Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -13,69 +13,40 @@ import torch.distributions as dists # Added import
|
|
13 |
# --- START: Copied Helper functions from generation_utils.py ---
|
14 |
# [Keep the copied functions: top_p_logits, top_k_logits, sample_tokens]
|
15 |
def top_p_logits(logits, top_p=None):
|
16 |
-
|
17 |
-
if top_p is None or top_p >= 1.0:
|
18 |
-
return logits
|
19 |
sorted_logits, sorted_indices = torch.sort(logits, descending=True)
|
20 |
cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
|
21 |
sorted_indices_to_remove = cumulative_probs > top_p
|
22 |
-
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
|
23 |
-
|
24 |
-
|
25 |
-
mask = mask.scatter_(-1, sorted_indices, sorted_indices_to_remove)
|
26 |
-
logits = logits.masked_fill(mask, torch.finfo(logits.dtype).min)
|
27 |
-
return logits
|
28 |
|
29 |
def top_k_logits(logits, top_k=None):
|
30 |
-
|
31 |
-
if top_k is None or top_k <= 0:
|
32 |
-
return logits
|
33 |
top_k = min(top_k, logits.size(-1))
|
34 |
indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
|
35 |
-
|
36 |
-
return logits
|
37 |
|
38 |
def sample_tokens(logits, temperature=0.0, top_p=None, top_k=None, margin_confidence=False, neg_entropy=False):
|
39 |
-
|
40 |
-
if
|
41 |
-
|
42 |
-
logits = logits / safe_temp
|
43 |
-
if top_p is not None and 0.0 < top_p < 1.0:
|
44 |
-
logits = top_p_logits(logits, top_p)
|
45 |
-
if top_k is not None and top_k > 0:
|
46 |
-
logits = top_k_logits(logits, top_k)
|
47 |
is_all_neg_inf = torch.all(logits == torch.finfo(logits.dtype).min, dim=-1, keepdim=True)
|
48 |
-
if torch.any(is_all_neg_inf):
|
49 |
-
uniform_logits = torch.zeros_like(logits)
|
50 |
-
logits = torch.where(is_all_neg_inf, uniform_logits, logits)
|
51 |
probs = torch.softmax(logits, dim=-1)
|
52 |
-
probs = torch.clamp(probs, min=0.0)
|
53 |
-
probs = probs / probs.sum(dim=-1, keepdim=True)
|
54 |
-
probs = torch.nan_to_num(probs, nan=0.0)
|
55 |
if temperature > 0:
|
56 |
-
try:
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
confidence, x0 = probs.max(dim=-1)
|
62 |
-
else:
|
63 |
-
confidence, x0 = probs.max(dim=-1)
|
64 |
-
if margin_confidence:
|
65 |
-
sorted_probs, _ = torch.sort(probs, dim=-1, descending=True)
|
66 |
-
top1_probs = sorted_probs[..., 0]
|
67 |
-
top2_probs = sorted_probs[..., 1] if sorted_probs.shape[-1] > 1 else top1_probs
|
68 |
-
confidence = top1_probs - top2_probs
|
69 |
-
if neg_entropy:
|
70 |
-
epsilon = 1e-10
|
71 |
-
log_probs = torch.log(probs + epsilon)
|
72 |
-
confidence = torch.sum(probs * log_probs, dim=-1)
|
73 |
confidence = torch.nan_to_num(confidence, nan=0.0)
|
74 |
return confidence, x0
|
75 |
# --- END: Copied Helper functions ---
|
76 |
|
77 |
-
|
78 |
-
#
|
79 |
config = AutoConfig.from_pretrained("Dream-org/Dream-v0-Instruct-7B", trust_remote_code=True)
|
80 |
model_path = "Dream-org/Dream-v0-Instruct-7B"
|
81 |
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
@@ -104,10 +75,8 @@ try:
|
|
104 |
SPECIAL_TOKEN_IDS.add(IM_END_ID)
|
105 |
except KeyError: IM_START_ID, IM_END_ID = None, None
|
106 |
|
107 |
-
|
108 |
# --- Helper Functions ---
|
109 |
def parse_constraints(constraints_text: str) -> Dict[int, List[int]]:
|
110 |
-
""" Parses word constraints. """
|
111 |
constraints = {}
|
112 |
if not constraints_text: return constraints
|
113 |
parts = constraints_text.split(',')
|
@@ -119,55 +88,26 @@ def parse_constraints(constraints_text: str) -> Dict[int, List[int]]:
|
|
119 |
pos = int(pos_str.strip())
|
120 |
word = word.strip()
|
121 |
token_ids = []
|
122 |
-
if word:
|
123 |
-
text_to_encode = (" " + word) if (pos > 0 and not word.startswith(" ")) else word
|
124 |
-
token_ids = tokenizer.encode(text_to_encode, add_special_tokens=False)
|
125 |
if token_ids and pos >= 0: constraints[pos] = token_ids
|
126 |
elif not token_ids and word: print(f"Warning: Could not tokenize constraint word '{word}'")
|
127 |
except ValueError: print(f"Warning: Invalid position '{pos_str}' in constraint part '{part}'")
|
128 |
except Exception as e: print(f"Warning: Error processing constraint '{part}': {e}")
|
129 |
return constraints
|
130 |
|
131 |
-
|
132 |
-
"""
|
133 |
-
Formats chat history [[user, bot], [user, bot]] into [{'role': 'user', 'content': ...}, ...]
|
134 |
-
for the tokenizer's chat template.
