Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -5,7 +5,7 @@ from threading import Thread
|
|
5 |
import os
|
6 |
import json
|
7 |
import uuid
|
8 |
-
from datasets import Dataset
|
9 |
from huggingface_hub import HfApi, login
|
10 |
import time
|
11 |
|
@@ -28,63 +28,6 @@ DATASET_FILENAME = "feedback.jsonl" # Filename for feedback data
|
|
28 |
# Ensure feedback directory exists
|
29 |
os.makedirs(DATASET_PATH, exist_ok=True)
|
30 |
|
31 |
-
# Sync existing dataset from Hub if available
|
32 |
-
def sync_dataset_from_hub():
|
33 |
-
"""Download existing dataset from Hub and merge with local data"""
|
34 |
-
try:
|
35 |
-
# Try to get token from environment variable
|
36 |
-
hf_token = os.environ.get("HF_TOKEN")
|
37 |
-
if hf_token:
|
38 |
-
login(token=hf_token)
|
39 |
-
|
40 |
-
# Check if the dataset exists on Hub
|
41 |
-
api = HfApi()
|
42 |
-
try:
|
43 |
-
dataset_info = api.dataset_info(DATASET_REPO)
|
44 |
-
# Dataset exists, download it
|
45 |
-
print(f"Syncing existing dataset from {DATASET_REPO}")
|
46 |
-
remote_dataset = load_dataset(DATASET_REPO)
|
47 |
-
|
48 |
-
# Convert to list of dictionaries
|
49 |
-
remote_data = [item for item in remote_dataset['train']]
|
50 |
-
|
51 |
-
# Check if local file exists
|
52 |
-
local_file = os.path.join(DATASET_PATH, DATASET_FILENAME)
|
53 |
-
local_data = []
|
54 |
-
|
55 |
-
if os.path.exists(local_file):
|
56 |
-
# Read local data
|
57 |
-
with open(local_file, 'r') as f:
|
58 |
-
for line in f:
|
59 |
-
try:
|
60 |
-
local_data.append(json.loads(line))
|
61 |
-
except json.JSONDecodeError:
|
62 |
-
continue
|
63 |
-
|
64 |
-
# Merge data (using IDs to avoid duplicates)
|
65 |
-
all_items = {}
|
66 |
-
for item in remote_data + local_data:
|
67 |
-
all_items[item['id']] = item
|
68 |
-
|
69 |
-
# Write back merged data
|
70 |
-
with open(local_file, 'w') as f:
|
71 |
-
for item in all_items.values():
|
72 |
-
f.write(json.dumps(item) + '\n')
|
73 |
-
|
74 |
-
print(f"Synced {len(all_items)} feedback items")
|
75 |
-
return True
|
76 |
-
|
77 |
-
except Exception as e:
|
78 |
-
print(f"Dataset {DATASET_REPO} does not exist yet or could not be accessed: {e}")
|
79 |
-
return False
|
80 |
-
|
81 |
-
except Exception as e:
|
82 |
-
print(f"Error syncing dataset: {e}")
|
83 |
-
return False
|
84 |
-
|
85 |
-
# Call sync on startup
|
86 |
-
sync_dataset_from_hub()
|
87 |
-
|
88 |
# Feedback storage functions
|
89 |
def save_feedback_locally(conversation, satisfaction, feedback_text):
|
90 |
"""Save feedback to a local JSONL file"""
|
@@ -150,17 +93,49 @@ def push_feedback_to_hub(hf_token=None):
|
|
150 |
print(f"Error pushing feedback data to Hub: {e}")
|
151 |
return False
|
152 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
153 |
# Function to handle the research feedback submission
|
154 |
-
def submit_research_feedback(
|
155 |
"""Save user feedback both locally and to HuggingFace Hub"""
|
156 |
-
# Print debug information
|
157 |
-
print(f"Saving feedback with conversation history containing {len(conv_history)} messages")
|
158 |
-
if conv_history and len(conv_history) > 0:
|
159 |
-
print(f"First message: {conv_history[0]['role']}: {conv_history[0]['content'][:30]}...")
|
160 |
-
print(f"Last message: {conv_history[-1]['role']}: {conv_history[-1]['content'][:30]}...")
