150,
                 label="Select the number of parameters (B)",
             ),
-            ColumnFilter(
-                AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True
-            ),
+            ColumnFilter(AutoEvalColumn.still_on_hub.name, type="boolean", label="Deleted/incomplete", default=True),
         ],
         bool_checkboxgroup_label="Hide models",
         interactive=False,
@@ -103,7 +97,7 @@ def init_leaderboard(dataframe):
 
 
 demo = gr.Blocks(css=custom_css)
-import json
+
 
 def create_score_df(json_path: Path, decimals: int):
     data = json.loads(json_path.read_text(encoding="utf-8"))
@@ -128,15 +122,20 @@ def create_score_df(json_path: Path, decimals: int):
     score_cols = [c for c in df.columns if c not in ("ID", "Model")]
     for col in score_cols:
         df[col] = df[col].apply(
-            lambda v: (
-                ""
-                if pd.isna(v)
-                else (f"{float(v):.{decimals}f}" if isinstance(v, (int, float)) else v)
-            )
+            lambda v: ("" if pd.isna(v) else (f"{float(v):.{decimals}f}" if isinstance(v, (int, float)) else v))
         )
     return df
 
 
+def dataframe_height(df: pd.DataFrame):
+    rows = 0 if df is None else int(getattr(df, "shape", (0, 0))[0])
+    row_px = 40
+    header_px = 44
+    padding_px = 96
+    height = header_px + (rows * row_px) + padding_px
+    return max(320, min(1200, height))
+
+
 def create_raw_score_df():
     raw_path = Path(__file__).resolve().parent / "src" / "raw_score.json"
     return create_score_df(raw_path, decimals=2)
@@ -220,8 +219,7 @@ def create_weights_table_html():
                 if (r, c) != (r1, c1):
                     covered.add((r, c))
 
-    parts = ["<div class=\"weights-scroll\"><table class=\"weights-table\">"
-    ]
+    parts = ['<div class="weights-scroll"><table class="weights-table">']
     for r_index, row in enumerate(rows, start=min_row):
         parts.append("<tr>")
         for c_index, value in enumerate(row, start=min_col):
@@ -247,7 +245,9 @@ def create_weights_table_html():
     parts.append("</table></div>")
     return "".join(parts)
 
+
 with demo:
+    gr.HTML(custom_js)
     gr.HTML(TITLE)
     gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
@@ -262,7 +262,7 @@ with demo:
                         wrap=True,
                         column_widths=column_widths,
                         row_count=(len(raw_score_df), "fixed"),
-                        height=700,
+                        height=dataframe_height(raw_score_df),
                         elem_id="raw-score-table",
                     )
                 with gr.TabItem("Weights"):
@@ -275,7 +275,7 @@ with demo:
                         wrap=True,
                         column_widths=column_widths,
                         row_count=(len(unweighted_z_df), "fixed"),
-                        height=700,
+                        height=dataframe_height(unweighted_z_df),
                         elem_id="unweighted-z-table",
                     )
                 with gr.TabItem("Weighted Z-score"):
@@ -286,7 +286,7 @@ with demo:
                         wrap=True,
                         column_widths=column_widths,
                         row_count=(len(weighted_z_df), "fixed"),
-                        height=700,
+                        height=dataframe_height(weighted_z_df),
                         elem_id="weighted-z-table",
                     )
 
@@ -294,9 +294,6 @@ with demo:
             gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
 
 
-
-
-
 scheduler = BackgroundScheduler()
 scheduler.add_job(restart_space, "interval", seconds=1800)
 scheduler.start()

src/about.py

@@ -1,6 +1,7 @@
 from dataclasses import dataclass
 from enum import Enum
 
+
 @dataclass
 class Task:
     benchmark: str
@@ -11,13 +12,13 @@ class Task:
 # 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 
+    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
     task0 = Task("anli_r1", "acc", "ANLI")
     task1 = Task("logiqa", "acc_norm", "LogiQA")
 
-NUM_FEWSHOT = 0 # Change with your few shot
-# ---------------------------------------------------
 
+NUM_FEWSHOT = 0  # Change with your few shot
+# ---------------------------------------------------
 
 
 # Your leaderboard name
@@ -28,7 +29,7 @@ INTRODUCTION_TEXT = """
 """
 
 # Which evaluations are you running? how can people reproduce what you have?
-LLM_BENCHMARKS_TEXT = f"""
+LLM_BENCHMARKS_TEXT = """
 ## About WebCoderBench
 
 Web applications (web apps) have become a key arena for large language models (LLMs) to demonstrate code generation capabilities and commercial potential. However, building a benchmark for LLM-generated web apps remains challenging due to:

src/display/css_html_js.py

@@ -99,10 +99,11 @@ custom_css = """
     overflow-x: auto !important;
     max-width: 100%;
     min-width: 0;
-    min-height: 78vh;
 }
 