|
135 |
-
"""
|
136 |
-
messages = []
|
137 |
-
# Ensure history is not empty and is properly structured
|
138 |
-
if not history:
|
139 |
-
return messages
|
140 |
-
for turn in history:
|
141 |
-
if not isinstance(turn, (list, tuple)) or len(turn) != 2:
|
142 |
-
print(f"Warning: Skipping malformed history turn: {turn}")
|
143 |
-
continue
|
144 |
-
user_msg, assistant_msg = turn
|
145 |
-
if user_msg is not None: # Check if user message exists
|
146 |
-
# Ensure content is a string
|
147 |
-
user_content = str(user_msg) if user_msg is not None else ""
|
148 |
-
messages.append({"role": "user", "content": user_content})
|
149 |
-
# Add assistant message only if it exists and is not None
|
150 |
-
if assistant_msg is not None:
|
151 |
-
assistant_content = str(assistant_msg) if assistant_msg is not None else ""
|
152 |
-
messages.append({"role": "assistant", "content": assistant_content})
|
153 |
-
# print(f"Formatted messages for template: {messages}") # Debug
|
154 |
-
return messages
|
155 |
|
156 |
def apply_constraints_to_state(
|
157 |
x: torch.Tensor, prompt_length: int, total_length: int,
|
158 |
parsed_constraints: Dict[int, List[int]], current_step: Optional[int] = None
|
159 |
) -> torch.Tensor:
|
160 |
-
""" Applies constraints to the state tensor `x`. """
|
161 |
modified_x = x.clone()
|
162 |
for rel_pos, word_token_ids in parsed_constraints.items():
|
163 |
-
abs_start_pos = prompt_length + rel_pos
|
164 |
-
abs_end_pos = abs_start_pos + len(word_token_ids)
|
165 |
if abs_start_pos < total_length and abs_end_pos <= total_length:
|
166 |
-
try:
|
167 |
-
|
168 |
-
|
169 |
-
except IndexError: print(f"Warning (Step {current_step}): Constraint OOB: {rel_pos}")
|
170 |
-
except Exception as e: print(f"Warning (Step {current_step}): Constraint failed {rel_pos}: {e}")
|
171 |
return modified_x
|
172 |
|
173 |
|
@@ -176,7 +116,7 @@ def apply_constraints_to_state(
|
|
176 |
@spaces.GPU
|
177 |
@torch.no_grad()
|
178 |
def generate_dream_response(
|
179 |
-
history: List[
|
180 |
gen_length: int,
|
181 |
steps: int,
|
182 |
constraints_text: str,
|
@@ -186,37 +126,32 @@ def generate_dream_response(
|
|
186 |
alg: str,
|
187 |
alg_temp: Optional[float],
|
188 |
visualization_delay: float
|
189 |
-
|
190 |
""" Generates text step-by-step and yields visualization states live. """
|
191 |
|
192 |
-
|
193 |
-
|
194 |
-
# Yield the current (potentially empty) history back
|
195 |
-
yield history, [("No valid input message found.", "red")], ""
|
196 |
return
|
197 |
|
198 |
# --- 1. Preparation ---
|
199 |
-
#
|
200 |
-
messages_for_template = format_chat_history(history)
|
201 |
parsed_constraints = parse_constraints(constraints_text)
|
202 |
|
203 |
try:
|
|
|
204 |
inputs = tokenizer.apply_chat_template(
|
205 |
-
|
206 |
return_tensors="pt",
|
207 |
return_dict=True,
|
208 |
-
add_generation_prompt=True #
|
209 |
)
|
210 |
input_ids = inputs.input_ids.to(device)
|
211 |
prompt_attention_mask = inputs.attention_mask.to(device) if 'attention_mask' in inputs else torch.ones_like(input_ids)
|
212 |
-
prompt_length = input_ids.shape[1]
|
213 |
-
# print(f"Prompt length for model: {prompt_length}") # Debug
|
214 |
-
# print(f"Input IDs to model (first 50): {input_ids[0, :50].tolist()}") # Debug
|
215 |
-
|
216 |
except Exception as e:
|
217 |
print(f"Error applying chat template: {e}")
|
218 |
-
# Yield
|
219 |
-
yield history, [("Error preparing input.", "red")]
|
220 |
return
|
221 |
|
222 |
eps = 1e-3
|
@@ -227,132 +162,111 @@ def generate_dream_response(
|
|
227 |
# --- 2. Initialize Generation State ---
|
228 |
total_length = prompt_length + gen_length
|
229 |
initial_generation_part = torch.full((1, gen_length), MASK_ID, dtype=torch.long, device=device)
|
|
|
230 |
x = torch.cat((input_ids, initial_generation_part), dim=1)
|
231 |
|
232 |
-
# --- Prepare Attention Mask ---
|
233 |
generation_attention_mask = torch.ones((1, gen_length), dtype=torch.long, device=device)
|
|
|
234 |
full_attention_mask_long = torch.cat((prompt_attention_mask, generation_attention_mask), dim=1)
|
|
|
235 |
attention_mask_for_model = full_attention_mask_long.to(model.dtype)
|
236 |
large_neg_val = torch.finfo(model.dtype).min
|
237 |
attention_mask_for_model = (1.0 - attention_mask_for_model) * large_neg_val
|
238 |
-
attention_mask_for_model = attention_mask_for_model.unsqueeze(1).unsqueeze(2) #
|
239 |
|
|
|
240 |
timesteps = torch.linspace(1, eps, steps + 1, device=device)
|
|
|
|
|
241 |
x = apply_constraints_to_state(x, prompt_length, total_length, parsed_constraints, current_step=-1)
|
242 |
|
243 |
-
# --- 3. Visualization &
|
244 |
previous_tokens_vis = None
|
245 |
-
#
|
246 |
-
|
247 |
-
# IMPORTANT: Gradio state needs the component to receive the *entire object* back if it's mutated.
|
248 |
-
# So yielding the modified `history` list itself should work.
|
249 |
-
history_for_yield = history # Reference the original list
|
250 |
|
251 |
# --- 4. Initial Yield (Masked State) ---
|
252 |
initial_generated_tokens = x[0, prompt_length:].cpu()
|
253 |
vis_data_initial = []
|
254 |
for tok_id in initial_generated_tokens.tolist():
|
255 |
-
|
|
|
|
|
256 |
previous_tokens_vis = initial_generated_tokens
|
257 |
-
# Yield the
|
258 |
-
yield
|
259 |
time.sleep(visualization_delay)
|
260 |
|
261 |
# --- 5. Step-by-Step Diffusion Loop ---
|
262 |
try:
|
263 |
start_time = time.time()
|
264 |
-
current_response_text = "" # Store intermediate text
|
265 |
-
|
266 |
for i in range(steps):
|
267 |
mask_index = (x == MASK_ID)
|
268 |
-
if not mask_index.any():
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
outputs = model(
|
273 |
-
input_ids=x,
|
274 |
-
attention_mask=attention_mask_for_model,
|
275 |
-
position_ids=None, use_cache=False, return_dict=True
|
276 |
-
)
|
277 |
logits = outputs.logits
|
278 |
-
logits = torch.cat([logits[:,:1], logits[:, :-1]], dim=1)
|
279 |
|
280 |
mask_logits = logits[mask_index]
|
281 |
-
if mask_logits.numel() == 0:
|
282 |
-
print(f"No masked tokens found for logit selection at step {i}. Stopping.")
