Spaces:
Runtime error
Runtime error
Update api.py
Browse files
api.py
CHANGED
|
@@ -7,40 +7,27 @@ app = Flask(__name__)
|
|
| 7 |
@app.route('/v1/chat/completions', methods=['POST'])
|
| 8 |
def chat_completions():
|
| 9 |
data = request.json
|
| 10 |
-
messages = data.get('messages',
|
| 11 |
-
max_tokens = data.get('max_tokens', 560)
|
| 12 |
-
temperature = data.get('temperature', 0.8)
|
| 13 |
-
|
| 14 |
if not messages or not isinstance(messages, list):
|
| 15 |
return jsonify({"error": "A valid 'messages' list is required."}), 400
|
| 16 |
|
| 17 |
try:
|
| 18 |
-
|
| 19 |
-
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
message_history=messages,
|
| 23 |
-
max_tokens=max_tokens,
|
| 24 |
-
temperature=temperature
|
| 25 |
)
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
assistant_msg = result[-1] if isinstance(result, list) else result
|
| 29 |
-
elapsed = time.time() - start_time
|
| 30 |
|
| 31 |
return jsonify({
|
| 32 |
-
"model": "
|
| 33 |
"choices": [{
|
| 34 |
-
"message": {
|
| 35 |
-
"role": "assistant",
|
| 36 |
-
"content": assistant_msg
|
| 37 |
-
}
|
| 38 |
}],
|
| 39 |
-
"usage": {
|
| 40 |
-
|
| 41 |
-
}
|
| 42 |
})
|
| 43 |
-
|
| 44 |
except Exception as e:
|
| 45 |
return jsonify({"error": str(e)}), 500
|
| 46 |
|
|
|
|
| 7 |
@app.route('/v1/chat/completions', methods=['POST'])
|
| 8 |
def chat_completions():
|
| 9 |
data = request.json
|
| 10 |
+
messages = data.get('messages') # list of {"role":..., "content":...}
|
|
|
|
|
|
|
|
|
|
| 11 |
if not messages or not isinstance(messages, list):
|
| 12 |
return jsonify({"error": "A valid 'messages' list is required."}), 400
|
| 13 |
|
| 14 |
try:
|
| 15 |
+
start = time.time()
|
| 16 |
+
new_history = generate_chat_completion(
|
| 17 |
+
message=messages[-1]['content'],
|
| 18 |
+
history=messages[:-1] # feed all except last user message
|
|
|
|
|
|
|
|
|
|
| 19 |
)
|
| 20 |
+
assistant_msg = new_history[-1]['content']
|
| 21 |
+
elapsed = time.time() - start
|
|
|
|
|
|
|
| 22 |
|
| 23 |
return jsonify({
|
| 24 |
+
"model": "tiiuae/falcon-rw-1b",
|
| 25 |
"choices": [{
|
| 26 |
+
"message": {"role": "assistant", "content": assistant_msg}
|
|
|
|
|
|
|
|
|
|
| 27 |
}],
|
| 28 |
+
"usage": {"generation_time": round(elapsed, 2)},
|
| 29 |
+
"history": new_history
|
|
|
|
| 30 |
})
|
|
|
|
| 31 |
except Exception as e:
|
| 32 |
return jsonify({"error": str(e)}), 500
|
| 33 |
|