Dataset for: Evaluating Large Language Models for Automatic Short Answer Grading: Performance and Bias Analysis

# Thesis Bias-LLM Data This archive contains the results of the LLM bias evaluation experiments used in the thesis. ## Contents - `experiment-results/`: Contains the outputs for various models. Each subfolder represents an experiment run for a specific model (e.g., `20260626_234841-gpt-oss-120b-decompose`). ## Data Formats Inside the experiment folders, you will find files typically following this naming convention: - `*_responses.jsonl`: Contains the raw or parsed JSON Lines output from the models. Each line is a JSON object representing a prompt and the model's response. - `*.csv`: Contains aggregated metrics, scoring, or tabular extraction of the experiment results for easier analysis. Files are often split into experimental conditions, such as: - `bias_control_FULL`: The control condition. - `bias_behavior_female_FULL` (and similar): The condition where specific behavioral or demographic prompts were introduced to measure bias. ## Data Manifest (Mapping to Models) Below is the mapping of experiment folders to their respective models and evaluation modes (Single-agent vs. Decompose): | Experiment Folder | Model & Mode | | :--- | :--- | | `20260507_365828` | Qwen3.6-35B (Single-agent) | | `20260607_005526` | Qwen3.6-35B (Decompose) | | `20260518_221517_gemma4E4Bit` | Gemma-4-E4B-IT (Single-agent) | | `20260607_005701` | Gemma-4-E4B-IT (Decompose) | | `20260518_221100_mistralsmall3224b` | Mistral-Small-3.2-24B (Single-agent) | | `20260607_011216` | Mistral-Small-3.2-24B (Decompose) | | `20260518_215807_deepseekv4flash` | DeepSeek-V4-Flash (Single-agent) | | `20260607_180255` | DeepSeek-V4-Flash (Decompose) | | `20260507_435434` | Gemma-4-31B-IT (Single-agent) | | `20260606_125208` | Gemma-4-31B-IT (Decompose) | | `20260507_355616` | Qwen3.5-397B (Single-agent) | | `20260626_234226-qwen35397-decompose` | Qwen3.5-397B (Decompose) | | `20260522_230045_gptoss120b` | GPT-OSS-120B (Single-agent) | | `20260626_234841-gpt-oss-120b-decompose` | GPT-OSS-120B (Decompose) |