version v0.1

.gradio/certificate.pem

@@ -0,0 +1,31 @@
+-----BEGIN CERTIFICATE-----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=
+-----END CERTIFICATE-----

README.md

@@ -1,48 +0,0 @@
----
-title: SearchAgent Leaderboard
-emoji: 🥇
-colorFrom: green
-colorTo: indigo
-sdk: gradio
-app_file: app.py
-pinned: true
-license: apache-2.0
-short_description: Duplicate this leaderboard to initialize your own!
-sdk_version: 5.43.1
-tags:
-- leaderboard
----
-
-# Start the configuration
-
-Most of the variables to change for a default leaderboard are in `src/env.py` (replace the path for your leaderboard) and `src/about.py` (for tasks).
-
-Results files should have the following format and be stored as json files:
-```json
-{
-    "config": {
-        "model_dtype": "torch.float16", # or torch.bfloat16 or 8bit or 4bit
-        "model_name": "path of the model on the hub: org/model",
-        "model_sha": "revision on the hub",
-    },
-    "results": {
-        "task_name": {
-            "metric_name": score,
-        },
-        "task_name2": {
-            "metric_name": score,
-        }
-    }
-}
-```
-
-Request files are created automatically by this tool.
-
-If you encounter problem on the space, don't hesitate to restart it to remove the create eval-queue, eval-queue-bk, eval-results and eval-results-bk created folder.
-
-# Code logic for more complex edits
-
-You'll find 
-- the main table' columns names and properties in `src/display/utils.py`
-- the logic to read all results and request files, then convert them in dataframe lines, in `src/leaderboard/read_evals.py`, and `src/populate.py`
-- the logic to allow or filter submissions in `src/submission/submit.py` and `src/submission/check_validity.py`

app.py

@@ -3,6 +3,10 @@ from gradio_leaderboard import Leaderboard, ColumnFilter, SelectColumns
 import pandas as pd
 from apscheduler.schedulers.background import BackgroundScheduler
 from huggingface_hub import snapshot_download
+import json
+import os
+import datetime
+import urllib.parse
 
 from src.about import (
     CITATION_BUTTON_LABEL,
@@ -22,7 +26,7 @@ from src.display.utils import (
     ModelType,
     fields,
     WeightType,
-    Precision
+    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
@@ -32,21 +36,95 @@ from src.submission.submit import add_new_eval
 def restart_space():
     API.restart_space(repo_id=REPO_ID)
 