|
161 |
-
|
162 |
# Save locally first
|
163 |
-
feedback_id = save_feedback_locally(
|
164 |
|
165 |
# Get token from environment variable
|
166 |
env_token = os.environ.get("HF_TOKEN")
|
@@ -175,114 +150,29 @@ def submit_research_feedback(conv_history, satisfaction, feedback_text):
|
|
175 |
|
176 |
return status_msg
|
177 |
|
178 |
-
# Initial state - set up at app start
|
179 |
-
def initialize_state():
|
180 |
-
"""Initialize the conversation state - this could load previous sessions or start fresh"""
|
181 |
-
return [] # Start with empty conversation history
|
182 |
-
|
183 |
# Create the Gradio blocks interface
|
184 |
with gr.Blocks() as demo:
|
185 |
-
#
|
186 |
-
|
187 |
|
188 |
with gr.Row():
|
189 |
with gr.Column(scale=3):
|
190 |
-
#
|
191 |
-
def
|
192 |
-
|
193 |
-
|
194 |
-
state =
|
195 |
-
|
196 |
-
|
197 |
-
# Copy history to state if state is empty but history exists
|
198 |
-
if len(state) == 0 and len(history) > 0:
|
199 |
-
state = history.copy()
|
200 |
-
print(f"Copied {len(history)} messages from history to state")
|
201 |
-
|
202 |
-
# Add user message to state
|
203 |
-
state.append({"role": "user", "content": message})
|
204 |
-
|
205 |
-
# Process with the model (this doesn't modify the original history)
|
206 |
-
input_text = tokenizer.apply_chat_template(state, tokenize=False, add_generation_prompt=True)
|
207 |
-
inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
|
208 |
-
|
209 |
-
# Create a streamer
|
210 |
-
streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
|
211 |
-
|
212 |
-
# Set up generation parameters
|
213 |
-
generation_kwargs = {
|
214 |
-
"input_ids": inputs,
|
215 |
-
"max_new_tokens": 1024,
|
216 |
-
"temperature": float(temperature),
|
217 |
-
"top_p": float(top_p),
|
218 |
-
"do_sample": True,
|
219 |
-
"streamer": streamer,
|
220 |
-
"eos_token_id": 128009,
|
221 |
-
}
|
222 |
-
|
223 |
-
# Run generation in a separate thread
|
224 |
-
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
225 |
-
thread.start()
|
226 |
-
|
227 |
-
# Yield from the streamer as tokens are generated
|
228 |
-
response = ""
|
229 |
-
for new_text in streamer:
|
230 |
-
response += new_text
|
231 |
-
# For each partial response, yield the text only
|
232 |
-
# We'll update the state after generation is complete
|
233 |
-
yield response
|
234 |
-
|
235 |
-
# After generation completes, update our state with the final response
|
236 |
-
state.append({"role": "assistant", "content": response})
|
237 |
-
|
238 |
-
# Return the updated state
|
239 |
-
return state
|
240 |
-
|
241 |
-
# Create a wrapper that connects to ChatInterface but also updates our state
|
242 |
-
def chat_with_state(message, history, temperature, top_p):
|
243 |
-
# This function is what interfaces with the ChatInterface
|
244 |
-
nonlocal conv_state
|
245 |
-
|
246 |
-
# Access the current state
|
247 |
-
current_state = conv_state.value if conv_state.value else []
|
248 |
-
|
249 |
-
# Call the main function that generates responses and updates state
|
250 |
-
# This is a generator function, so we need to iterate through its outputs
|
251 |
-
response_gen = enhanced_predict(message, history, temperature, top_p, current_state)
|
252 |
-
|
253 |
-
# For each response, yield it and also update our state at the end
|
254 |
-
last_response = None
|
255 |
-
for response in response_gen:
|
256 |
-
last_response = response
|
257 |
-
yield response
|
258 |
-
|
259 |
-
# After generation is complete, update our state
|
260 |
-
if last_response is not None:
|
261 |
-
# Create a full copy of the history plus the new exchange
|
262 |
-
updated_state = []
|
263 |
-
# Add all previous history
|
264 |
-
for msg in history:
|
265 |
-
updated_state.append(msg.copy())
|
266 |
-
# Add new exchange
|
267 |
-
updated_state.append({"role": "user", "content": message})
|
268 |
-
updated_state.append({"role": "assistant", "content": last_response})
|
269 |
-
|
270 |
-
# Store in our state
|
271 |
-
conv_state.value = updated_state
|
272 |
-
|
273 |
-
# Debug
|
274 |
-
print(f"Updated conversation state with {len(updated_state)} messages")
|
275 |
-
if updated_state:
|
276 |
-
last_msg = updated_state[-1]
|
277 |
-
print(f"Last message: {last_msg['role']}: {last_msg['content'][:30]}...")