|
283 |
-
break
|
284 |
|
285 |
t = timesteps[i]; s = timesteps[i + 1]
|
286 |
x_new_masked_part = torch.full_like(x[mask_index], MASK_ID, device=device, dtype=torch.long)
|
287 |
|
288 |
-
# [
|
289 |
if alg == 'origin':
|
290 |
p_transfer = (1.0 - s / t) if i < steps - 1 else 1.0
|
291 |
num_masked = mask_logits.shape[0]
|
292 |
transfer_indices_relative = torch.rand(num_masked, device=device) < p_transfer
|
293 |
logits_to_sample = mask_logits[transfer_indices_relative]
|
294 |
-
if logits_to_sample.numel() > 0:
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
use_margin = (alg == 'topk_margin'); use_entropy = (alg == 'entropy')
|
299 |
-
confidence, x0_candidates = sample_tokens(
|
300 |
-
mask_logits, temperature=temperature, top_p=top_p_val, top_k=top_k_val,
|
301 |
-
margin_confidence=use_margin, neg_entropy=use_entropy
|
302 |
-
)
|
303 |
num_mask_token = mask_logits.shape[0]
|
304 |
target_num_revealed_float = num_mask_token * (1.0 - s / t)
|
305 |
number_transfer_tokens = int(target_num_revealed_float) if i < steps - 1 else num_mask_token
|
306 |
if number_transfer_tokens > 0:
|
307 |
num_samples = min(number_transfer_tokens, num_mask_token)
|
308 |
if num_samples > 0:
|
309 |
-
transfer_indices_relative = torch.tensor([], dtype=torch.long, device=device)
|
310 |
-
if alg_temp_val is None or alg_temp_val <= 0: # Top-k
|
311 |
sort_metric = confidence if alg != 'entropy' else -confidence
|
312 |
k_topk = min(num_samples, sort_metric.numel())
|
313 |
if k_topk > 0: _, transfer_indices_relative = torch.topk(sort_metric, k=k_topk)
|
314 |
-
else: #
|
315 |
if confidence.numel() > 0:
|
316 |
-
conf_probs = confidence / alg_temp_val
|
317 |
-
|
318 |
-
|
319 |
-
conf_probs = F.softmax(conf_probs, dim=-1)
|
320 |
-
conf_probs = torch.clamp(conf_probs, min=0.0)
|
321 |
-
conf_probs = torch.nan_to_num(conf_probs, nan=0.0)
|
322 |
-
prob_sum = conf_probs.sum()
|
323 |
-
target_sum_tensor = torch.tensor(1.0, device=device, dtype=prob_sum.dtype)
|
324 |
-
if not torch.isclose(prob_sum, target_sum_tensor, atol=1e-4) and prob_sum > 0:
|
325 |
-
safe_prob_sum = torch.max(prob_sum, torch.tensor(1e-12, device=device, dtype=prob_sum.dtype))
|
326 |
-
conf_probs = conf_probs / safe_prob_sum
|
327 |
final_prob_sum_check = conf_probs.sum()
|
328 |
if conf_probs.numel() > 0 and num_samples > 0 and torch.all(conf_probs >= 0) and torch.isclose(final_prob_sum_check, target_sum_tensor, atol=1e-4):
|
329 |
try: transfer_indices_relative = torch.multinomial(conf_probs, num_samples=num_samples, replacement=False)
|
330 |
-
except RuntimeError as e: print(f"
|
331 |
-
if
|
332 |
-
sort_metric = confidence if alg != 'entropy' else -confidence
|
333 |
-
k_fallback = min(num_samples, sort_metric.numel())
|
334 |
-
if k_fallback > 0: _, transfer_indices_relative = torch.topk(sort_metric, k=k_fallback)
|
335 |
# Apply transfer
|
336 |
if transfer_indices_relative.numel() > 0:
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
|
|
|
342 |
|
343 |
-
|
|
|
344 |
x = apply_constraints_to_state(x, prompt_length, total_length, parsed_constraints, current_step=i)
|
345 |
|
346 |
# --- Yield Visualization ---
|
347 |
current_generated_tokens = x[0, prompt_length:].cpu()
|
348 |
vis_data = []
|
349 |
-
# [
|
350 |
for j in range(gen_length):
|
351 |
current_tok_id = current_generated_tokens[j].item()
|
352 |
previous_tok_id = previous_tokens_vis[j].item() if previous_tokens_vis is not None and j < len(previous_tokens_vis) else MASK_ID
|
353 |
-
try:
|
354 |
-
decoded_token = tokenizer.decode([current_tok_id], skip_special_tokens=False, clean_up_tokenization_spaces=False)
|
355 |
-
display_token = MASK_TOKEN if current_tok_id == MASK_ID else decoded_token
|
356 |
except Exception: display_token = f"[ID:{current_tok_id}]"
|
357 |
color = None; token_to_display = display_token
|
358 |
if current_tok_id == MASK_ID: color = "#444444"
|
@@ -361,27 +275,17 @@ def generate_dream_response(
|
|
361 |
should_hide = (PAD_ID is not None and current_tok_id == PAD_ID) or (EOS_ID is not None and current_tok_id == EOS_ID)
|
362 |
if should_hide and previous_tok_id == current_tok_id: token_to_display = ""; color = None
|
363 |
if token_to_display: vis_data.append((token_to_display, color))
|
364 |
-
# ---
|
365 |
|
366 |
previous_tokens_vis = current_generated_tokens
|
367 |
|
368 |
-
#
|
369 |
intermediate_response_tokens = x[0, prompt_length:]
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
-
|
374 |
-
|
375 |
-
|
376 |
-
# --- Update history for yield ---
|
377 |
-
# Update the placeholder in the *last turn* of the history list
|
378 |
-
if history_for_yield and history_for_yield[-1]:
|
379 |
-
history_for_yield[-1][1] = current_response_text + "..." # Indicate streaming
|
380 |
-
|
381 |
-
# --- Yield current state ---
|
382 |
-
yield history_for_yield, vis_data, current_response_text
|
383 |
time.sleep(visualization_delay)
|
384 |
-
# --- End loop iteration ---
|
385 |
|
386 |
end_time = time.time()
|
387 |
print(f"Dream generation finished in {end_time - start_time:.2f} seconds.")