+
+def save_submission_and_notify(model_name, contact_email, weight_link, json_results, paper_link, description):
+    """Save submission to file and provide instructions for email"""
+    try:
+        # Validate JSON format if provided
+        if json_results.strip():
+            try:
+                json.loads(json_results)
+            except json.JSONDecodeError:
+                return "❌ Invalid JSON format in results field"
+
+        # Create submission data
+        submission_data = {
+            "timestamp": datetime.datetime.now().isoformat(),
+            "model_name": model_name,
+            "contact_email": contact_email,
+            "weight_link": weight_link,
+            "paper_link": paper_link,
+            "description": description,
+            "json_results": json_results,
+        }
+
+        # Save to submissions directory
+        os.makedirs("submissions", exist_ok=True)
+        filename = (
+            f"submissions/{model_name.replace('/', '_')}_{datetime.datetime.now().strftime('%Y%m%d_%H%M%S')}.json"
+        )
+
+        with open(filename, "w") as f:
+            json.dump(submission_data, f, indent=2)
+
+        # Create mailto link for user
+        subject = f"SearchAgent Leaderboard Submission: {model_name}"
+        body = f"""New model submission for SearchAgent Leaderboard:
+
+Model Name: {model_name}
+Contact Email: {contact_email}
+Weight Link: {weight_link}
+Paper Link: {paper_link}
+Description: {description}
+
+JSON Results:
+{json_results}"""
+
+        # URL encode the email content
+        mailto_link = (
+            f"mailto:[email protected]?subject={urllib.parse.quote(subject)}&body={urllib.parse.quote(body[:500])}"
+        )
+
+        return f"""✅ Submission saved successfully!
+        
+📧 **Please send your submission to: [email protected]**
+
+You can either:
+1. Click here to open your email client: [Send Email](mailto:[email protected])
+2. Or copy the submission details above and send manually
+
+Your submission has been saved to: {filename}
+
+We'll review your model and get back to you at {contact_email}."""
+
+    except Exception as e:
+        return f"❌ Failed to save submission: {str(e)}"
+
+
 ### Space initialisation
+# Use local data for demo purposes
 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
-    )
-except Exception:
-    restart_space()
+    # For demo, use local eval-queue directory if it exists
+    import os
+
+    if not os.path.exists(EVAL_REQUESTS_PATH):
+        os.makedirs(EVAL_REQUESTS_PATH, exist_ok=True)
+    # snapshot_download(
+    #     repo_id=QUEUE_REPO, local_dir=EVAL_REQUESTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+    # )
+except Exception as e:
+    print(f"Could not setup eval requests path: {e}")
 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
-    )
-except Exception:
-    restart_space()
+    # For demo, use local eval-results directory if it exists
+    if not os.path.exists(EVAL_RESULTS_PATH):
+        os.makedirs(EVAL_RESULTS_PATH, exist_ok=True)
+    # snapshot_download(
+    #     repo_id=RESULTS_REPO, local_dir=EVAL_RESULTS_PATH, repo_type="dataset", tqdm_class=None, etag_timeout=30, token=TOKEN
+    # )
+except Exception as e:
+    print(f"Could not setup eval results path: {e}")
 
 
 LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS, BENCHMARK_COLS)
@@ -57,6 +135,7 @@ LEADERBOARD_DF = get_leaderboard_df(EVAL_RESULTS_PATH, EVAL_REQUESTS_PATH, COLS,
     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.")
@@ -68,21 +147,10 @@ def init_leaderboard(dataframe):
             cant_deselect=[c.name for c in fields(AutoEvalColumn) if c.never_hidden],
             label="Select Columns to Display:",
         ),
-        search_columns=[AutoEvalColumn.model.name, AutoEvalColumn.license.name],
+        search_columns=[AutoEvalColumn.model.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.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=True
-            ),
+            ColumnFilter(AutoEvalColumn.model_size.name, type="checkboxgroup", label="Model Size"),
         ],
         bool_checkboxgroup_label="Hide models",
         interactive=False,
@@ -92,101 +160,18 @@ def init_leaderboard(dataframe):
 demo = gr.Blocks(css=custom_css)
 with demo:
     gr.HTML(TITLE)
-    gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
     with gr.Tabs(elem_classes="tab-buttons") as tabs:
-        with gr.TabItem("🏅 LLM Benchmark", elem_id="llm-benchmark-tab-table", id=0):
+        with gr.TabItem("🏅 SearchAgent Benchmark", elem_id="llm-benchmark-tab-table", id=0):
             leaderboard = init_leaderboard(LEADERBOARD_DF)
+            gr.Markdown(INTRODUCTION_TEXT, elem_classes="markdown-text")
 
         with gr.TabItem("📝 About", elem_id="llm-benchmark-tab-table", id=2):
             gr.Markdown(LLM_BENCHMARKS_TEXT, elem_classes="markdown-text")
 
