TESTE_5

app.py

@@ -1,36 +1,35 @@
 import gradio as gr
 from huggingface_hub import InferenceClient
-from transformers import AutoTokenizer, AutoModelForQuestionAnswering, pipeline
 from sentence_transformers import SentenceTransformer, util
+from transformers import pipeline
 
 # Modelos
-model_name = "deepset/roberta-base-squad2"
-qa_pipeline = pipeline("question-answering", model=model_name, tokenizer=model_name)
+chat_model_zephyr = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
+chat_model_gemma = pipeline("text-generation", model="declare-lab/gemma-v2", max_new_tokens=256)
 
-client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
-
-# Modelo para comparação semântica (cosine similarity)
+# Similaridade
 similarity_model = SentenceTransformer("all-MiniLM-L6-v2")
 
 
-def get_qa_pipeline_answer(question, context):
-    return qa_pipeline({"question": question, "context": context})["answer"]
-
-
-def get_zephyr_answer(question, context):
+def get_zephyr_response(question):
     messages = [
         {"role": "system", "content": "You are a helpful assistant."},
-        {"role": "user", "content": f"Context: {context}\n\nQuestion: {question}"},
+        {"role": "user", "content": question}
     ]
-    response = client.chat_completion(
+    response = chat_model_zephyr.chat_completion(
         messages,
-        max_tokens=512,
+        max_tokens=256,
         temperature=0.7,
         top_p=0.95,
     )
     return response.choices[0].message.content.strip()
 
 
+def get_gemma_response(question):
+    generated = chat_model_gemma(question)[0]["generated_text"]
+    return generated.strip()
+
+
 def compare_answers(answer1, answer2):
     emb1 = similarity_model.encode(answer1, convert_to_tensor=True)
     emb2 = similarity_model.encode(answer2, convert_to_tensor=True)
@@ -38,28 +37,25 @@ def compare_answers(answer1, answer2):
     return round(similarity, 3)
 
 
-def respond(question, context):
-    answer1 = get_qa_pipeline_answer(question, context)
-    answer2 = get_zephyr_answer(question, context)
-    similarity_score = compare_answers(answer1, answer2)
+def respond(question):
+    answer_zephyr = get_zephyr_response(question)
+    answer_gemma = get_gemma_response(question)
+    similarity = compare_answers(answer_zephyr, answer_gemma)
 
     return (
-        f"📘 Roberta-base-squad2:\n{answer1}\n\n"
-        f"🧠 Zephyr-7b:\n{answer2}\n\n"
-        f"🔍 Similaridade Semântica: **{similarity_score}**"
+        f"🧠 Zephyr-7b:\n{answer_zephyr}\n\n"
+        f"🤖 Gemma-v2:\n{answer_gemma}\n\n"
+        f"🔍 Similaridade Semântica: **{similarity}**"
     )
 
 
-# Interface Gradio
 with gr.Blocks() as demo:
-    gr.Markdown("# 🔎 Perguntas com dois modelos\nCompare duas respostas e veja a similaridade.")
-    with gr.Row():
-        question = gr.Textbox(label="Pergunta")
-        context = gr.Textbox(label="Contexto")
-    submit_btn = gr.Button("Obter Respostas")
-    output = gr.Textbox(label="Respostas e Similaridade")
+    gr.Markdown("# 🤖 Comparador de Respostas (sem contexto)\nDigite uma pergunta e veja as respostas de dois modelos.")
+    question = gr.Textbox(label="Pergunta")
+    submit = gr.Button("Comparar Respostas")
+    output = gr.Textbox(label="Respostas e Similaridade", lines=15)
 
-    submit_btn.click(respond, inputs=[question, context], outputs=output)
+    submit.click(respond, inputs=question, outputs=output)
 
 if __name__ == "__main__":
     demo.launch()