|
@@ -389,49 +293,38 @@ def generate_dream_response(
|
|
389 |
# --- 6. Final Processing & Yield ---
|
390 |
final_sequence = x[0]
|
391 |
response_tokens = final_sequence[prompt_length:]
|
392 |
-
final_response_text = tokenizer.decode(
|
393 |
-
|
394 |
-
|
395 |
-
clean_up_tokenization_spaces=True
|
396 |
-
).strip()
|
397 |
|
398 |
-
# Update the history definitively with the final text
|
399 |
-
if history_for_yield and history_for_yield[-1]:
|
400 |
-
history_for_yield[-1][1] = final_response_text
|
401 |
-
|
402 |
-
# Format final visualization
|
403 |
final_generated_tokens = x[0, prompt_length:].cpu()
|
404 |
vis_data_final = []
|
405 |
-
# [
|
406 |
for j in range(gen_length):
|
407 |
-
|
408 |
-
|
409 |
-
|
410 |
-
|
411 |
-
|
412 |
-
|
413 |
-
|
414 |
-
|
415 |
-
|
416 |
-
|
417 |
-
|
418 |
-
|
419 |
-
|
420 |
-
|
421 |
-
|
422 |
-
# Yield the final state
|
423 |
-
yield history_for_yield, vis_data_final, final_response_text
|
424 |
print("Visualization streaming complete.")
|
425 |
|
426 |
except Exception as e:
|
427 |
print(f"Error during generation or processing: {e}")
|
428 |
import traceback
|
429 |
traceback.print_exc()
|
430 |
-
#
|
431 |
-
#
|
432 |
-
|
433 |
-
|
434 |
-
yield history_for_yield, [("Error during generation.", "red")], ""
|
435 |
return
|
436 |
|
437 |
|
@@ -448,17 +341,17 @@ def create_chatbot_demo():
|
|
448 |
"[[Blog](https://hkunlp.github.io/blog/2025/dream/)]"
|
449 |
)
|
450 |
|
451 |
-
#
|
452 |
-
chat_history_state = gr.State([])
|
453 |
|
454 |
with gr.Row():
|
455 |
with gr.Column(scale=3):
|
|
|
456 |
chatbot_ui = gr.Chatbot(
|
457 |
label="Conversation",
|
|
|
458 |
height=500,
|
459 |
show_copy_button=True,
|
460 |
bubble_full_width=False,
|
461 |
-
# value=[] # Initial value set by state binding later
|
462 |
)
|
463 |
with gr.Group():
|
464 |
with gr.Row():
|
@@ -474,12 +367,10 @@ def create_chatbot_demo():
|
|
474 |
)
|
475 |
with gr.Column(scale=2):
|
476 |
output_vis = gr.HighlightedText(
|
477 |
-
label="Denoising Process Visualization",
|
478 |
-
show_legend=False, interactive=False
|
479 |
-
)
|
480 |
-
response_text_display = gr.Textbox(
|
481 |
-
label="Generated Response (Live)", interactive=False, lines=5
|
482 |
)
|
|
|
483 |
|
484 |
with gr.Accordion("Generation Settings", open=False):
|
485 |
# [Settings sliders remain the same]
|
@@ -497,88 +388,75 @@ def create_chatbot_demo():
|
|
497 |
with gr.Row():
|
498 |
visualization_delay = gr.Slider(minimum=0.0, maximum=0.5, value=0.03, step=0.01, label="Visualization Delay (seconds)")
|
499 |
|
500 |
-
|
501 |
clear_btn = gr.Button("Clear Conversation")
|
502 |
|
503 |
-
# --- Event
|
504 |
|
505 |
-
|
506 |
-
|
507 |
-
Adds
|
508 |
-
for the bot's response (clearing previous outputs).
|
509 |
-
"""
|
510 |
if not message.strip():
|
511 |
gr.Warning("Please enter a message.")
|
512 |
-
# Return unchanged history
|
513 |
-
|
514 |
-
|
515 |
-
history
|
516 |
-
|
517 |
-
# empty input, empty vis, empty response text.
|
518 |
-
return history, history, "", [], ""
|
519 |
|
520 |
def clear_all():
|
521 |
-
"""Clears
|
522 |
-
return [], [], ""
|
523 |
|
524 |
# --- Connect UI elements ---
|
525 |
|
526 |
-
# Define inputs
|
|
|
527 |
generation_inputs = [
|
528 |
-
|
|
|
529 |
temperature, top_p, top_k, remasking_strategy, alg_temp,
|
530 |
visualization_delay
|
531 |
]
|
532 |
-
#
|
533 |
-
|
534 |
-
|
535 |
-
# Chain the actions: Submit/Click -> add_user_message -> generate_dream_response
|
536 |
-
|
537 |
-
# 1. User submits message (Enter or Button)
|
538 |
-
user_interaction = [user_input, chat_history_state]
|
539 |
-
outputs_after_user_add = [
|
540 |
-
chat_history_state, # Update the state
|
541 |
-
chatbot_ui, # Update chatbot UI immediately
|
542 |
-
user_input, # Clear user input box
|
543 |
-
output_vis, # Clear visualization
|
544 |
-
response_text_display # Clear response text box
|
545 |
-
]
|
546 |
|
|
|
547 |
submit_listener = user_input.submit(
|
548 |
-
fn=add_user_message,
|
549 |
-
inputs=
|
550 |
-
outputs=
|
551 |
-
|
|
|
552 |
fn=generate_dream_response,
|
553 |
-
inputs=generation_inputs,
|
554 |
-
outputs=generation_outputs, # Stream
|
555 |
show_progress="hidden"
|
556 |
)
|
557 |
|
|
|
558 |
click_listener = send_btn.click(
|
559 |
-
fn=add_user_message,
|
560 |
-
inputs=
|
561 |
-
outputs=
|
562 |
-
|
|
|
563 |
fn=generate_dream_response,
|
564 |
inputs=generation_inputs,
|
565 |
-
outputs=generation_outputs,
|
566 |
show_progress="hidden"
|
567 |
)
|
568 |
|
569 |
-
#
|
570 |
clear_btn.click(
|
571 |
-
clear_all,
|
572 |
inputs=[],
|
573 |
-
outputs=[
|
574 |
-
|
575 |
-
output_vis, response_text_display
|
576 |
-
]
|
577 |
)
|
578 |
|
579 |
return demo
|
580 |
|
581 |
-
|
582 |
# --- Launch ---
|
583 |
if __name__ == "__main__":
|
584 |
demo = create_chatbot_demo()
|
|
|
13 |
# --- START: Copied Helper functions from generation_utils.py ---
|
14 |
# [Keep the copied functions: top_p_logits, top_k_logits, sample_tokens]
|
15 |
def top_p_logits(logits, top_p=None):
|
16 |
+
if top_p is None or top_p >= 1.0: return logits
|
|
|
|
|
17 |
sorted_logits, sorted_indices = torch.sort(logits, descending=True)
|
18 |
cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
|
19 |
sorted_indices_to_remove = cumulative_probs > top_p
|
20 |
+
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone(); sorted_indices_to_remove[..., 0] = 0
|
21 |
+
mask = torch.zeros_like(logits, dtype=torch.bool, device=logits.device).scatter_(-1, sorted_indices, sorted_indices_to_remove)
|
22 |
+
return logits.masked_fill(mask, torch.finfo(logits.dtype).min)
|
|
|
|
|
|
|
23 |
|
24 |