-        with gr.TabItem("🚀 Submit here! ", elem_id="llm-benchmark-tab-table", id=3):
+        with gr.TabItem("📤 Submit Model", elem_id="llm-benchmark-tab-table", id=3):
             with gr.Column():
-                with gr.Row():
-                    gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
-
-                with gr.Column():
-                    with gr.Accordion(
-                        f"✅ Finished Evaluations ({len(finished_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            finished_eval_table = gr.components.Dataframe(
-                                value=finished_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-                    with gr.Accordion(
-                        f"🔄 Running Evaluation Queue ({len(running_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            running_eval_table = gr.components.Dataframe(
-                                value=running_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-
-                    with gr.Accordion(
-                        f"⏳ Pending Evaluation Queue ({len(pending_eval_queue_df)})",
-                        open=False,
-                    ):
-                        with gr.Row():
-                            pending_eval_table = gr.components.Dataframe(
-                                value=pending_eval_queue_df,
-                                headers=EVAL_COLS,
-                                datatype=EVAL_TYPES,
-                                row_count=5,
-                            )
-            with gr.Row():
-                gr.Markdown("# ✉️✨ Submit your model here!", elem_classes="markdown-text")
-
-            with gr.Row():
-                with gr.Column():
-                    model_name_textbox = gr.Textbox(label="Model name")
-                    revision_name_textbox = gr.Textbox(label="Revision commit", placeholder="main")
-                    model_type = gr.Dropdown(
-                        choices=[t.to_str(" : ") for t in ModelType if t != ModelType.Unknown],
-                        label="Model type",
-                        multiselect=False,
-                        value=None,
-                        interactive=True,
-                    )
-
-                with gr.Column():
-                    precision = gr.Dropdown(
-                        choices=[i.value.name for i in Precision if i != Precision.Unknown],
-                        label="Precision",
-                        multiselect=False,
-                        value="float16",
-                        interactive=True,
-                    )
-                    weight_type = gr.Dropdown(
-                        choices=[i.value.name for i in WeightType],
-                        label="Weights type",
-                        multiselect=False,
-                        value="Original",
-                        interactive=True,
-                    )
-                    base_model_name_textbox = gr.Textbox(label="Base model (for delta or adapter weights)")
-
-            submit_button = gr.Button("Submit Eval")
-            submission_result = gr.Markdown()
-            submit_button.click(
-                add_new_eval,
-                [
-                    model_name_textbox,
-                    base_model_name_textbox,
-                    revision_name_textbox,
-                    precision,
-                    weight_type,
-                    model_type,
-                ],
-                submission_result,
-            )
+                gr.Markdown(EVALUATION_QUEUE_TEXT, elem_classes="markdown-text")
 
     with gr.Row():
         with gr.Accordion("📙 Citation", open=False):
@@ -201,4 +186,4 @@ with demo:
 scheduler = BackgroundScheduler()
 scheduler.add_job(restart_space, "interval", seconds=1800)
 scheduler.start()
-demo.queue(default_concurrency_limit=40).launch()
+demo.queue(default_concurrency_limit=40).launch(share=True)

requirements.txt

@@ -1,16 +0,0 @@
-APScheduler
-black
-datasets
-gradio
-gradio[oauth]
-gradio_leaderboard==0.0.13
-gradio_client
-huggingface-hub>=0.18.0
-matplotlib
-numpy
-pandas
-python-dateutil
-tqdm
-transformers
-tokenizers>=0.15.0
-sentencepiece

src/about.py

@@ -1,6 +1,7 @@
 from dataclasses import dataclass
 from enum import Enum
 
+
 @dataclass
 class Task:
     benchmark: str
@@ -11,62 +12,192 @@ 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 
-    task0 = Task("anli_r1", "acc", "ANLI")
-    task1 = Task("logiqa", "acc_norm", "LogiQA")
-
-NUM_FEWSHOT = 0 # Change with your few shot
+    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
+    # General QA tasks
+    nq = Task("nq", "exact_match", "NQ")
+    triviaqa = Task("triviaqa", "exact_match", "TriviaQA")
+    popqa = Task("popqa", "exact_match", "PopQA")
+    # Multi-hop QA tasks
+    hotpotqa = Task("hotpotqa", "exact_match", "HotpotQA")
+    twowiki = Task("2wiki", "exact_match", "2wiki")
+    musique = Task("musique", "exact_match", "Musique")
+    bamboogle = Task("bamboogle", "exact_match", "Bamboogle")
+    fictionalhot = Task("fictionalhot", "exact_match", "FictionalHot")
+
+
+NUM_FEWSHOT = 0  # Change with your few shot
 # ---------------------------------------------------
 
