Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -1,183 +1,218 @@
|
|
1 |
-
import streamlit as st
|
2 |
-
import torch
|
3 |
-
from peft import AutoPeftModelForCausalLM
|
4 |
-
from transformers import AutoTokenizer, TextStreamer
|
5 |
-
|
6 |
-
import
|
7 |
-
import
|
8 |
-
import
|
9 |
-
import
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
st.
|
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 |
-
st.header("
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
72 |
-
|
73 |
-
|
74 |
-
|
75 |
-
|
76 |
-
|
77 |
-
#
|
78 |
-
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
|
96 |
-
|
97 |
-
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
106 |
-
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
-
|
112 |
-
|
113 |
-
|
114 |
-
|
115 |
-
|
116 |
-
|
117 |
-
|
118 |
-
|
119 |
-
|
120 |
-
|
121 |
-
|
122 |
-
|
123 |
-
|
124 |
-
|
125 |
-
|
126 |
-
|
127 |
-
|
128 |
-
|
129 |
-
|
130 |
-
|
131 |
-
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
thread
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
#
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
181 |
-
|
182 |
-
|
183 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import torch
|
3 |
+
from peft import AutoPeftModelForCausalLM
|
4 |
+
from transformers import AutoTokenizer, TextStreamer
|
5 |
+
# bitsandbytes is no longer needed
|
6 |
+
import io
|
7 |
+
import sys
|
8 |
+
import threading
|
9 |
+
import time
|
10 |
+
import queue # Import the queue module
|
11 |
+
|
12 |
+
# --- Configuration ---
|
13 |
+
DEFAULT_MODEL_PATH = "lora_model" # Or your default path
|
14 |
+
# DEFAULT_LOAD_IN_4BIT is removed as we are not using quantization
|
15 |
+
|
16 |
+
# --- Page Configuration ---
|
17 |
+
st.set_page_config(page_title="Fine-tuned LLM Chat Interface (CPU)", layout="wide")
|
18 |
+
st.title("Fine-tuned LLM Chat Interface (CPU Mode)")
|
19 |
+
st.warning("Running in CPU mode. Expect slower generation times and higher RAM usage.", icon="⚠️")
|
20 |
+
|
21 |
+
# --- Model Loading (Cached for CPU) ---
|
22 |
+
@st.cache_resource(show_spinner="Loading model and tokenizer onto CPU...")
|
23 |
+
def load_model_and_tokenizer_cpu(model_path):
|
24 |
+
"""Loads the PEFT model and tokenizer onto the CPU."""
|
25 |
+
try:
|
26 |
+
# Use standard float32 for CPU compatibility and stability
|
27 |
+
torch_dtype = torch.float32
|
28 |
+
|
29 |
+
model = AutoPeftModelForCausalLM.from_pretrained(
|
30 |
+
model_path,
|
31 |
+
torch_dtype=torch_dtype,
|
32 |
+
# load_in_4bit=False, # Explicitly removed/not needed
|
33 |
+
device_map="cpu", # Force loading onto CPU
|
34 |
+
)
|
35 |
+
tokenizer = AutoTokenizer.from_pretrained(model_path)
|
36 |
+
model.eval() # Set model to evaluation mode
|
37 |
+
print("Model and tokenizer loaded successfully onto CPU.")
|
38 |
+
return model, tokenizer
|
39 |
+
except Exception as e:
|
40 |
+
st.error(f"Error loading model from path '{model_path}' onto CPU: {e}", icon="🚨")
|
41 |
+
print(f"Error loading model onto CPU: {e}")
|
42 |
+
return None, None
|
43 |
+
|
44 |
+
# --- Custom Streamer Class (Modified for Queue) ---
|
45 |
+
class QueueStreamer(TextStreamer):
|
46 |
+
def __init__(self, tokenizer, skip_prompt, q):
|
47 |
+
super().__init__(tokenizer, skip_prompt=skip_prompt)
|
48 |
+
self.queue = q
|
49 |
+
self.stop_signal = None # Can be used if needed, but queue is primary
|
50 |
+
|
51 |
+
def on_finalized_text(self, text: str, stream_end: bool = False):
|
52 |
+
"""Puts the text onto the queue."""
|
53 |
+
self.queue.put(text)
|
54 |
+
if stream_end:
|
55 |
+
self.end()
|
56 |
+
|
57 |
+
def end(self):
|
58 |
+
"""Signals the end of generation by putting None in the queue."""
