from gradio import Interface import gradio as gr import aranizer from aranizer import aranizer_bpe50k, aranizer_bpe64k, aranizer_bpe86k, aranizer_sp32k, aranizer_sp50k, aranizer_sp64k, aranizer_sp86k from transformers import AutoTokenizer, HfFolder, logging import os # Retrieve your Hugging Face token from the environment variable HF_TOKEN = os.getenv('HF_TOKEN') if HF_TOKEN: HfFolder.save_token(HF_TOKEN) os.environ["HUGGINGFACEHUB_API_TOKEN"] = HF_TOKEN # Load additional tokenizers from transformers gpt_13b_tokenizer = AutoTokenizer.from_pretrained("FreedomIntelligence/AceGPT-13B") gpt_7b_tokenizer = AutoTokenizer.from_pretrained("FreedomIntelligence/AceGPT-7B") jais_13b_tokenizer = AutoTokenizer.from_pretrained("inception-mbzuai/jais-13b") arabert_tokenizer = AutoTokenizer.from_pretrained("aubmindlab/bert-base-arabertv2") # Try to load the gated tokenizer try: meta_llama_tokenizer = AutoTokenizer.from_pretrained("meta-llama/Meta-Llama-3-8B") except Exception as e: meta_llama_tokenizer = None logging.warning(f"Could not load meta-llama/Meta-Llama-3-8B tokenizer: {e}") cohere_command_r_v01_tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-v01") cohere_command_r_plus_tokenizer = AutoTokenizer.from_pretrained("CohereForAI/c4ai-command-r-plus") # List of available tokenizers and a dictionary to load them tokenizer_options = [ "aranizer_bpe50k", "aranizer_bpe64k", "aranizer_bpe86k", "aranizer_sp32k", "aranizer_sp50k", "aranizer_sp64k", "aranizer_sp86k", "FreedomIntelligence/AceGPT-13B", "FreedomIntelligence/AceGPT-7B", "inception-mbzuai/jais-13b", "aubmindlab/bert-base-arabertv2", "CohereForAI/c4ai-command-r-v01", "CohereForAI/c4ai-command-r-plus" ] if meta_llama_tokenizer: tokenizer_options.append("meta-llama/Meta-Llama-3-8B") tokenizers = { "aranizer_bpe50k": aranizer_bpe50k.get_tokenizer, "aranizer_bpe64k": aranizer_bpe64k.get_tokenizer, "aranizer_bpe86k": aranizer_bpe86k.get_tokenizer, "aranizer_sp32k": aranizer_sp32k.get_tokenizer, "aranizer_sp50k": aranizer_sp50k.get_tokenizer, "aranizer_sp64k": aranizer_sp64k.get_tokenizer, "aranizer_sp86k": aranizer_sp86k.get_tokenizer, "FreedomIntelligence/AceGPT-13B": lambda: gpt_13b_tokenizer, "FreedomIntelligence/AceGPT-7B": lambda: gpt_7b_tokenizer, "inception-mbzuai/jais-13b": lambda: jais_13b_tokenizer, "aubmindlab/bert-base-arabertv2": lambda: arabert_tokenizer, "CohereForAI/c4ai-command-r-v01": lambda: cohere_command_r_v01_tokenizer, "CohereForAI/c4ai-command-r-plus": lambda: cohere_command_r_plus_tokenizer } if meta_llama_tokenizer: tokenizers["meta-llama/Meta-Llama-3-8B"] = lambda: meta_llama_tokenizer def compare_tokenizers(tokenizer_name, text): # Handle the transformer tokenizers separately due to API differences if tokenizer_name in [ "FreedomIntelligence/AceGPT-13B", "FreedomIntelligence/AceGPT-7B", "inception-mbzuai/jais-13b", "aubmindlab/bert-base-arabertv2", "meta-llama/Meta-Llama-3-8B", "CohereForAI/c4ai-command-r-v01", "CohereForAI/c4ai-command-r-plus" ]: tokenizer = tokenizers[tokenizer_name]() tokens = tokenizer.tokenize(text) encoded_output = tokenizer.encode(text, add_special_tokens=True) decoded_text = tokenizer.decode(encoded_output, skip_special_tokens=True) else: # AraNizer tokenizers tokenizer = tokenizers[tokenizer_name]() tokens = tokenizer.tokenize(text) encoded_output = tokenizer.encode(text, add_special_tokens=True) decoded_text = tokenizer.decode(encoded_output) # Prepare the results to be displayed in HTML format tokens_arabic = [token.encode('utf-8').decode('utf-8') if isinstance(token, bytes) else token for token in tokens] results_html = f"""
Tokens: {tokens_arabic}
Encoded: {encoded_output}
Decoded: {decoded_text}