 
-
 # Your leaderboard name
-TITLE = """<h1 align="center" id="space-title">Demo leaderboard</h1>"""
+TITLE = """<h1 align="center" id="space-title">🔍 SearchAgent Leaderboard</h1>"""
 
 # What does your leaderboard evaluate?
 INTRODUCTION_TEXT = """
-Intro text
+# 🔍 SearchAgent Leaderboard
+
+This leaderboard evaluates the performance of **search-augmented question answering systems** across various tasks, ranging from simple factual QA to complex multi-hop reasoning. Our evaluation addresses the inconsistency in experimental settings across prior works by providing a standardized comparison framework.
+
+## 📊 Evaluation Tasks
+
+We evaluate on a comprehensive set of benchmarks that test different aspects of search-augmented QA:
+
+### General QA (Set A)
+- **NQ**: Natural Questions - QA based on real Google search queries from Wikipedia
+- **TriviaQA**: Trivia questions requiring document-based answer extraction  
+- **PopQA**: Popular culture QA testing knowledge breadth and parametric vs. non-parametric memory
+
+### Multi-Hop QA (Set B)
+- **HotpotQA**: Complex QA requiring reasoning across multiple documents with explainable reasoning chains
+- **2wiki**: Multi-hop reasoning based on Wikipedia requiring compositional reasoning
+- **Musique**: Multi-step compositional reasoning QA via single-hop question composition
+- **Bamboogle**: Adversarial search QA designed to test compositionality gaps in language models
+
+### Novel Evaluation: FictionalHot
+- **FictionalHot**: A closed-world benchmark grounding questions in synthetic fictional entities to mitigate data contamination and enable reproducible evaluation. Questions are transformed from real-world scenarios to fictional ones while preserving reasoning structure.
+
+## 🎯 Evaluation Metrics
+Following standardized practices, we primarily use **Exact Match (EM)** as the main metric. A prediction is correct if its normalized string exactly matches any normalized reference answer (with lowercasing, punctuation removal, and whitespace normalization).
+
 """
 
 # Which evaluations are you running? how can people reproduce what you have?
 LLM_BENCHMARKS_TEXT = f"""
-## How it works
+## 🔬 Evaluation Methodology
 
-## Reproducibility
-To reproduce our results, here is the commands you can run:
+This leaderboard addresses the challenge of inconsistent experimental settings in search agent evaluation by providing standardized comparisons. Prior works vary significantly in:
 
-"""
+1. **Corpora**: From static Wikipedia snapshots (2018, 2019) to live Internet access
+2. **Test Sets**: Broad evaluation (Set A) vs. focused multi-hop evaluation (Set B)  
+3. **Training Regimes**: No training to multi-dataset fine-tuning approaches
+4. **Metrics**: Exact Match, F1, Substring matching, and LLM-as-a-judge evaluations
 
-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.
+## 📋 Dataset Details & Challenges
+
+### Data Contamination Problem
+A critical issue in current benchmarks is **data contamination**, where high scores may reflect memorized pretraining knowledge rather than genuine procedural reasoning capabilities.
+
+### Our Solution: FictionalHot
+We introduce **FictionalHot**, a novel closed-world benchmark that:
+- Grounds all questions in newly generated synthetic fictional entities
+- Uses a three-step construction pipeline: sampling → GPT-based entity replacement → synthetic document generation
+- Forces models to rely on procedural reasoning over provided documents
+- Enables reproducible evaluation with a fixed knowledge source
 
-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!
+### Benchmark Coverage
+- **Corpus**: 2018 Wikipedia snapshot for reproducibility
+- **Retrieval**: Top-k=3 with maximum T=4 tool-use turns per question
 