|
59 |
+
self.queue.put(self.stop_signal) # Put None (or a specific sentinel)
|
60 |
+
|
61 |
+
|
62 |
+
# --- Sidebar for Settings ---
|
63 |
+
with st.sidebar:
|
64 |
+
st.header("Model Configuration")
|
65 |
+
st.info(f"Model loaded on startup: `{DEFAULT_MODEL_PATH}` (CPU Mode).")
|
66 |
+
|
67 |
+
st.header("Generation Settings")
|
68 |
+
temperature = st.slider("Temperature", min_value=0.0, max_value=2.0, value=0.7, step=0.05)
|
69 |
+
# min_p might not be as commonly used or effective without top_p/top_k,
|
70 |
+
# but keeping it allows experimentation. Consider using top_k or top_p instead.
|
71 |
+
# Example: top_p = st.slider("Top P", min_value=0.01, max_value=1.0, value=0.9, step=0.01)
|
72 |
+
min_p = st.slider("Min P", min_value=0.01, max_value=1.0, value=0.1, step=0.01) # Keep for now
|
73 |
+
max_tokens = st.slider("Max New Tokens", min_value=50, max_value=2048, value=256, step=50) # Reduced default for CPU
|
74 |
+
|
75 |
+
if st.button("Clear Chat History"):
|
76 |
+
st.session_state.messages = []
|
77 |
+
st.rerun() # Rerun to clear display immediately
|
78 |
+
|
79 |
+
|
80 |
+
# --- Load Model (runs only once on first run or if cache is cleared) ---
|
81 |
+
model, tokenizer = load_model_and_tokenizer_cpu(DEFAULT_MODEL_PATH)
|
82 |
+
|
83 |
+
# --- Initialize Session State ---
|
84 |
+
if "messages" not in st.session_state:
|
85 |
+
st.session_state.messages = []
|
86 |
+
|
87 |
+
# --- Main Chat Interface ---
|
88 |
+
if model is None or tokenizer is None:
|
89 |
+
st.error("CPU Model loading failed. Please check the path, available RAM, and logs. Cannot proceed.")
|
90 |
+
st.stop()
|
91 |
+
|
92 |
+
# Display conversation history
|
93 |
+
for message in st.session_state.messages:
|
94 |
+
with st.chat_message(message["role"]):
|
95 |
+
st.markdown(message["content"])
|
96 |
+
|
97 |
+
# Handle user input
|
98 |
+
user_input = st.chat_input("Ask the fine-tuned model (CPU)...")
|
99 |
+
|
100 |
+
if user_input:
|
101 |
+
# Add user message to history and display it
|
102 |
+
st.session_state.messages.append({"role": "user", "content": user_input})
|
103 |
+
with st.chat_message("user"):
|
104 |
+
st.markdown(user_input)
|
105 |
+
|
106 |
+
# Prepare for model response
|
107 |
+
with st.chat_message("assistant"):
|
108 |
+
response_placeholder = st.empty()
|
109 |
+
response_placeholder.markdown("Generating response on CPU... please wait... ▌") # Initial message
|
110 |
+
text_queue = queue.Queue() # Create a queue for this specific response
|
111 |
+
# Initialize the modified streamer
|
112 |
+
text_streamer = QueueStreamer(tokenizer, skip_prompt=True, q=text_queue)
|
113 |
+
|
114 |
+
# Prepare input for the model
|
115 |
+
messages_for_model = st.session_state.messages
|
116 |
+
|
117 |
+
try:
|
118 |
+
# Ensure inputs are on the CPU (model.device should be 'cpu' now)
|
119 |
+
target_device = model.device
|
120 |
+
# print(f"Model device: {target_device}") # Debugging: should print 'cpu'
|
121 |
+
|
122 |
+
if tokenizer.chat_template:
|
123 |
+
inputs = tokenizer.apply_chat_template(
|
124 |
+
messages_for_model,
|
125 |
+
tokenize=True,
|
126 |
+
add_generation_prompt=True,
|
127 |
+
return_tensors="pt"
|
128 |
+
).to(target_device) # Send input tensors to CPU
|
129 |
+
else:
|
130 |
+
prompt_text = "\n".join([f"{msg['role']}: {msg['content']}" for msg in messages_for_model]) + "\nassistant:"
|
131 |
+
inputs = tokenizer(prompt_text, return_tensors="pt").input_ids.to(target_device) # Send input tensors to CPU
|
132 |
+
|
133 |
+
# Generation arguments
|
134 |
+
generation_kwargs = dict(
|
135 |
+
input_ids=inputs,
|
136 |
+
streamer=text_streamer, # Use the QueueStreamer
|
137 |
+
max_new_tokens=max_tokens,
|
138 |
+
use_cache=True, # Caching can still help CPU generation speed
|
139 |
+
temperature=temperature if temperature > 0 else None,
|
140 |
+
top_p=None, # Consider adding top_p slider in UI
|
141 |
+