|
278 |
|
279 |
# Create ChatInterface
|
280 |
chatbot = gr.ChatInterface(
|
281 |
chat_with_state,
|
282 |
additional_inputs=[
|
|
|
283 |
gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature"),
|
284 |
gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-P")
|
285 |
],
|
|
|
286 |
type="messages"
|
287 |
)
|
288 |
|
@@ -317,10 +207,10 @@ with gr.Blocks() as demo:
|
|
317 |
feedback_modal
|
318 |
)
|
319 |
|
320 |
-
# Connect the submit button to the submit_research_feedback function
|
321 |
submit_button.click(
|
322 |
submit_research_feedback,
|
323 |
-
inputs=[
|
324 |
outputs=response_text
|
325 |
)
|
326 |
|
|
|
5 |
import os
|
6 |
import json
|
7 |
import uuid
|
8 |
+
from datasets import Dataset
|
9 |
from huggingface_hub import HfApi, login
|
10 |
import time
|
11 |
|
|
|
28 |
# Ensure feedback directory exists
|
29 |
os.makedirs(DATASET_PATH, exist_ok=True)
|
30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
31 |
# Feedback storage functions
|
32 |
def save_feedback_locally(conversation, satisfaction, feedback_text):
|
33 |
"""Save feedback to a local JSONL file"""
|
|
|
93 |
print(f"Error pushing feedback data to Hub: {e}")
|
94 |
return False
|
95 |
|
96 |
+
# Modified predict function to update conversation state
|
97 |
+
@spaces.GPU(duration=120)
|
98 |
+
def predict(message, history, state, temperature, top_p):
|
99 |
+
# Update history with user message
|
100 |
+
history.append({"role": "user", "content": message})
|
101 |
+
|
102 |
+
|
103 |
+
input_text = tokenizer.apply_chat_template(history, tokenize=False, add_generation_prompt=True)
|
104 |
+
inputs = tokenizer.encode(input_text, return_tensors="pt").to(device)
|
105 |
+
|
106 |
+
# Create a streamer
|
107 |
+
streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
|
108 |
+
|
109 |
+
# Set up generation parameters
|
110 |
+
generation_kwargs = {
|
111 |
+
"input_ids": inputs,
|
112 |
+
"max_new_tokens": 1024,
|
113 |
+
"temperature": float(temperature),
|
114 |
+
"top_p": float(top_p),
|
115 |
+
"do_sample": True,
|
116 |
+
"streamer": streamer,
|
117 |
+
}
|
118 |
+
|
119 |
+
# Run generation in a separate thread
|
120 |
+
thread = Thread(target=model.generate, kwargs=generation_kwargs)
|
121 |
+
thread.start()
|
122 |
+
|
123 |
+
# Yield from the streamer as tokens are generated
|
124 |
+
partial_text = ""
|
125 |
+
for new_text in streamer:
|
126 |
+
partial_text += new_text
|
127 |
+
yield partial_text, state
|
128 |
+
|
129 |
+
# After full generation, update state with assistant's response
|
130 |
+
history.append({"role": "assistant", "content": partial_text})
|
131 |
+
state = history.copy()
|
132 |
+
return partial_text, state
|
133 |
+
|
134 |
# Function to handle the research feedback submission
|
135 |
+
def submit_research_feedback(conversation_state, satisfaction, feedback_text):
|
136 |
"""Save user feedback both locally and to HuggingFace Hub"""
|
|
|
|
|
|
|
|
|
|
|
|
|
137 |
# Save locally first
|
138 |
+
feedback_id = save_feedback_locally(conversation_state, satisfaction, feedback_text)
|
139 |
|
140 |
# Get token from environment variable
|
141 |
env_token = os.environ.get("HF_TOKEN")
|
|
|
150 |
|
151 |
return status_msg
|
152 |
|
|
|
|
|
|
|
|
|
|
|
153 |
# Create the Gradio blocks interface
|
154 |
with gr.Blocks() as demo:
|
155 |
+
# State to track conversation history
|
156 |
+
conversation_state = gr.State([])
|
157 |
|
158 |
with gr.Row():
|
159 |
with gr.Column(scale=3):
|
160 |
+
# Custom chat function wrapper to update state
|
161 |
+
def chat_with_state(message, history, state, temperature, top_p):
|
162 |
+
for partial_response, updated_state in predict(message, history, state, temperature, top_p):
|
163 |
+
# Update our state with each yield
|
164 |
+
state = updated_state
|
165 |
+
yield partial_response, state
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
166 |
|
167 |
# Create ChatInterface
|
168 |
chatbot = gr.ChatInterface(
|
169 |
chat_with_state,
|
170 |
additional_inputs=[
|
171 |
+
conversation_state,
|
172 |
gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature"),
|
173 |
gr.Slider(0.1, 1.0, value=0.9, step=0.05, label="Top-P")
|
174 |
],
|
175 |
+
additional_outputs=[conversation_state],
|
176 |
type="messages"
|
177 |
)
|
178 |
|
|
|
207 |
feedback_modal
|
208 |
)
|
209 |
|
210 |
+
# Connect the submit button to the submit_research_feedback function with the current conversation state
|
211 |
submit_button.click(
|
212 |
submit_research_feedback,
|
213 |
+
inputs=[conversation_state, satisfaction, feedback_text],
|
214 |
outputs=response_text
|
215 |
)
|
216 |
|