-### 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`!
+## 🔄 Experimental Setup
 
-### 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 🤗
+Following established practices, we:
+- Fine-tune on unified NQ + HotpotQA training data
+- Evaluate on Qwen2.5-3B-Instruct and Qwen2.5-7B-Instruct models  
+- Use E5 embeddings for retrieval backend
+- Apply standard Exact Match evaluation with string normalization
 
-### 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_QUEUE_TEXT = """
+## 📣 Model Submission via Community
+
+We now accept submissions via the Space's Community (Discussions). This keeps the process simple and transparent.
+
+- Go to the Community tab of this leaderboard Space:  
+  https://huggingface.co/spaces/TencentBAC/SearchAgent_Leaderboard
+- Create a new Discussion with title:  
+  `Submission: <YourMethod>-<model_name>-<model_size>`
+- Include the following in the post:
+  - Model weights link (HF or GitHub)
+  - Short method description
+  - Evaluation JSON (inline or attached)
+
+Example JSON:
+```json
+{
+  "config": {
+    "model_dtype": "torch.float16",
+    "model_name": "YourMethod-Qwen2.5-7b-Instruct",
+    "model_sha": "main"
+  },
+  "results": {
+    "nq": {"exact_match": 0.45},
+    "triviaqa": {"exact_match": 0.62},
+    "popqa": {"exact_match": 0.38},
+    "hotpotqa": {"exact_match": 0.41},
+    "2wiki": {"exact_match": 0.33},
+    "musique": {"exact_match": 0.15},
+    "bamboogle": {"exact_match": 0.28},
+    "fictionalhot": {"exact_match": 0.06}
+  }
+}
+```
+
+We will review your post and add your model to the leaderboard.
 """
 
 CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
 CITATION_BUTTON_TEXT = r"""
+% Key Search-Augmented QA Methods
+@article{luo2024search,
+  title={Search-o1: Agentic Search-Enhanced Large Reasoning Models},
+  author={Xiaoxi Li and Guanting Dong and Jiajie Jin and Yuyao Zhang and Yujia Zhou and Yutao Zhu and Peitian Zhang and Zhicheng Dou},
+  journal={arXiv preprint arXiv:2501.05366},
+  year={2025}
+}
+
+@article{songR1SearcherIncentivizingSearch2025,
+  title={R1-Searcher: Incentivizing the Search Capability in LLMs via Reinforcement Learning},
+  author={Song, Huatong and Jiang, Jinhao and Min, Yingqian and Chen, Jie and Chen, Zhipeng and Zhao, Wayne Xin and Fang, Lei and Wen, Ji-Rong},
+  journal={arXiv preprint arXiv:2503.05592},
+  year={2025}
+}
+
+@article{jin2025search,
+  title={Search-r1: Training llms to reason and leverage search engines with reinforcement learning},
+  author={Jin, Bowen and Zeng, Hansi and Yue, Zhenrui and Yoon, Jinsung and Arik, Sercan and Wang, Dong and Zamani, Hamed and Han, Jiawei},
+  journal={arXiv preprint arXiv:2503.09516},
+  year={2025}
+}
+
+@article{sunZeroSearchIncentivizeSearch2025,
+  title={ZeroSearch: Incentivize the Search Capability of LLMs without Searching},
+  author={Sun, Hao and Qiao, Zile and Guo, Jiayan and Fan, Xuanbo and Hou, Yingyan and Jiang, Yong and Xie, Pengjun and Zhang, Yan and Huang, Fei and Zhou, Jingren},
+  journal={arXiv preprint arXiv:2505.04588},
+  year={2025}
+}
+
+@article{zheng2025deepresearcher,
+  title={Deepresearcher: Scaling deep research via reinforcement learning in real-world environments},
+  author={Zheng, Yuxiang and Fu, Dayuan and Hu, Xiangkun and Cai, Xiaojie and Ye, Lyumanshan and Lu, Pengrui and Liu, Pengfei},
+  journal={arXiv preprint arXiv:2504.03160},
+  year={2025}
+}
+
+% Benchmark Datasets
+@article{kwiatkowskiNaturalQuestionsBenchmark2019,
+  title={Natural Questions: A Benchmark for Question Answering Research},
+  author={Kwiatkowski, Tom and Palomaki, Jennimaria and Redfield, Olivia and Collins, Michael and Parikh, Ankur and Alberti, Chris and Epstein, Danielle and Polosukhin, Illia and Devlin, Jacob and Lee, Kenton and others},
+  journal={Transactions of the Association for Computational Linguistics},
+  volume={7},
+  pages={453--466},
+  year={2019}
+}
+
+@article{yangHotpotQADatasetDiverse2018,
+  title={HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering},
+  author={Yang, Zhilin and Qi, Peng and Zhang, Saizheng and Bengio, Yoshua and Cohen, William and Salakhutdinov, Ruslan and Manning, Christopher D.},
+  booktitle={Proceedings of EMNLP},
+  year={2018}
+}
+
+@article{trivediMuSiQueMultihopQuestions2022,
+  title={MuSiQue: Multihop Questions via Single-hop Question Composition},
+  author={Trivedi, Harsh and Balasubramanian, Niranjan and Khot, Tushar and Sabharwal, Ashish},
+  journal={Transactions of the Association for Computational Linguistics},
+  volume={10},
+  pages={539--554},
+  year={2022}
+}
 """