def top_k_logits(logits, top_k=None):
|
25 |
+
if top_k is None or top_k <= 0: return logits
|
|
|
|
|
26 |
top_k = min(top_k, logits.size(-1))
|
27 |
indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
|
28 |
+
return logits.masked_fill(indices_to_remove, torch.finfo(logits.dtype).min)
|
|
|
29 |
|
30 |
def sample_tokens(logits, temperature=0.0, top_p=None, top_k=None, margin_confidence=False, neg_entropy=False):
|
31 |
+
if temperature > 0: safe_temp = max(temperature, 1e-6); logits = logits / safe_temp
|
32 |
+
if top_p is not None and 0.0 < top_p < 1.0: logits = top_p_logits(logits, top_p)
|
33 |
+
if top_k is not None and top_k > 0: logits = top_k_logits(logits, top_k)
|
|
|
|
|
|
|
|
|
|
|
34 |
is_all_neg_inf = torch.all(logits == torch.finfo(logits.dtype).min, dim=-1, keepdim=True)
|
35 |
+
if torch.any(is_all_neg_inf): uniform_logits = torch.zeros_like(logits); logits = torch.where(is_all_neg_inf, uniform_logits, logits)
|
|
|
|
|
36 |
probs = torch.softmax(logits, dim=-1)
|
37 |
+
probs = torch.clamp(probs, min=0.0); probs = probs / probs.sum(dim=-1, keepdim=True); probs = torch.nan_to_num(probs, nan=0.0)
|
|
|
|
|
38 |
if temperature > 0:
|
39 |
+
try: x0 = dists.Categorical(probs=probs).sample(); confidence = torch.gather(probs, -1, x0.unsqueeze(-1)).squeeze(-1)
|
40 |
+
except Exception as e: print(f"Warning: Sampling failed: {e}. Argmax fallback."); confidence, x0 = probs.max(dim=-1)
|
41 |
+
else: confidence, x0 = probs.max(dim=-1)
|
42 |
+
if margin_confidence: sorted_probs, _ = torch.sort(probs, dim=-1, descending=True); top1_probs = sorted_probs[..., 0]; top2_probs = sorted_probs[..., 1] if sorted_probs.shape[-1] > 1 else top1_probs; confidence = top1_probs - top2_probs
|
43 |
+
if neg_entropy: epsilon = 1e-10; log_probs = torch.log(probs + epsilon); confidence = torch.sum(probs * log_probs, dim=-1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
44 |
confidence = torch.nan_to_num(confidence, nan=0.0)
|
45 |
return confidence, x0
|
46 |
# --- END: Copied Helper functions ---
|
47 |
|
48 |
+
# [Keep model loading, constants as before]
|
49 |
+
# Load model configuration to get special token IDs
|
50 |
config = AutoConfig.from_pretrained("Dream-org/Dream-v0-Instruct-7B", trust_remote_code=True)
|
51 |
model_path = "Dream-org/Dream-v0-Instruct-7B"
|
52 |
device = 'cuda' if torch.cuda.is_available() else 'cpu'
|
|
|
75 |
SPECIAL_TOKEN_IDS.add(IM_END_ID)
|
76 |
except KeyError: IM_START_ID, IM_END_ID = None, None
|
77 |
|
|
|
78 |
# --- Helper Functions ---
|
79 |
def parse_constraints(constraints_text: str) -> Dict[int, List[int]]:
|
|
|
80 |
constraints = {}
|
81 |
if not constraints_text: return constraints
|
82 |
parts = constraints_text.split(',')
|
|
|
88 |
pos = int(pos_str.strip())
|
89 |
word = word.strip()
|
90 |
token_ids = []
|
91 |
+
if word: text_to_encode = (" " + word) if (pos > 0 and not word.startswith(" ")) else word; token_ids = tokenizer.encode(text_to_encode, add_special_tokens=False)
|
|
|
|
|
92 |
if token_ids and pos >= 0: constraints[pos] = token_ids
|
93 |
elif not token_ids and word: print(f"Warning: Could not tokenize constraint word '{word}'")
|
94 |
except ValueError: print(f"Warning: Invalid position '{pos_str}' in constraint part '{part}'")
|
95 |
except Exception as e: print(f"Warning: Error processing constraint '{part}': {e}")
|
96 |
return constraints
|
97 |
|
98 |
+
# Removed format_chat_history as history will be in the correct format
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
|
100 |
def apply_constraints_to_state(
|
101 |
x: torch.Tensor, prompt_length: int, total_length: int,
|
102 |
parsed_constraints: Dict[int, List[int]], current_step: Optional[int] = None
|
103 |
) -> torch.Tensor:
|
|
|
104 |
modified_x = x.clone()
|
105 |
for rel_pos, word_token_ids in parsed_constraints.items():
|
106 |
+
abs_start_pos = prompt_length + rel_pos; abs_end_pos = abs_start_pos + len(word_token_ids)
|
|
|
107 |
if abs_start_pos < total_length and abs_end_pos <= total_length:
|
108 |
+
try: constraint_tensor = torch.tensor(word_token_ids, dtype=torch.long, device=modified_x.device); modified_x[0, abs_start_pos:abs_end_pos] = constraint_tensor
|
109 |
+
except IndexError: print(f"Warning (Step {current_step}): Constraint idx error at {rel_pos}")
|
110 |
+
except Exception as e: print(f"Warning (Step {current_step}): Constraint apply error at {rel_pos}: {e}")
|
|
|
|
|
111 |
return modified_x
|
112 |
|
113 |
|
|
|
116 |
@spaces.GPU
|
117 |
@torch.no_grad()
|
118 |
def generate_dream_response(
|
119 |
+
history: List[Dict[str, str]], # MODIFIED: Expect List[Dict]
|
120 |
gen_length: int,
|
121 |
steps: int,
|
122 |
constraints_text: str,
|
|
|
126 |
alg: str,
|
127 |
alg_temp: Optional[float],
|
128 |
visualization_delay: float
|
129 |
+
): # Removed -> type hint for cleaner yield handling
|
130 |
""" Generates text step-by-step and yields visualization states live. """
|
131 |
|
132 |
+
if not history or history[-1]["role"] != "user": # Check last message is from user
|
133 |
+
yield history, [("No user message found to respond to.", "red")]
|
|
|
|
|
134 |
return
|
135 |
|
136 |
# --- 1. Preparation ---
|
137 |
+
# History is already formatted for the template
|
|
|
138 |
parsed_constraints = parse_constraints(constraints_text)
|
139 |
|
140 |
try:
|
141 |
+
# apply_chat_template expects List[Dict[str, str]]
|
142 |
inputs = tokenizer.apply_chat_template(
|
143 |
+
history, # Use history directly
|
144 |
return_tensors="pt",
|
145 |
return_dict=True,
|
146 |
+
add_generation_prompt=True # Crucial: Adds the "<|im_start|>assistant\n" prompt
|
147 |
)
|
148 |
