OttoCapi commited on
Commit
74f331c
·
verified ·
1 Parent(s): f5d7717

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +35 -48
app.py CHANGED
@@ -2,49 +2,29 @@ import subprocess
2
  import sys
3
  import os
4
 
5
- # Self-install dependencies with error handling
6
- REQUIRED = ["gradio==4.44.1", "deep-translator==1.11.4", "llama-cpp-python==0.2.90"]
7
  for pkg in REQUIRED:
8
  try:
9
- __import__(pkg.split("==")[0])
10
  except ImportError:
11
  print(f" Installing {pkg}...")
12
- subprocess.check_call([sys.executable, "-m", "pip", "install", "--no-cache-dir", pkg])
13
 
14
  import gradio as gr
15
  from deep_translator import GoogleTranslator
 
16
 
17
- # Check if llama_cpp imported successfully
18
- try:
19
- from llama_cpp import Llama
20
- except ImportError as e:
21
- print(f" FATAL: llama_cpp failed to import: {e}")
22
- print(" This usually means it was compiled without AVX2 support.")
23
- print(" Trying to reinstall with CMAKE_ARGS...")
24
- os.environ["CMAKE_ARGS"] = "-DLLAMA_BLAS=OFF -DLLAMA_AVX2=ON"
25
- subprocess.check_call([sys.executable, "-m", "pip", "install", "--no-cache-dir", "--force-reinstall", "llama-cpp-python==0.2.90"])
26
- from llama_cpp import Llama
27
-
28
- # Verify model file exists
29
- MODEL_PATH = "Potodoo-V1.Q4_K_M.gguf"
30
- if not os.path.exists(MODEL_PATH):
31
- print(f" FATAL: Model file '{MODEL_PATH}' not found!")
32
- print(f" Files in /app: {os.listdir('/app')}")
33
- raise FileNotFoundError(f"Model file missing: {MODEL_PATH}")
34
-
35
  print("🥔 Loading Potodoo...")
36
- try:
37
- llm = Llama(
38
- model_path=MODEL_PATH,
39
- n_ctx=512,
40
- n_gpu_layers=0,
41
- n_threads=2,
42
- verbose=True # Enable verbose to see actual errors
43
- )
44
- print("✅ Potodoo ready!")
45
- except Exception as e:
46
- print(f" FATAL: Failed to load model: {e}")
47
- raise
48
 
49
  LANG_MAP = {
50
  "Português (BR)": "pt",
@@ -56,6 +36,7 @@ LANG_MAP = {
56
  }
57
 
58
  def respond(message, history, language):
 
59
  english_input = message
60
  target_lang = LANG_MAP.get(language)
61
 
@@ -66,24 +47,26 @@ def respond(message, history, language):
66
  except:
67
  pass
68
 
69
- chat_history = []
 
 
 
70
  if history:
71
- for turn in history:
72
- chat_history.append({"role": "user", "content": turn[0]})
73
- chat_history.append({"role": "assistant", "content": turn[1]})
74
-
75
- messages = [{"role": "system", "content": "You are Potodoo."}]
76
- messages.extend(chat_history[-4:])
77
  messages.append({"role": "user", "content": english_input})
78
 
 
79
  prompt = ""
80
  for msg in messages:
81
  prompt += f"<|im_start|>{msg['role']}\n{msg['content']}<|im_end|>\n"
82
  prompt += "<|im_start|>assistant\n"
83
 
 
84
  output = llm(prompt, max_tokens=128, stop=["<|im_end|>"], temperature=0.7)
85
  eng_resp = output['choices'][0]['text'].strip()
86
 
 
87
  final_resp = eng_resp
88
  if target_lang:
89
  try:
@@ -94,18 +77,22 @@ def respond(message, history, language):
94
 
95
  return final_resp
96
 
97
- with gr.Blocks(theme=gr.themes.Soft(primary_hue="purple")) as demo:
98
- gr.Markdown("# 🥔 Potodoo")
99
- chatbot = gr.Chatbot(height=400, type="tuples")
 