src/display/formatting.py

@@ -3,8 +3,28 @@ def model_hyperlink(link, model_name):
 
 
 def make_clickable_model(model_name):
-    link = f"https://huggingface.co/{model_name}"
-    return model_hyperlink(link, model_name)
+    # Custom link mappings for each model
+    custom_links = {
+        "ReSeek-Qwen2.5-7b-Instruct": "https://your-custom-link.com/reseek-7b",
+        "ReSeek-Qwen2.5-3b-Instruct": "https://your-custom-link.com/reseek-3b",
+        "ZeroSearch-Qwen2.5-3b-Instruct": "https://huggingface.co/Alibaba-NLP/ZeroSearch_wiki_V2_Qwen2.5_3B_Instruct",
+        "ZeroSearch-Qwen2.5-7b-Instruct": "https://huggingface.co/Alibaba-NLP/ZeroSearch_wiki_V2_Qwen2.5_7B_Instruct",
+        "Search-R1-Qwen2.5-7b-Instruct": "https://huggingface.co/PeterJinGo/SearchR1-nq_hotpotqa_train-qwen2.5-7b-it-em-ppo",
+        "Search-R1-Qwen2.5-3b-Instruct": "https://huggingface.co/PeterJinGo/SearchR1-nq_hotpotqa_train-qwen2.5-3b-em-grpo",
+        "Search-o1-Qwen2.5-7b-Instruct": "https://github.com/RUC-NLPIR/Search-o1",
+        "RAG-Qwen2.5-7b-Instruct": "",
+        "R1-Qwen2.5-7b-Instruct": "",
+        "SFT-Qwen2.5-7b-Instruct": "",
+        "CoT-Qwen2.5-7b-Instruct": "",
+        "Direct-Inference-Qwen2.5-7b-Instruct": "",
+    }
+
+    if model_name in custom_links:
+        link = custom_links[model_name]
+        return model_hyperlink(link, model_name)
+    else:
+        # If no custom link, just return the model name
+        return model_name
 
 
 def styled_error(error):

src/display/utils.py

@@ -5,6 +5,7 @@ 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:] != "__"]
 
