TELEIA Datasets Leaderboard

These dataset contains the answers of different LLMs to the TELEIA (Spanish Language Benchmark for Artificial Intelligence Models) dataset.

LLMs evaluated:

• Yi-6B-Chat
• Meta-Llama-3-8B-Instruct
• Llama-2-7b-chat-hf
• gemma-7b-it
• Mistral-7B-Instruct-v0.1
• occiglot-7b-es-en-instruct
• GPT3.5
• GPT4

Files:

• TELEIA_Cervantes_AVE_results.xlsx: vocabulary and grammatical structures, following the format of the Cervantes AVE exam
• TELEIA_PCE_results.xlsx: test on morphology and semantics resembling the style of the PCE exam, consisting of short questions or sentences to be completed
• TELEIA_SIELE_results.xlsx: different texts with questions related to them, based on the reading comprehension task of the SIELE exam

Each .xlsx contains a sheet with the results of each model and the following columns:

• question: question from TELEIA
• option_a: possible answer from TELEIA    
• option_b: possible answer from TELEIA       
• option_c: possible answer from TELEIA        
• option_d: possible answer from TELEIA        
• correct_answer: correct answer form TELEIA  
• llm_question: complete question made to the LLM    
• tokens_in: list of tokens that compound the question    
• tokens_in_count: number of tokens that compound the question    
• llm_answer: raw answer from the LLM    
• llm_answer_filtered: answer in format {A,B,C,D} from the LLM    
• tokens_out : list of tokens that compound the raw answer    
• tokens_out_count: number of tokens that compound the raw answer    
• word_count :  number of words that compound the raw answer