fixing stuffs

app.py

@@ -1,7 +1,45 @@
 import gradio as gr
+from ingestion.pdf import process_pdf
+from rag.pipeline import run_rag
 
-def echo(text):
-    return text
+vectorstore = None
 
-demo = gr.Interface(fn=echo, inputs="text", outputs="text")
-demo.launch()
+def load_document(file):
+    global vectorstore
+    if file is None:
+        return "Please upload a PDF file."
+    try:
+        vectorstore = process_pdf(file.name)
+        return "Document processed successfully."
+    except Exception as e:
+        return f"Error: {str(e)}"
+
+def ask(question):
+    if vectorstore is None:
+        return "Upload a document first", "", ""
+    if not question.strip():
+        return "Please enter a question", "", ""
+    try:
+        return run_rag(question, vectorstore)
+    except Exception as e:
+        return f"Error: {str(e)}", "", ""
+
+with gr.Blocks(title="Tech Explainer RAG") as demo:
+    gr.Markdown("# Tech Explainer — RAG with Automatic Evaluation")
+
+    file = gr.File(label="Upload PDF", file_types=[".pdf"])
+    load_btn = gr.Button("Process PDF")
+    status = gr.Textbox(label="Status")
+
+    question = gr.Textbox(label="Question", placeholder="Ask a question about the document...")
+    ask_btn = gr.Button("Ask")
+
+    answer = gr.Textbox(label="Answer", lines=5)
+    sources = gr.Textbox(label="Sources", lines=2)
+    evaluation = gr.Textbox(label="Evaluation", lines=3)
+
+    load_btn.click(load_document, inputs=file, outputs=status)
+    ask_btn.click(ask, inputs=question, outputs=[answer, sources, evaluation])
+
+if __name__ == "__main__":
+    demo.launch()

evaluation/metrics.py

@@ -1,33 +1,43 @@
 import json
+import os
 from datetime import datetime
 from rag.llm import generate
 
-
 LOG_PATH = "logs/metrics.jsonl"
 
 
+def ensure_log_dir():
+    """Cria o diretório de logs se não existir"""
+    os.makedirs(os.path.dirname(LOG_PATH), exist_ok=True)
+
+
 def evaluate_and_log(question, context, answer):
+    ensure_log_dir()
+    
     prompt = (
-    "Evaluate the answer based on faithfulness, clarity, usefulness and completeness. "
-    "Provide a short justification.\n"
-    f"Context:\n{context}\n"
-    f"Answer:\n{answer}"
+        "Evaluate the answer based on faithfulness, clarity, usefulness and completeness. "
+        "Provide a short justification.\n\n"
+        f"Context:\n{context}\n\n"
+        f"Answer:\n{answer}\n\n"
+        "Evaluation:"
     )
 
-
-    evaluation = generate(prompt, max_tokens=200)
-
+    try:
+        evaluation = generate(prompt, max_tokens=200)
+    except Exception as e:
+        evaluation = f"Evaluation failed: {str(e)}"
 
     record = {
-    "timestamp": datetime.utcnow().isoformat(),
-    "question": question,
-    "answer": answer,
-    "evaluation": evaluation
+        "timestamp": datetime.utcnow().isoformat(),
+        "question": question,
+        "answer": answer,
+        "evaluation": evaluation
     }
 
+    try:
+        with open(LOG_PATH, "a", encoding="utf-8") as f:
+            f.write(json.dumps(record, ensure_ascii=False) + "\n")
+    except Exception as e:
+        print(f"Warning: Could not write to log file: {e}")
 
-    with open(LOG_PATH, "a") as f:
-        f.write(json.dumps(record) + "\n")
-
-
-    return evaluation
+    return evaluation

ingestion/pdf.py

@@ -1,16 +1,22 @@
 from pypdf import PdfReader
-from langchain_text_splitters import RecursiveCharacterTextSplitter
-from langchain_core.documents import Document
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+from langchain.schema import Document
 from rag.retriever import build_vectorstore
 
 
-def process_pdf(file):
-    reader = PdfReader(file.name)
+def process_pdf(file_path):
+    """
+    file_path: string - caminho para o arquivo PDF
+    """
+    reader = PdfReader(file_path)
     text = "".join(page.extract_text() or "" for page in reader.pages)
 
-    splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
+    splitter = RecursiveCharacterTextSplitter(
+        chunk_size=500,
+        chunk_overlap=100
+    )
 
     chunks = splitter.split_text(text)
     documents = [Document(page_content=c) for c in chunks]
 
-    return build_vectorstore(documents)
+    return build_vectorstore(documents)

rag/llm.py

@@ -6,23 +6,52 @@ from config import PRIMARY_LLM, FALLBACK_LLM
 def load_model():
     try:
         tokenizer = AutoTokenizer.from_pretrained(PRIMARY_LLM)
+        if tokenizer.pad_token is None:
+            tokenizer.pad_token = tokenizer.eos_token
+        
         model = AutoModelForCausalLM.from_pretrained(
-            PRIMARY_LLM, device_map="auto"
+            PRIMARY_LLM,
+            device_map="auto",
+            load_in_8bit=True
         )
-    except Exception:
+        print(f"Loaded primary model: {PRIMARY_LLM}")
+    except Exception as e:
+        print(f"Primary model failed: {e}")
+        print(f"Loading fallback: {FALLBACK_LLM}")
+        
         tokenizer = AutoTokenizer.from_pretrained(FALLBACK_LLM, trust_remote_code=True)
-        model = AutoModelForCausalLM.from_pretrained(FALLBACK_LLM)
+        if tokenizer.pad_token is None:
+            tokenizer.pad_token = tokenizer.eos_token
+            
+        model = AutoModelForCausalLM.from_pretrained(
+            FALLBACK_LLM,
+            trust_remote_code=True,
+            torch_dtype=torch.float16,
+            device_map="auto"
+        )
 
     return tokenizer, model
 
+
 tokenizer, model = load_model()
 
 
 def generate(prompt, max_tokens=400):
-    inputs = tokenizer(prompt, return_tensors="pt", truncation=True)
+    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=2048)
     inputs = {k: v.to(model.device) for k, v in inputs.items()}
 
     with torch.no_grad():
-        output = model.generate(**inputs, max_new_tokens=max_tokens)
+        output = model.generate(
+            **inputs,
+            max_new_tokens=max_tokens,
+            do_sample=True,
+            temperature=0.7,
+            top_p=0.9
+        )
 
-    return tokenizer.decode(output[0], skip_special_tokens=True)
+    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
+    
+    if prompt in generated_text:
+        generated_text = generated_text.replace(prompt, "").strip()
+    
+    return generated_text

rag/pipeline.py

@@ -1,19 +1,21 @@
 from rag.llm import generate
 from evaluation.metrics import evaluate_and_log
 
+
 def run_rag(question, vectorstore):
     docs = vectorstore.similarity_search(question, k=3)
-    context = "\n".join(d.page_content for d in docs)
-
+    context = "\n\n".join(d.page_content for d in docs)
 
     prompt = (
-    "Use the context below to answer the question clearly and simply.\n"
-    f"Context:\n{context}\n"
-    f"Question: {question}"
+        "Use the context below to answer the question clearly and simply.\n\n"
+        f"Context:\n{context}\n\n"
+        f"Question: {question}\n\n"
+        "Answer:"
     )
 
-
     answer = generate(prompt)
     evaluation = evaluate_and_log(question, context, answer)
 
-    return answer, "Sources retrieved from document", evaluation
+    sources = f"Retrieved {len(docs)} relevant passages from document"
+
+    return answer, sources, evaluation

requirements.txt

@@ -1,14 +1,10 @@
-gradio
-gradio_client>=0.8.0
+gradio==4.44.0
 transformers
 sentence-transformers
-torch
-pypdf
-
 langchain
-langchain-core
 langchain-community
-langchain-text-splitters
-langchain-huggingface
-
 faiss-cpu
+pypdf
+torch
+accelerate
+bitsandbytes