-#raw-score-table > div {
+#raw-score-table,
+#unweighted-z-table,
+#weighted-z-table {
     overflow-x: auto !important;
     max-width: 100%;
     min-width: 0;
@@ -153,6 +154,69 @@ custom_css = """
 }
 """
 
+custom_js = """
+<script>
+(function () {
+    const tableIds = ["raw-score-table", "unweighted-z-table", "weighted-z-table"];
+
+    const nudgeScroll = (el) => {
+        const scroller =
+            el.querySelector(".table-wrap") ||
+            el.querySelector(".wrap") ||
+            el.querySelector(".wrap-inner") ||
+            el;
+
+        const x = scroller.scrollLeft || 0;
+        scroller.scrollLeft = x + 1;
+        scroller.scrollLeft = x;
+        scroller.dispatchEvent(new Event("scroll", { bubbles: true }));
+    };
+
+    const forceRelayout = () => {
+        window.dispatchEvent(new Event("resize"));
+        for (const id of tableIds) {
+            const el = document.getElementById(id);
+            if (!el) continue;
+            nudgeScroll(el);
+        }
+    };
+
+    const schedule = () => {
+        forceRelayout();
+        setTimeout(forceRelayout, 50);
+        setTimeout(forceRelayout, 200);
+        setTimeout(forceRelayout, 800);
+    };
+
+    const observer = new MutationObserver(() => {
+        schedule();
+    });
+    observer.observe(document.documentElement, {
+        subtree: true,
+        childList: true,
+        attributes: true,
+        attributeFilter: ["style", "class", "aria-selected"],
+    });
+
+    document.addEventListener(
+        "click",
+        (e) => {
+            const target = e.target;
+            if (!(target instanceof Element)) return;
+            if (target.closest("button, [role='tab'], .tab-buttons")) {
+                setTimeout(schedule, 0);
+            }
+        },
+        true
+    );
+
+    window.addEventListener("resize", () => setTimeout(forceRelayout, 0));
+    window.addEventListener("load", () => setTimeout(schedule, 0));
+    schedule();
+})();
+</script>
+"""
+
 get_window_url_params = """
     function(url_params) {
         const params = new URLSearchParams(window.location.search);

src/display/utils.py

@@ -1,10 +1,9 @@
 from dataclasses import dataclass, make_dataclass
 from enum import Enum
 
-import pandas as pd
-
 from src.about import Tasks
 
+
 def fields(raw_class):
     return [v for k, v in raw_class.__dict__.items() if k[:2] != "__" and k[-2:] != "__"]
 
@@ -20,12 +19,13 @@ class ColumnContent:
     hidden: bool = False
     never_hidden: bool = False
 
+
 ## Leaderboard columns
 auto_eval_column_dict = []
 # Init
 auto_eval_column_dict.append(["model_type_symbol", ColumnContent, ColumnContent("T", "str", True, never_hidden=True)])
 auto_eval_column_dict.append(["model", ColumnContent, ColumnContent("Model", "markdown", True, never_hidden=True)])
-#Scores
+# Scores
 auto_eval_column_dict.append(["average", ColumnContent, ColumnContent("Average ⬆️", "number", True)])
 for task in Tasks:
     auto_eval_column_dict.append([task.name, ColumnContent, ColumnContent(task.value.col_name, "number", True)])
@@ -43,6 +43,7 @@ auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sh
 # We use make dataclass to dynamically fill the scores from Tasks
 AutoEvalColumn = make_dataclass("AutoEvalColumn", auto_eval_column_dict, frozen=True)
 
+
 ## For the queue columns in the submission tab
 @dataclass(frozen=True)
 class EvalQueueColumn:  # Queue column
@@ -53,12 +54,13 @@ class EvalQueueColumn:  # Queue column
     weight_type = ColumnContent("weight_type", "str", "Original")
     status = ColumnContent("status", "str", True)
 
+
 ## All the model information that we might need
 @dataclass
 class ModelDetails:
     name: str
     display_name: str = ""
-    symbol: str = "" # emoji
+    symbol: str = ""  # emoji
 
 
 class ModelType(Enum):
@@ -83,11 +85,13 @@ class ModelType(Enum):
             return ModelType.IFT
         return ModelType.Unknown
 
+
 class WeightType(Enum):
     Adapter = ModelDetails("Adapter")
     Original = ModelDetails("Original")
     Delta = ModelDetails("Delta")
 
+
 class Precision(Enum):
     float16 = ModelDetails("float16")
     bfloat16 = ModelDetails("bfloat16")
@@ -100,6 +104,7 @@ class Precision(Enum):
             return Precision.bfloat16
         return Precision.Unknown
 