input_ids = inputs.input_ids.to(device)
|
149 |
prompt_attention_mask = inputs.attention_mask.to(device) if 'attention_mask' in inputs else torch.ones_like(input_ids)
|
150 |
+
prompt_length = input_ids.shape[1] # Length *after* adding the generation prompt
|
|
|
|
|
|
|
151 |
except Exception as e:
|
152 |
print(f"Error applying chat template: {e}")
|
153 |
+
# Yield current history and error message for visualization
|
154 |
+
yield history, [("Error preparing input.", "red")]
|
155 |
return
|
156 |
|
157 |
eps = 1e-3
|
|
|
162 |
# --- 2. Initialize Generation State ---
|
163 |
total_length = prompt_length + gen_length
|
164 |
initial_generation_part = torch.full((1, gen_length), MASK_ID, dtype=torch.long, device=device)
|
165 |
+
# input_ids already includes the assistant prompt, so just append masks
|
166 |
x = torch.cat((input_ids, initial_generation_part), dim=1)
|
167 |
|
168 |
+
# --- Prepare Attention Mask for SDPA ---
|
169 |
generation_attention_mask = torch.ones((1, gen_length), dtype=torch.long, device=device)
|
170 |
+
# prompt_attention_mask corresponds to input_ids (which includes assistant prompt)
|
171 |
full_attention_mask_long = torch.cat((prompt_attention_mask, generation_attention_mask), dim=1)
|
172 |
+
|
173 |
attention_mask_for_model = full_attention_mask_long.to(model.dtype)
|
174 |
large_neg_val = torch.finfo(model.dtype).min
|
175 |
attention_mask_for_model = (1.0 - attention_mask_for_model) * large_neg_val
|
176 |
+
attention_mask_for_model = attention_mask_for_model.unsqueeze(1).unsqueeze(2) # [B, 1, 1, N]
|
177 |
|
178 |
+
# --- Timesteps ---
|
179 |
timesteps = torch.linspace(1, eps, steps + 1, device=device)
|
180 |
+
|
181 |
+
# Apply initial constraints (relative to start of generation = prompt_length)
|
182 |
x = apply_constraints_to_state(x, prompt_length, total_length, parsed_constraints, current_step=-1)
|
183 |
|
184 |
+
# --- 3. Visualization & History Setup ---
|
185 |
previous_tokens_vis = None
|
186 |
+
# MODIFIED: Append placeholder assistant message to the history state *before* looping
|
187 |
+
history.append({"role": "assistant", "content": ""})
|
|
|
|
|
|
|
188 |
|
189 |
# --- 4. Initial Yield (Masked State) ---
|
190 |
initial_generated_tokens = x[0, prompt_length:].cpu()
|
191 |
vis_data_initial = []
|
192 |
for tok_id in initial_generated_tokens.tolist():
|
193 |
+
display_token = MASK_TOKEN; color = "#444444"
|
194 |
+
vis_data_initial.append((display_token, color))
|
195 |
+
|
196 |
previous_tokens_vis = initial_generated_tokens
|
197 |
+
# Yield the history (which now includes the empty assistant message) and initial vis
|
198 |
+
yield history, vis_data_initial
|
199 |
time.sleep(visualization_delay)
|
200 |
|
201 |
# --- 5. Step-by-Step Diffusion Loop ---
|
202 |
try:
|
203 |
start_time = time.time()
|
|
|
|
|
204 |
for i in range(steps):
|
205 |
mask_index = (x == MASK_ID)
|
206 |
+
if not mask_index.any(): break # Stop early
|
207 |
+
|
208 |
+
outputs = model(input_ids=x, attention_mask=attention_mask_for_model, return_dict=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
209 |
logits = outputs.logits
|
210 |
+
logits = torch.cat([logits[:,:1], logits[:, :-1]], dim=1) # Align logits
|
211 |
|
212 |
mask_logits = logits[mask_index]
|
213 |
+
if mask_logits.numel() == 0: break # Stop early
|
|
|
|
|
214 |
|
215 |
t = timesteps[i]; s = timesteps[i + 1]
|
216 |
x_new_masked_part = torch.full_like(x[mask_index], MASK_ID, device=device, dtype=torch.long)
|
217 |
|
218 |
+
# [Keep sampling/remasking logic ('origin' and confidence-based) exactly the same]
|
219 |
if alg == 'origin':
|
220 |
p_transfer = (1.0 - s / t) if i < steps - 1 else 1.0
|
221 |
num_masked = mask_logits.shape[0]
|
222 |
transfer_indices_relative = torch.rand(num_masked, device=device) < p_transfer
|
223 |
logits_to_sample = mask_logits[transfer_indices_relative]
|
224 |
+
if logits_to_sample.numel() > 0: _, sampled_tokens = sample_tokens(logits_to_sample, temperature=temperature, top_p=top_p_val, top_k=top_k_val); x_new_masked_part[transfer_indices_relative] = sampled_tokens
|
225 |
+
else:
|
226 |
+
use_margin=(alg == 'topk_margin'); use_entropy=(alg == 'entropy')
|
227 |
+
confidence, x0_candidates = sample_tokens(mask_logits, temperature=temperature, top_p=top_p_val, top_k=top_k_val, margin_confidence=use_margin, neg_entropy=use_entropy)
|
|
|
|
|
|
|
|
|
|
|
228 |
num_mask_token = mask_logits.shape[0]
|
229 |
target_num_revealed_float = num_mask_token * (1.0 - s / t)
|
230 |
number_transfer_tokens = int(target_num_revealed_float) if i < steps - 1 else num_mask_token
|
231 |
if number_transfer_tokens > 0:
|
232 |
num_samples = min(number_transfer_tokens, num_mask_token)
|
233 |
if num_samples > 0:
|
234 |
+
transfer_indices_relative = torch.tensor([], dtype=torch.long, device=device)
|
235 |
+
if alg_temp_val is None or alg_temp_val <= 0: # Top-k confidence
|
236 |
sort_metric = confidence if alg != 'entropy' else -confidence
|
237 |
k_topk = min(num_samples, sort_metric.numel())
|
238 |
if k_topk > 0: _, transfer_indices_relative = torch.topk(sort_metric, k=k_topk)
|
239 |
+
else: # Sample based on confidence temperature
|
240 |
if confidence.numel() > 0:
|
241 |
+
conf_probs = confidence / alg_temp_val; conf_probs = torch.nan_to_num(conf_probs, nan=0.0, posinf=1e9, neginf=-1e9); conf_probs = torch.clamp(conf_probs - conf_probs.max(), min=-30); conf_probs = F.softmax(conf_probs, dim=-1); conf_probs = torch.clamp(conf_probs, min=0.0); conf_probs = torch.nan_to_num(conf_probs, nan=0.0)
|
242 |
+
prob_sum = conf_probs.sum(); target_sum_tensor = torch.tensor(1.0, device=device, dtype=prob_sum.dtype)