 
100
 
101
  with gr.Row():
102
- msg = gr.Textbox(scale=4, show_label=False, placeholder="Type...")
103
  lang = gr.Dropdown(
104
  choices=["English", "Português (BR)", "Español", "Français", "Italiano", "日本語", "한국어"],
105
- value="English", scale=1
 
106
  )
107
 
108
  msg.submit(respond, [msg, chatbot, lang], [msg, chatbot])
109
- gr.ClearButton([msg, chatbot])
110
 
111
- demo.launch(server_name="0.0.0.0", server_port=7860)
 
 
2
  import sys
3
  import os
4
 
5
+ # Self-install dependencies (Gradio is already installed by HF)
6
+ REQUIRED = ["deep-translator==1.11.4", "llama-cpp-python==0.2.90"]
7
  for pkg in REQUIRED:
8
  try:
9
+ __import__(pkg.split("==")[0].replace("-", "_"))
10
  except ImportError:
11
  print(f" Installing {pkg}...")
12
+ subprocess.check_call([sys.executable, "-m", "pip", "install", pkg])
13
 
14
  import gradio as gr
15
  from deep_translator import GoogleTranslator
16
+ from llama_cpp import Llama
17
 
18
+ # Load Potodoo (UPDATED FILENAME!)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  print("🥔 Loading Potodoo...")
20
+ llm = Llama(
21
+ model_path="Potodoo-V1.Q4_K_M.gguf",
22
+ n_ctx=512,
23
+ n_gpu_layers=0,
24
+ n_threads=2,
25
+ verbose=False
26
+ )
27
+ print("✅ Potodoo ready!")
 
 
 
 
28
 
29
  LANG_MAP = {
30
  "Português (BR)": "pt",
 
36
  }
37
 
38
  def respond(message, history, language):
39
+ # Gradio 6.0 passes history as a list of dicts: [{"role": "user", "content": "..."}]
40
  english_input = message
41
  target_lang = LANG_MAP.get(language)
42
 
 
47
  except:
48
  pass
49
 
50
+ # Build prompt
51
+ messages = [{"role": "system", "content": "You are Potodoo, a helpful AI assistant."}]
52
+
53
+ # Add history (last 4 turns to save memory)
54
  if history:
55
+ messages.extend(history[-4:])
56
+
 
 
 
 
57
  messages.append({"role": "user", "content": english_input})
58
 
59
+ # Format as ChatML for SmolLM2
60
  prompt = ""
61
  for msg in messages:
62
  prompt += f"<|im_start|>{msg['role']}\n{msg['content']}<|im_end|>\n"
63
  prompt += "<|im_start|>assistant\n"
64
 
65
+ # Generate
66
  output = llm(prompt, max_tokens=128, stop=["<|im_end|>"], temperature=0.7)
67
  eng_resp = output['choices'][0]['text'].strip()
68
 
69
+ # Translate back
70
  final_resp = eng_resp
71
  if target_lang:
72
  try:
 
77
 
78
  return final_resp
79
 
80
+ # Gradio 6.0 UI (No 'type' or 'theme' arguments here!)
81
+ with gr.Blocks() as demo:
82
+ gr.Markdown("# 🥔 Potodoo\n*Multilingual AI Assistant*")
83
+
84
+ chatbot = gr.Chatbot(height=400) # ✅ Default 'messages' format in Gradio 6
85
 
86
  with gr.Row():
87
+ msg = gr.Textbox(scale=4, show_label=False, placeholder="Type a message...")
88
  lang = gr.Dropdown(
89
  choices=["English", "Português (BR)", "Español", "Français", "Italiano", "日本語", "한국어"],
90
+ value="English",
91
+ scale=1
92
  )
93
 
94
  msg.submit(respond, [msg, chatbot, lang], [msg, chatbot])
95
+ gr.ClearButton([msg, chatbot], value="Clear Chat")
96
 
97
+ # ✅ Theme is passed to launch() in Gradio 6.0!
98
+ demo.launch(server_name="0.0.0.0", server_port=7860, theme=gr.themes.Soft(primary_hue="purple"))