+
 # Column selection
 COLS = [c.name for c in fields(AutoEvalColumn) if not c.hidden]
 
@@ -107,4 +112,3 @@ EVAL_COLS = [c.name for c in fields(EvalQueueColumn)]
 EVAL_TYPES = [c.type for c in fields(EvalQueueColumn)]
 
 BENCHMARK_COLS = [t.value.col_name for t in Tasks]
-

src/envs.py

@@ -4,9 +4,9 @@ from huggingface_hub import HfApi
 
 # Info to change for your repository
 # ----------------------------------
-TOKEN = os.environ.get("HF_TOKEN") # A read/write token for your org
+TOKEN = os.environ.get("HF_TOKEN")  # A read/write token for your org
 
-OWNER = "demo-leaderboard-backend" # Change to your org - don't forget to create a results and request dataset, with the correct format!
+OWNER = "demo-leaderboard-backend"  # Change to your org - don't forget to create a results and request dataset, with the correct format!
 # ----------------------------------
 
 REPO_ID = f"{OWNER}/leaderboard"
@@ -14,7 +14,7 @@ QUEUE_REPO = f"{OWNER}/requests"
 RESULTS_REPO = f"{OWNER}/results"
 
 # If you setup a cache later, just change HF_HOME
-CACHE_PATH=os.getenv("HF_HOME", ".")
+CACHE_PATH = os.getenv("HF_HOME", ".")
 
 # Local caches
 EVAL_REQUESTS_PATH = os.path.join(CACHE_PATH, "eval-queue")

src/leaderboard/read_evals.py

@@ -1,6 +1,5 @@
 import glob
 import json
-import math
 import os
 from dataclasses import dataclass
 
@@ -8,28 +7,28 @@ import dateutil
 import numpy as np
 
 from src.display.formatting import make_clickable_model
-from src.display.utils import AutoEvalColumn, ModelType, Tasks, Precision, WeightType
+from src.display.utils import AutoEvalColumn, ModelType, Precision, Tasks, WeightType
 from src.submission.check_validity import is_model_on_hub
 
 
 @dataclass
 class EvalResult:
-    """Represents one full evaluation. Built from a combination of the result and request file for a given run.
-    """
-    eval_name: str # org_model_precision (uid)
-    full_model: str # org/model (path on hub)
-    org: str 
+    """Represents one full evaluation. Built from a combination of the result and request file for a given run."""
+
+    eval_name: str  # org_model_precision (uid)
+    full_model: str  # org/model (path on hub)
+    org: str
     model: str
-    revision: str # commit hash, "" if main
+    revision: str  # commit hash, "" if main
     results: dict
     precision: Precision = Precision.Unknown
-    model_type: ModelType = ModelType.Unknown # Pretrained, fine tuned, ...
-    weight_type: WeightType = WeightType.Original # Original or Adapter
-    architecture: str = "Unknown" 
+    model_type: ModelType = ModelType.Unknown  # Pretrained, fine tuned, ...
+    weight_type: WeightType = WeightType.Original  # Original or Adapter
+    architecture: str = "Unknown"
     license: str = "?"
     likes: int = 0
     num_params: int = 0
-    date: str = "" # submission date of request file
+    date: str = ""  # submission date of request file
     still_on_hub: bool = False
 
     @classmethod
@@ -85,10 +84,10 @@ class EvalResult:
             org=org,
             model=model,
             results=results,
-            precision=precision,  
-            revision= config.get("model_sha", ""),
+            precision=precision,
+            revision=config.get("model_sha", ""),
             still_on_hub=still_on_hub,
-            architecture=architecture
+            architecture=architecture,
         )
 
     def update_with_request_file(self, requests_path):
@@ -105,7 +104,9 @@ class EvalResult:
             self.num_params = request.get("params", 0)
             self.date = request.get("submitted_time", "")
         except Exception:
-            print(f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}")
+            print(
+                f"Could not find request file for {self.org}/{self.model} with precision {self.precision.value.name}"
+            )
 
     def to_dict(self):
         """Converts the Eval Result to a dict compatible with our dataframe display"""
@@ -146,10 +147,7 @@ def get_request_file_for_model(requests_path, model_name, precision):
     for tmp_request_file in request_files:
         with open(tmp_request_file, "r") as f:
             req_content = json.load(f)
-            if (
-                req_content["status"] in ["FINISHED"]
-                and req_content["precision"] == precision.split(".")[-1]
-            ):
+            if req_content["status"] in ["FINISHED"] and req_content["precision"] == precision.split(".")[-1]:
                 request_file = tmp_request_file
     return request_file
 