|
243 |
+
if not torch.isclose(prob_sum, target_sum_tensor, atol=1e-4) and prob_sum > 0: safe_prob_sum = torch.max(prob_sum, torch.tensor(1e-12, device=device, dtype=prob_sum.dtype)); conf_probs = conf_probs / safe_prob_sum
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
244 |
final_prob_sum_check = conf_probs.sum()
|
245 |
if conf_probs.numel() > 0 and num_samples > 0 and torch.all(conf_probs >= 0) and torch.isclose(final_prob_sum_check, target_sum_tensor, atol=1e-4):
|
246 |
try: transfer_indices_relative = torch.multinomial(conf_probs, num_samples=num_samples, replacement=False)
|
247 |
+
except RuntimeError as e: print(f"Warning step {i}: Multinomial failed ('{e}'). Fallback."); sort_metric = confidence if alg != 'entropy' else -confidence; k_fallback = min(num_samples, sort_metric.numel()); if k_fallback > 0: _, transfer_indices_relative = torch.topk(sort_metric, k=k_fallback)
|
248 |
+
else: sort_metric = confidence if alg != 'entropy' else -confidence; k_fallback = min(num_samples, sort_metric.numel()); if k_fallback > 0: _, transfer_indices_relative = torch.topk(sort_metric, k=k_fallback)
|
|
|
|
|
|
|
249 |
# Apply transfer
|
250 |
if transfer_indices_relative.numel() > 0:
|
251 |
+
valid_indices = transfer_indices_relative < x0_candidates.shape[0]; valid_transfer_indices = transfer_indices_relative[valid_indices]
|
252 |
+
if valid_transfer_indices.numel() > 0:
|
253 |
+
if valid_transfer_indices.max() < x_new_masked_part.shape[0]: x_new_masked_part[valid_transfer_indices] = x0_candidates[valid_transfer_indices].clone()
|
254 |
+
else: print(f"Warning step {i}: transfer_indices OOB for x_new_masked_part.")
|
255 |
|
256 |
+
x[mask_index] = x_new_masked_part # Update state
|
257 |
|
258 |
+
# --- Apply Constraints ---
|
259 |
+
# Remember prompt_length now includes the assistant prompt turn
|
260 |
x = apply_constraints_to_state(x, prompt_length, total_length, parsed_constraints, current_step=i)
|
261 |
|
262 |
# --- Yield Visualization ---
|
263 |
current_generated_tokens = x[0, prompt_length:].cpu()
|
264 |
vis_data = []
|
265 |
+
# [Keep visualization formatting logic the same]
|
266 |
for j in range(gen_length):
|
267 |
current_tok_id = current_generated_tokens[j].item()
|
268 |
previous_tok_id = previous_tokens_vis[j].item() if previous_tokens_vis is not None and j < len(previous_tokens_vis) else MASK_ID
|
269 |
+
try: decoded_token = tokenizer.decode([current_tok_id], skip_special_tokens=False, clean_up_tokenization_spaces=False); display_token = MASK_TOKEN if current_tok_id == MASK_ID else decoded_token
|
|
|
|
|
270 |
except Exception: display_token = f"[ID:{current_tok_id}]"
|
271 |
color = None; token_to_display = display_token
|
272 |
if current_tok_id == MASK_ID: color = "#444444"
|
|
|
275 |
should_hide = (PAD_ID is not None and current_tok_id == PAD_ID) or (EOS_ID is not None and current_tok_id == EOS_ID)
|
276 |
if should_hide and previous_tok_id == current_tok_id: token_to_display = ""; color = None
|
277 |
if token_to_display: vis_data.append((token_to_display, color))
|
|
|
278 |
|
279 |
previous_tokens_vis = current_generated_tokens
|
280 |
|
281 |
+
# MODIFIED: Update the *content* of the last history item
|
282 |
intermediate_response_tokens = x[0, prompt_length:]
|
283 |
+
intermediate_response_text = tokenizer.decode(intermediate_response_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=True).strip()
|
284 |
+
history[-1]["content"] = intermediate_response_text # Update last dict entry
|
285 |
+
|
286 |
+
# Yield the updated history list and current vis data
|
287 |
+
yield history, vis_data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
288 |
time.sleep(visualization_delay)
|
|
|
289 |
|
290 |
end_time = time.time()
|
291 |
print(f"Dream generation finished in {end_time - start_time:.2f} seconds.")
|
|
|
293 |
# --- 6. Final Processing & Yield ---
|
294 |
final_sequence = x[0]
|
295 |
response_tokens = final_sequence[prompt_length:]
|
296 |
+
final_response_text = tokenizer.decode(response_tokens, skip_special_tokens=True, clean_up_tokenization_spaces=True).strip()
|
297 |
+
# Update the final content in the history object
|
298 |
+
history[-1]["content"] = final_response_text
|
|
|
|
|
299 |
|
|
|
|
|
|
|
|
|
|
|
300 |
final_generated_tokens = x[0, prompt_length:].cpu()
|
301 |
vis_data_final = []
|
302 |
+
# [Keep final visualization formatting logic the same]
|
303 |
for j in range(gen_length):
|
304 |
+
current_tok_id = final_generated_tokens[j].item()
|
305 |
+
previous_tok_id = previous_tokens_vis[j].item() if previous_tokens_vis is not None and j < len(previous_tokens_vis) else MASK_ID
|
306 |
+
try: decoded_token = tokenizer.decode([current_tok_id], skip_special_tokens=False, clean_up_tokenization_spaces=False); display_token = MASK_TOKEN if current_tok_id == MASK_ID else decoded_token
|
307 |
+
except Exception: display_token = f"[ID:{current_tok_id}]"
|
308 |
+
color = None; token_to_display = display_token
|
309 |
+
if current_tok_id == MASK_ID: color = "#444444"
|
310 |
+
elif previous_tok_id == MASK_ID: color = "#66CC66"
|
311 |
+
else: color = "#6699CC"
|
312 |
+
should_hide = (PAD_ID is not None and current_tok_id == PAD_ID) or (EOS_ID is not None and current_tok_id == EOS_ID)
|
313 |
+
if should_hide and previous_tok_id == current_tok_id: token_to_display = ""; color = None
|
314 |
+
if token_to_display: vis_data_final.append((token_to_display, color))
|
315 |
+
|
316 |
+
# Yield final history and visualization
|
317 |
+
yield history, vis_data_final
|
|
|
|
|
|
|
318 |
print("Visualization streaming complete.")
