torahCodes/Torah_Codes

1from transformers import AutoTokenizer, AutoModelForTokenClassification, pipeline2from diffusers import DiffusionPipeline3from transformers import AutoModelForSeq2SeqLM4from samplings import top_p_sampling, temperature_sampling5import torch6from sentence_transformers import SentenceTransformer, util7from datasets import load_dataset8import soundfile as sf9import unicodedata10import itertools11 12 13class AIAssistant:14    def __init__(self):15        pass16 17 18    ## generate regexp for search over memory19    def gen_search_expr(self,palabras_unidas):20 21        combinaciones = []22 23        for i in range(1, len(palabras_unidas) + 1):24            for combinacion in itertools.combinations(palabras_unidas, i):25                regex = ".*?".join(combinacion)26                combinaciones.append(regex)27 28        return combinaciones29 30    ## join taggued tokens into words31    def process_list(self,lista):32        palabras_unidas = []33        palabra_actual = ""34 35        for token in lista:36            if token.startswith("##"):37                palabra_actual += token[2:]38            else:39                if palabra_actual:40                    palabras_unidas.append(palabra_actual)41                    palabra_actual = ""42                palabra_actual += token43 44        if palabra_actual:45            palabras_unidas.append(palabra_actual)46 47        return [unicodedata.normalize("NFKD", palabra).encode("ASCII", "ignore").decode("ASCII").lower() for palabra in palabras_unidas]48 49   50    ## gramatical classificator51    def grammatical_pos_tagger(self, text):52        nlp_pos = pipeline("token-classification", model="QCRI/bert-base-multilingual-cased-pos-english", tokenizer="QCRI/bert-base-multilingual-cased-pos-english")53        res = nlp_pos(text)54        return res55 56 57    ## entity classifier58    def entity_pos_tagger(self, txt):59        tokenizer = AutoTokenizer.from_pretrained("Davlan/bert-base-multilingual-cased-ner-hrl")60        model = AutoModelForTokenClassification.from_pretrained("Davlan/bert-base-multilingual-cased-ner-hrl")61        nlp = pipeline("ner", model=model, tokenizer=tokenizer)62        ner_results = nlp(txt)63        return ner_results64 65 66    ## sentiment analysis67    def sentiment_tags(self,text):68        distilled_student_sentiment_classifier = pipeline(69            model="lxyuan/distilbert-base-multilingual-cased-sentiments-student", 70            return_all_scores=True71        )72 73        # english74        return distilled_student_sentiment_classifier(text)75 76    ## check similarity among sentences (group of tokens (words))77    def similarity_tag(self, sentenceA,sentenceB):78        res=[]79        model = SentenceTransformer('abbasgolestani/ag-nli-bert-mpnet-base-uncased-sentence-similarity-v1') 80 81        # Two lists of sentences82        #sentences1 = ['I am honored to be given the opportunity to help make our company better',83        #            'I love my job and what I do here',84        #            'I am excited about our company’s vision']85 86        #sentences2 = ['I am hopeful about the future of our company',87        #            'My work is aligning with my passion',88        #            'Definitely our company vision will be the next breakthrough to change the world and I’m so happy and proud to work here']89 90        sentences1 = sentenceA91        sentences2 = sentenceB92        #Compute embedding for both lists93        embeddings1 = model.encode(sentences1, convert_to_tensor=True)94        embeddings2 = model.encode(sentences2, convert_to_tensor=True)95 96        #Compute cosine-similarities97        cosine_scores = util.cos_sim(embeddings1, embeddings2)98 99        #Output the pairs with their score100        for i in range(len(sentences1)):101            try:102                res.append({"A": sentences1[i], "B":sentences2[i], "score":cosine_scores[i][i]})103            except:104                pass105 106            #print("{} \t\t {} \t\t Score: {:.4f}".format(sentences1[i], sentences2[i], cosine_scores[i][i]))107 108        return res109 110 111 112    ## text to speech113    def texto_to_speech(self,txt):    114        synthesiser = pipeline("text-to-speech", "microsoft/speecht5_tts")115 116        embeddings_dataset = load_dataset("Matthijs/cmu-arctic-xvectors", split="validation")117        speaker_embedding = torch.tensor(embeddings_dataset[7306]["xvector"]).unsqueeze(0)118        # You can replace this embedding with your own as well.119 120        speech = synthesiser(txt, forward_params={"speaker_embeddings": speaker_embedding})121        sf.write("speech.wav", speech["audio"], samplerate=speech["sampling_rate"])122 123        return speech   124    ## text to stable difusor generated image125    def text_to_image_generation(self, prompt, n_steps=40, high_noise_frac=0.8):126        base = DiffusionPipeline.from_pretrained(127            "stabilityai/stable-diffusion-xl-base-1.0", torch_dtype=torch.float16, variant="fp16", use_safetensors=True128        )129        base.to("cuda")130        refiner = DiffusionPipeline.from_pretrained(131            "stabilityai/stable-diffusion-xl-refiner-1.0",132            text_encoder_2=base.text_encoder_2,133            vae=base.vae,134            torch_dtype=torch.float16,135            use_safetensors=True,136            variant="fp16",137        )138        refiner.to("cuda")139 140        image = base(141            prompt=prompt,142            num_inference_steps=n_steps,143            denoising_end=high_noise_frac,144            output_type="latent",145        ).images146        image = refiner(147            prompt=prompt,148            num_inference_steps=n_steps,149            denoising_start=high_noise_frac,150            image=image,151        ).images[0]152        return image153 154 155    ## pass text prompt to music156    def text_to_music(self, text, max_length=1024, top_p=0.9, temperature=1.0):157        tokenizer = AutoTokenizer.from_pretrained('sander-wood/text-to-music')158        model = AutoModelForSeq2SeqLM.from_pretrained('sander-wood/text-to-music')159 160        input_ids = tokenizer(text,161                              return_tensors='pt',162                              truncation=True,163                              max_length=max_length)['input_ids']164 165        decoder_start_token_id = model.config.decoder_start_token_id166        eos_token_id = model.config.eos_token_id167 168        decoder_input_ids = torch.tensor([[decoder_start_token_id]])169 170        for t_idx in range(max_length):171            outputs = model(input_ids=input_ids,172                            decoder_input_ids=decoder_input_ids)173            probs = outputs.logits[0][-1]174            probs = torch.nn.Softmax(dim=-1)(probs).detach().numpy()175            sampled_id = temperature_sampling(probs=top_p_sampling(probs,176                                                                top_p=top_p,177                                                                return_probs=True),178                                            temperature=temperature)179            decoder_input_ids = torch.cat((decoder_input_ids, torch.tensor([[sampled_id]])), 1)180            if sampled_id!=eos_token_id:181                continue182            else:183                tune = "X:1\n"184                tune += tokenizer.decode(decoder_input_ids[0], skip_special_tokens=True)185                return tune186                break187 188 189if __name__ == "__main__":190 191    # Ejemplo de uso192    assistant = AIAssistant()193    ner_results = assistant.entity_pos_tagger("Nader Jokhadar had given Syria the lead with a well-struck header in the seventh minute.")194    print(ner_results)195 196    image = assistant.text_to_image_generation("A majestic lion jumping from a big stone at night")197    print(image)198 199    pos_tags = assistant.grammatical_pos_tagger('Mis amigos están pensando en viajar a Londres este verano')200    print(pos_tags)201 202    tune = assistant.text_to_music("This is a traditional Irish dance music.")203    print(tune)204 205