@@ -188,7 +186,7 @@ def get_raw_eval_results(results_path: str, requests_path: str) -> list[EvalResu
     results = []
     for v in eval_results.values():
         try:
-            v.to_dict() # we test if the dict version is complete
+            v.to_dict()  # we test if the dict version is complete
             results.append(v)
         except KeyError:  # not all eval values present
             continue

src/populate.py

@@ -39,7 +39,9 @@ def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
             all_evals.append(data)
         elif ".md" not in entry:
             # this is a folder
-            sub_entries = [e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")]
+            sub_entries = [
+                e for e in os.listdir(f"{save_path}/{entry}") if os.path.isfile(e) and not e.startswith(".")
+            ]
             for sub_entry in sub_entries:
                 file_path = os.path.join(save_path, entry, sub_entry)
                 with open(file_path) as fp:

src/submission/check_validity.py

@@ -1,8 +1,6 @@
 import json
 import os
-import re
 from collections import defaultdict
-from datetime import datetime, timedelta, timezone
 
 import huggingface_hub
 from huggingface_hub import ModelCard
@@ -10,6 +8,7 @@ from huggingface_hub.hf_api import ModelInfo
 from transformers import AutoConfig
 from transformers.models.auto.tokenization_auto import AutoTokenizer
 
+
 def check_model_card(repo_id: str) -> tuple[bool, str]:
     """Checks if the model card and license exist and have been filled"""
     try:
@@ -31,31 +30,38 @@ def check_model_card(repo_id: str) -> tuple[bool, str]:
 
     return True, ""
 
-def is_model_on_hub(model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False) -> tuple[bool, str]:
+
+def is_model_on_hub(
+    model_name: str, revision: str, token: str = None, trust_remote_code=False, test_tokenizer=False
+) -> tuple[bool, str]:
     """Checks if the model model_name is on the hub, and whether it (and its tokenizer) can be loaded with AutoClasses."""
     try:
-        config = AutoConfig.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+        config = AutoConfig.from_pretrained(
+            model_name, revision=revision, trust_remote_code=trust_remote_code, token=token
+        )
         if test_tokenizer:
             try:
-                tk = AutoTokenizer.from_pretrained(model_name, revision=revision, trust_remote_code=trust_remote_code, token=token)
+                AutoTokenizer.from_pretrained(
+                    model_name, revision=revision, trust_remote_code=trust_remote_code, token=token
+                )
             except ValueError as e:
+                return (False, f"uses a tokenizer which is not in a transformers release: {e}", None)
+            except Exception:
                 return (
                     False,
-                    f"uses a tokenizer which is not in a transformers release: {e}",
-                    None
+                    "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?",
+                    None,
                 )
-            except Exception as e:
-                return (False, "'s tokenizer cannot be loaded. Is your tokenizer class in a stable transformers release, and correctly configured?", None)
         return True, None, config
 
     except ValueError:
         return (
             False,
             "needs to be launched with `trust_remote_code=True`. For safety reason, we do not allow these models to be automatically submitted to the leaderboard.",
-            None
+            None,
         )
 
-    except Exception as e:
+    except Exception:
         return False, "was not found on hub!", None
 
 
@@ -70,10 +76,12 @@ def get_model_size(model_info: ModelInfo, precision: str):
     model_size = size_factor * model_size
     return model_size
 
+
 def get_model_arch(model_info: ModelInfo):
     """Gets the model architecture from the configuration"""
     return model_info.config.get("architectures", "Unknown")
 
+
 def already_submitted_models(requested_models_dir: str) -> set[str]:
     """Gather a list of already submitted models to avoid duplicates"""
     depth = 1

src/submission/submit.py

@@ -3,7 +3,7 @@ import os
 from datetime import datetime, timezone
 
 from src.display.formatting import styled_error, styled_message, styled_warning
-from src.envs import API, EVAL_REQUESTS_PATH, TOKEN, QUEUE_REPO
+from src.envs import API, EVAL_REQUESTS_PATH, QUEUE_REPO, TOKEN
 from src.submission.check_validity import (
     already_submitted_models,
     check_model_card,
@@ -14,6 +14,7 @@ from src.submission.check_validity import (
 REQUESTED_MODELS = None
 USERS_TO_SUBMISSION_DATES = None
 
+
 def add_new_eval(
     model: str,
     base_model: str,
@@ -45,7 +46,9 @@ def add_new_eval(
 
     # Is the model on the hub?
     if weight_type in ["Delta", "Adapter"]:
-        base_model_on_hub, error, _ = is_model_on_hub(model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True)
+        base_model_on_hub, error, _ = is_model_on_hub(
+            model_name=base_model, revision=revision, token=TOKEN, test_tokenizer=True
+        )
         if not base_model_on_hub:
             return styled_error(f'Base model "{base_model}" {error}')