Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,16 +1,12 @@
|
|
| 1 |
from flask import Flask, request, jsonify
|
| 2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
|
|
|
| 3 |
import torch
|
| 4 |
-
import requests
|
| 5 |
|
| 6 |
app = Flask(__name__)
|
| 7 |
-
|
| 8 |
-
# Allow CORS for everything (so TurboWarp can connect)
|
| 9 |
-
from flask_cors import CORS
|
| 10 |
CORS(app)
|
| 11 |
|
| 12 |
-
|
| 13 |
-
print("π Loading Phi model... this may take a minute.")
|
| 14 |
model_name = "microsoft/phi-2"
|
| 15 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 16 |
model = AutoModelForCausalLM.from_pretrained(
|
|
@@ -19,20 +15,18 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
| 19 |
low_cpu_mem_usage=True
|
| 20 |
)
|
| 21 |
model.to("cuda" if torch.cuda.is_available() else "cpu")
|
| 22 |
-
print("β
|
| 23 |
|
| 24 |
-
# === Main API ===
|
| 25 |
@app.route("/api/ask", methods=["POST"])
|
| 26 |
def ask():
|
| 27 |
-
data = request.get_json()
|
| 28 |
prompt = data.get("prompt", "")
|
| 29 |
|
| 30 |
chat_prompt = f"### Instruction:\nYou are Acla, a helpful AI powered by phi-3 mini that can reason about math, code, and logic.\n\n### Input:\n{prompt}\n\n### Response:"
|
| 31 |
inputs = tokenizer(chat_prompt, return_tensors="pt").to(model.device)
|
| 32 |
-
|
| 33 |
outputs = model.generate(
|
| 34 |
**inputs,
|
| 35 |
-
max_new_tokens=
|
| 36 |
temperature=0.7,
|
| 37 |
top_p=0.9,
|
| 38 |
do_sample=True
|
|
@@ -42,22 +36,9 @@ def ask():
|
|
| 42 |
response = text.split("### Response:")[-1].strip()
|
| 43 |
return jsonify({"reply": response})
|
| 44 |
|
| 45 |
-
|
| 46 |
-
# === Proxy endpoint ===
|
| 47 |
-
@app.route("/proxy", methods=["POST"])
|
| 48 |
-
def proxy():
|
| 49 |
-
"""Forward TurboWarp requests to /api/ask internally."""
|
| 50 |
-
try:
|
| 51 |
-
data = request.get_json()
|
| 52 |
-
r = requests.post("http://localhost:7860/api/ask", json=data)
|
| 53 |
-
return jsonify(r.json())
|
| 54 |
-
except Exception as e:
|
| 55 |
-
return jsonify({"error": str(e)}), 500
|
| 56 |
-
|
| 57 |
-
|
| 58 |
@app.route("/")
|
| 59 |
def home():
|
| 60 |
-
return "π§ Phi-2
|
| 61 |
|
| 62 |
if __name__ == "__main__":
|
| 63 |
app.run(host="0.0.0.0", port=7860)
|
|
|
|
| 1 |
from flask import Flask, request, jsonify
|
| 2 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 3 |
+
from flask_cors import CORS
|
| 4 |
import torch
|
|
|
|
| 5 |
|
| 6 |
app = Flask(__name__)
|
|
|
|
|
|
|
|
|
|
| 7 |
CORS(app)
|
| 8 |
|
| 9 |
+
print("π Loading Phi model (microsoft/phi-2)...")
|
|
|
|
| 10 |
model_name = "microsoft/phi-2"
|
| 11 |
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
| 12 |
model = AutoModelForCausalLM.from_pretrained(
|
|
|
|
| 15 |
low_cpu_mem_usage=True
|
| 16 |
)
|
| 17 |
model.to("cuda" if torch.cuda.is_available() else "cpu")
|
| 18 |
+
print("β
Phi model loaded!")
|
| 19 |
|
|
|
|
| 20 |
@app.route("/api/ask", methods=["POST"])
|
| 21 |
def ask():
|
| 22 |
+
data = request.get_json(force=True)
|
| 23 |
prompt = data.get("prompt", "")
|
| 24 |
|
| 25 |
chat_prompt = f"### Instruction:\nYou are Acla, a helpful AI powered by phi-3 mini that can reason about math, code, and logic.\n\n### Input:\n{prompt}\n\n### Response:"
|
| 26 |
inputs = tokenizer(chat_prompt, return_tensors="pt").to(model.device)
|
|
|
|
| 27 |
outputs = model.generate(
|
| 28 |
**inputs,
|
| 29 |
+
max_new_tokens=250,
|
| 30 |
temperature=0.7,
|
| 31 |
top_p=0.9,
|
| 32 |
do_sample=True
|
|
|
|
| 36 |
response = text.split("### Response:")[-1].strip()
|
| 37 |
return jsonify({"reply": response})
|
| 38 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 39 |
@app.route("/")
|
| 40 |
def home():
|
| 41 |
+
return "π§ Phi-2 chatbot is running! POST JSON to /api/ask with {'prompt': 'your question'}."
|
| 42 |
|
| 43 |
if __name__ == "__main__":
|
| 44 |
app.run(host="0.0.0.0", port=7860)
|