@@ -20,29 +21,23 @@ 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(["rank", ColumnContent, ColumnContent("Rank", "number", 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)])
 # Model information
-auto_eval_column_dict.append(["model_type", ColumnContent, ColumnContent("Type", "str", False)])
-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(["still_on_hub", ColumnContent, ColumnContent("Available on the hub", "bool", False)])
-auto_eval_column_dict.append(["revision", ColumnContent, ColumnContent("Model sha", "str", False, False)])
+auto_eval_column_dict.append(["model_size", ColumnContent, ColumnContent("Model Size", "str", True)])
 
 # 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 +48,13 @@ 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 +79,13 @@ 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 +98,7 @@ 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 +106,3 @@ 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

@@ -6,7 +6,7 @@ from huggingface_hub import HfApi
 # ----------------------------------
 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 = "searchagent-leaderboard" # Change to your org - don't forget to create a results and request dataset, with the correct format!
 # ----------------------------------
 
 REPO_ID = f"{OWNER}/leaderboard"

src/leaderboard/read_evals.py

@@ -14,22 +14,22 @@ 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
@@ -57,9 +57,12 @@ class EvalResult:
             result_key = f"{org}_{model}_{precision.value.name}"
         full_model = "/".join(org_and_model)
 
-        still_on_hub, _, model_config = is_model_on_hub(
-            full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
-        )
+        # For demo purposes, assume models are available
+        still_on_hub = True
+        model_config = None
+        # still_on_hub, _, model_config = is_model_on_hub(
+        #     full_model, config.get("model_sha", "main"), trust_remote_code=True, test_tokenizer=False
+        # )
         architecture = "?"
         if model_config is not None:
             architectures = getattr(model_config, "architectures", None)
@@ -85,10 +88,10 @@ 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):
@@ -109,25 +112,25 @@ class EvalResult:
 
     def to_dict(self):
         """Converts the Eval Result to a dict compatible with our dataframe display"""
-        average = sum([v for v in self.results.values() if v is not None]) / len(Tasks)
+        valid_results = [v for v in self.results.values() if v is not None]
+        average = sum(valid_results) / len(valid_results) if valid_results else 0
+        # Extract model size from model name
+        model_size = "Unknown"
+        if "3b" in self.full_model.lower():
+            model_size = "3B"
+        elif "7b" in self.full_model.lower():
+            model_size = "7B"
+
         data_dict = {
             "eval_name": self.eval_name,  # not a column, just a save name,
-            AutoEvalColumn.precision.name: self.precision.value.name,
-            AutoEvalColumn.model_type.name: self.model_type.value.name,
-            AutoEvalColumn.model_type_symbol.name: self.model_type.value.symbol,
-            AutoEvalColumn.weight_type.name: self.weight_type.value.name,
-            AutoEvalColumn.architecture.name: self.architecture,
+            AutoEvalColumn.rank.name: 0,  # Will be set later based on average ranking
             AutoEvalColumn.model.name: make_clickable_model(self.full_model),
-            AutoEvalColumn.revision.name: self.revision,
+            AutoEvalColumn.model_size.name: model_size,
             AutoEvalColumn.average.name: average,
-            AutoEvalColumn.license.name: self.license,
-            AutoEvalColumn.likes.name: self.likes,
-            AutoEvalColumn.params.name: self.num_params,
-            AutoEvalColumn.still_on_hub.name: self.still_on_hub,
         }
 
         for task in Tasks:
-            data_dict[task.value.col_name] = self.results[task.value.benchmark]
+            data_dict[task.value.col_name] = self.results.get(task.value.benchmark, None)
 
         return data_dict
 
@@ -146,10 +149,7 @@ 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 +188,7 @@ 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

@@ -15,6 +15,10 @@ def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchm
 
     df = pd.DataFrame.from_records(all_data_json)
     df = df.sort_values(by=[AutoEvalColumn.average.name], ascending=False)
+
+    # Add ranking based on average score
+    df[AutoEvalColumn.rank.name] = range(1, len(df) + 1)
+
     df = df[cols].round(decimals=2)
 
     # filter out if any of the benchmarks have not been produced