src/about.py

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("perplexity", "perplexity", "Perplexity")

NUM_FEWSHOT = 0 # Not used for perplexity
# ---------------------------------------------------

# Your leaderboard name
TITLE = """<h1 align="center" id="space-title">Model Perplexity Leaderboard</h1>"""

# What does your leaderboard evaluate?
INTRODUCTION_TEXT = """
This leaderboard evaluates language models based on their perplexity scores on a fixed test passage and 
structural similarity to GPT-2 using model tracing analysis.

- **Perplexity**: Lower perplexity scores indicate better performance - it means the model is better at predicting the next token in the text.
- **Match P-Value**: Lower p-values indicate the model preserves structural similarity to GPT-2 after fine-tuning (neuron organization is maintained).
"""

# Which evaluations are you running?
LLM_BENCHMARKS_TEXT = """
## How it works

The evaluation runs two types of analysis on language models:

### 1. Perplexity Evaluation
Perplexity tests using a fixed test passage about artificial intelligence.
Perplexity measures how well a model predicts text - lower scores mean better predictions.

### 2. Model Tracing Analysis
Compares each model's internal structure to GPT-2 using the "match" statistic with alignment:
- **Base Model**: GPT-2 (`openai-community/gpt2`)
- **Comparison**: Each model on the leaderboard  
- **Method**: Neuron matching analysis across transformer layers
- **Alignment**: Models are aligned before comparison using the Hungarian algorithm
- **Output**: P-value indicating structural similarity (lower = more similar to GPT-2)

The match statistic tests whether neurons in corresponding layers maintain similar functional roles 
between the base model and fine-tuned variants.

## Test Text

The evaluation uses the following passage:
```
Artificial intelligence has transformed the way we live and work, bringing both opportunities and challenges. 
From autonomous vehicles to language models that can engage in human-like conversation, AI technologies are becoming increasingly 
sophisticated. However, with this advancement comes the responsibility to ensure these systems are developed and deployed ethically, 
with careful consideration for privacy, fairness, and transparency. The future of AI will likely depend on how well we balance innovation 
with these important social considerations.
```
"""

EVALUATION_QUEUE_TEXT = """
## Before submitting a model

1. Make sure your model is public on the Hugging Face Hub
2. The model should be loadable with AutoModelForCausalLM
3. The model should support text generation tasks
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = ""