Spaces:
Sleeping
Sleeping
File size: 2,304 Bytes
785c4f7 7d3dbed fbd71f6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 |
import torch
from evo_model import EvoTransformer
from transformers import AutoTokenizer, pipeline
from rag_utils import RAGRetriever, extract_text_from_file
import os
# Load Evo model
def load_evo_model(model_path="evo_hellaswag.pt", device=None):
if device is None:
device = "cuda" if torch.cuda.is_available() else "cpu"
model = EvoTransformer()
model.load_state_dict(torch.load(model_path, map_location=device))
model.to(device)
model.eval()
return model, device
evo_model, device = load_evo_model()
tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
# Load GPT-3.5 (via OpenAI API)
import openai
openai.api_key = os.getenv("OPENAI_API_KEY")
# RAG Retriever
retriever = RAGRetriever()
def get_context_from_file(file_obj):
file_path = file_obj.name
text = extract_text_from_file(file_path)
retriever.add_document(text)
return text
# Evo prediction
def get_evo_response(prompt, file=None):
# Step 1: augment context if document is uploaded
context = ""
if file is not None:
context_list = retriever.retrieve(prompt)
context = "\n".join(context_list)
full_prompt = f"{prompt}\n{context}"
# Step 2: use Evo to predict
options = ["Yes, proceed with the action.", "No, maintain current strategy."]
inputs = [f"{full_prompt} {opt}" for opt in options]
encoded = tokenizer(inputs, padding=True, truncation=True, return_tensors="pt").to(device)
with torch.no_grad():
logits = evo_model(encoded["input_ids"]).squeeze(-1)
probs = torch.softmax(logits, dim=0)
best = torch.argmax(probs).item()
return f"Evo suggests: {options[best]} (Confidence: {probs[best]:.2f})"
# GPT-3.5 response
def get_gpt_response(prompt, file=None):
context = ""
if file is not None:
context_list = retriever.retrieve(prompt)
context = "\n".join(context_list)
full_prompt = (
f"Question: {prompt}\n"
f"Relevant Context:\n{context}\n"
f"Answer like a financial advisor."
)
response = openai.ChatCompletion.create(
model="gpt-3.5-turbo",
messages=[
{"role": "user", "content": full_prompt}
],
temperature=0.4,
)
return response.choices[0].message.content.strip()
#
|