|
319 |
|
320 |
except Exception as e:
|
321 |
print(f"Error during generation or processing: {e}")
|
322 |
import traceback
|
323 |
traceback.print_exc()
|
324 |
+
# Set error message in the last history item? Or yield separate error?
|
325 |
+
# Let's just yield the current history and error vis
|
326 |
+
history[-1]["content"] = f"Error: {e}" # Put error in assistant message
|
327 |
+
yield history, [("Error during generation.", "red")]
|
|
|
328 |
return
|
329 |
|
330 |
|
|
|
341 |
"[[Blog](https://hkunlp.github.io/blog/2025/dream/)]"
|
342 |
)
|
343 |
|
344 |
+
# STATE: No explicit state needed if chatbot manages it via input/output
|
|
|
345 |
|
346 |
with gr.Row():
|
347 |
with gr.Column(scale=3):
|
348 |
+
# MODIFIED: Use type="messages"
|
349 |
chatbot_ui = gr.Chatbot(
|
350 |
label="Conversation",
|
351 |
+
type="messages", # Use dictionary format
|
352 |
height=500,
|
353 |
show_copy_button=True,
|
354 |
bubble_full_width=False,
|
|
|
355 |
)
|
356 |
with gr.Group():
|
357 |
with gr.Row():
|
|
|
367 |
)
|
368 |
with gr.Column(scale=2):
|
369 |
output_vis = gr.HighlightedText(
|
370 |
+
label="Denoising Process Visualization",
|
371 |
+
combine_adjacent=True, show_legend=False, interactive=False
|
|
|
|
|
|
|
372 |
)
|
373 |
+
# REMOVED: Separate response text display
|
374 |
|
375 |
with gr.Accordion("Generation Settings", open=False):
|
376 |
# [Settings sliders remain the same]
|
|
|
388 |
with gr.Row():
|
389 |
visualization_delay = gr.Slider(minimum=0.0, maximum=0.5, value=0.03, step=0.01, label="Visualization Delay (seconds)")
|
390 |
|
|
|
391 |
clear_btn = gr.Button("Clear Conversation")
|
392 |
|
393 |
+
# --- Event Handlers ---
|
394 |
|
395 |
+
# MODIFIED: add_user_message uses dictionary format
|
396 |
+
def add_user_message(message: str, history: List[Dict[str, str]]):
|
397 |
+
"""Adds user message in dictionary format, clears input."""
|
|
|
|
|
398 |
if not message.strip():
|
399 |
gr.Warning("Please enter a message.")
|
400 |
+
return history, "" # Return unchanged history, don't clear input here
|
401 |
+
# Append user message as a dictionary
|
402 |
+
history.append({"role": "user", "content": message})
|
403 |
+
# Return updated history, clear input box
|
404 |
+
return history, ""
|
|
|
|
|
405 |
|
406 |
def clear_all():
|
407 |
+
"""Clears chatbot, visualization, and input."""
|
408 |
+
return [], [], "" # Chatbot, Vis, Input
|
409 |
|
410 |
# --- Connect UI elements ---
|
411 |
|
412 |
+
# Define the inputs for the generation function
|
413 |
+
# MODIFIED: Input is chatbot_ui (provides List[Dict])
|
414 |
generation_inputs = [
|
415 |
+
chatbot_ui, # Get history directly from chatbot component
|
416 |
+
gen_length, steps, constraints_input,
|
417 |
temperature, top_p, top_k, remasking_strategy, alg_temp,
|
418 |
visualization_delay
|
419 |
]
|
420 |
+
# Define the outputs for the generation function
|
421 |
+
# MODIFIED: Output history (List[Dict]) to chatbot_ui, vis_data to output_vis
|
422 |
+
generation_outputs = [chatbot_ui, output_vis]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
423 |
|
424 |
+
# Handle Textbox Submission (Enter key)
|
425 |
submit_listener = user_input.submit(
|
426 |
+
fn=add_user_message, # Use modified function
|
427 |
+
inputs=[user_input, chatbot_ui], # Pass chatbot state
|
428 |
+
outputs=[chatbot_ui, user_input], # Update chatbot state, clear input
|
429 |
+
queue=False # User message add should be quick
|
430 |
+
).then(
|
431 |
fn=generate_dream_response,
|
432 |
+
inputs=generation_inputs,
|
433 |
+
outputs=generation_outputs, # Stream history to chatbot, vis to output_vis
|
434 |
show_progress="hidden"
|
435 |
)
|
436 |
|
437 |
+
# Handle Send Button Click
|
438 |
click_listener = send_btn.click(
|
439 |
+
fn=add_user_message, # Use modified function
|
440 |
+
inputs=[user_input, chatbot_ui], # Pass chatbot state
|
441 |
+
outputs=[chatbot_ui, user_input], # Update chatbot state, clear input
|
442 |
+
queue=False # User message add should be quick
|
443 |
+
).then(
|
444 |
fn=generate_dream_response,
|
445 |
inputs=generation_inputs,
|
446 |
+
outputs=generation_outputs, # Stream history to chatbot, vis to output_vis
|
447 |
show_progress="hidden"
|
448 |
)
|
449 |
|
450 |
+
# Clear Button Action
|
451 |
clear_btn.click(
|
452 |
+
clear_all, # Use modified clear function
|
453 |
inputs=[],
|
454 |
+
outputs=[chatbot_ui, output_vis, user_input], # Clear chatbot, vis, input
|
455 |
+
queue=False
|
|
|
|
|
456 |
)
|
457 |
|
458 |
return demo
|
459 |
|
|
|
460 |
# --- Launch ---
|
461 |
if __name__ == "__main__":
|
462 |
demo = create_chatbot_demo()
|