# top_k=50, # Example: Or use top_k
|
142 |
+
min_p=min_p,
|
143 |
+
do_sample=True if temperature > 0 else False,
|
144 |
+
eos_token_id=tokenizer.eos_token_id,
|
145 |
+
pad_token_id=tokenizer.pad_token_id if tokenizer.pad_token_id else tokenizer.eos_token_id
|
146 |
+
)
|
147 |
+
|
148 |
+
# Define the target function for the thread
|
149 |
+
def generation_thread_func():
|
150 |
+
try:
|
151 |
+
# Run generation in the background thread (on CPU)
|
152 |
+
# Wrap in torch.no_grad() to save memory during inference
|
153 |
+
with torch.no_grad():
|
154 |
+
model.generate(**generation_kwargs)
|
155 |
+
except Exception as e:
|
156 |
+
# If error occurs in thread, signal stop and maybe log
|
157 |
+
print(f"Error in generation thread: {e}")
|
158 |
+
# Attempt to put error message in queue? Or just rely on main thread error handling
|
159 |
+
st.error(f"Error during generation: {e}") # Show error in UI too
|
160 |
+
finally:
|
161 |
+
# Ensure the queue loop terminates even if error occurred
|
162 |
+
text_streamer.end()
|
163 |
+
|
164 |
+
|
165 |
+
# Start the generation thread
|
166 |
+
thread = threading.Thread(target=generation_thread_func)
|
167 |
+
thread.start()
|
168 |
+
|
169 |
+
# --- Main thread: Read from queue and update UI ---
|
170 |
+
generated_text = ""
|
171 |
+
while True:
|
172 |
+
try:
|
173 |
+
# Get the next text chunk from the queue
|
174 |
+
# Use timeout to prevent blocking indefinitely if thread hangs
|
175 |
+
chunk = text_queue.get(block=True, timeout=1) # Short timeout OK for slow CPU gen
|
176 |
+
if chunk is text_streamer.stop_signal: # Check for end signal (None)
|
177 |
+
break
|
178 |
+
generated_text += chunk
|
179 |
+
response_placeholder.markdown(generated_text + "▌") # Update placeholder
|
180 |
+
except queue.Empty:
|
181 |
+
# If queue is empty, check if the generation thread is still running
|
182 |
+
if not thread.is_alive():
|
183 |
+
# Thread finished, but maybe didn't put the stop signal (error?)
|
184 |
+
break # Exit loop
|
185 |
+
# Otherwise, continue waiting for next chunk
|
186 |
+
continue
|
187 |
+
except Exception as e:
|
188 |
+
st.error(f"Error reading from generation queue: {e}")
|
189 |
+
print(f"Error reading from queue: {e}")
|
190 |
+
break # Exit loop on queue error
|
191 |
+
|
192 |
+
# Final update without the cursor
|
193 |
+
response_placeholder.markdown(generated_text)
|
194 |
+
|
195 |
+
# Add the complete assistant response to history *after* generation
|
196 |
+
if generated_text: # Only add if something was generated
|
197 |
+
st.session_state.messages.append({"role": "assistant", "content": generated_text})
|
198 |
+
else:
|
199 |
+
# Handle case where generation failed silently in thread or produced nothing
|
200 |
+
if not any(m['role'] == 'assistant' and m['content'].startswith("*Error") for m in st.session_state.messages):
|
201 |
+
st.warning("Assistant produced no output.", icon="⚠️")
|
202 |
+
|
203 |
+
|
204 |
+
# Wait briefly for the thread to finish if it hasn't already
|
205 |
+
thread.join(timeout=5.0) # Longer timeout might be needed if cleanup is slow
|
206 |
+
|
207 |
+
|
208 |
+
except Exception as e:
|
209 |
+
st.error(f"Error during generation setup or queue handling: {e}", icon="🔥")
|
210 |
+
print(f"Error setting up generation or handling queue: {e}")
|
211 |
+
# Add error to chat history for context
|
212 |
+
error_message = f"*Error generating response: {e}*"
|
213 |
+
if not generated_text: # Add if no text was generated at all
|
214 |
+
st.session_state.messages.append({"role": "assistant", "content": error_message})
|
215 |
+
response_placeholder.error(f"Error generating response: {e}")
|
216 |
+
else: # Append error notice if some text was generated before error
|
217 |
+
st.session_state.messages.append({"role": "assistant", "content": generated_text + "\n\n" + error_message})
|
218 |
+
response_placeholder.markdown(generated_text + f"\n\n*{error_message}*")
|