taxonomy_data.json

{"DeepSeek-V2-Lite": {"correct": 0.62, "layer_error": 0.075, "span_error": 0.0775, "wrong_color": 0.0775, "wrong_position": 0.0225, "other": 0.1275}, "Gemma-2-9B-Instruct": {"correct": 0.6125, "layer_error": 0.1, "span_error": 0.0575, "wrong_color": 0.055, "wrong_position": 0.005, "other": 0.17}, "Llama-3.1-8B-Instruct": {"correct": 0.6566, "layer_error": 0.101, "span_error": 0.0303, "wrong_color": 0.053, "wrong_position": 0.0051, "other": 0.154}, "Mistral-7B-Instruct": {"correct": 0.6649, "layer_error": 0.0825, "span_error": 0.0747, "wrong_color": 0.0464, "wrong_position": 0.0077, "other": 0.1237}, "Qwen2.5-Instruct-14B": {"correct": 0.5575, "layer_error": 0.1893, "span_error": 0.0793, "wrong_color": 0.046, "wrong_position": 0.0102, "other": 0.1176}, "Qwen2.5-Instruct-32B": {"correct": 0.4175, "layer_error": 0.2825, "span_error": 0.0925, "wrong_color": 0.035, "wrong_position": 0.0825, "other": 0.09}, "Qwen2.5-Instruct-72B": {"correct": 0.6115, "layer_error": 0.1529, "span_error": 0.0727, "wrong_color": 0.0426, "wrong_position": 0.015, "other": 0.1053}, "Qwen2.5-Instruct-7B": {"correct": 0.6872, "layer_error": 0.059, "span_error": 0.0282, "wrong_color": 0.0513, "wrong_position": 0.0154, "other": 0.159}, "Claude Sonnet-4.6": {"correct": 0.44, "layer_error": 0.2525, "span_error": 0.04, "wrong_color": 0.1, "wrong_position": 0.1275, "other": 0.04}, "Gemini 2.5-Flash": {"correct": 0.41, "layer_error": 0.2825, "span_error": 0.0475, "wrong_color": 0.075, "wrong_position": 0.155, "other": 0.03}, "Gemini 3-Flash": {"correct": 0.6987, "layer_error": 0.0519, "span_error": 0.0805, "wrong_color": 0.0078, "wrong_position": 0.0104, "other": 0.1506}, "Gemini 3.1-Flash-Lite": {"correct": 0.395, "layer_error": 0.36, "span_error": 0.045, "wrong_color": 0.055, "wrong_position": 0.0725, "other": 0.0725}, "GPT-4.1-Mini": {"correct": 0.4275, "layer_error": 0.2775, "span_error": 0.06, "wrong_color": 0.0525, "wrong_position": 0.085, "other": 0.0975}, "GPT-4o-Mini": {"correct": 0.48, "layer_error": 0.2625, "span_error": 0.06, "wrong_color": 0.0625, "wrong_position": 0.0475, "other": 0.0875}, "GPT-4o": {"correct": 0.6599, "layer_error": 0.1184, "span_error": 0.063, "wrong_color": 0.0403, "wrong_position": 0.0126, "other": 0.1058}}