Spaces:
Running
on
Zero
Running
on
Zero
Add pause length
Browse files
app.py
CHANGED
|
@@ -144,7 +144,7 @@ def generate_diffusion_text(input_ids):
|
|
| 144 |
return sampled, conf
|
| 145 |
|
| 146 |
# --- Inference Wrapper ---
|
| 147 |
-
def diffusion_chat(question, eot_weight, max_it, sharpness,
|
| 148 |
placeholder = "What do you know about the city of New York?"
|
| 149 |
if question.strip() == "":
|
| 150 |
question = placeholder
|
|
@@ -193,7 +193,7 @@ def diffusion_chat(question, eot_weight, max_it, sharpness, noise_clipping, use_
|
|
| 193 |
|
| 194 |
prev_decoded_tokens = decoded_tokens
|
| 195 |
yield f"<b>Iteration {i+1}/{max_it} (after generation):</b><br>" + "".join(highlighted).replace('\n', '<br>')
|
| 196 |
-
time.sleep(
|
| 197 |
|
| 198 |
# --- Early stopping ---
|
| 199 |
last_tokens.append(current_tokens)
|
|
@@ -235,7 +235,7 @@ def diffusion_chat(question, eot_weight, max_it, sharpness, noise_clipping, use_
|
|
| 235 |
highlighted.append(token_str)
|
| 236 |
|
| 237 |
yield f"<b>Iteration {i+1}/{max_it} (before noising):</b><br>" + "".join(highlighted).replace('\n', '<br>')
|
| 238 |
-
time.sleep(
|
| 239 |
|
| 240 |
|
| 241 |
final_tokens = tokenizer.convert_ids_to_tokens(current_tokens[answer_start:])
|
|
@@ -255,10 +255,12 @@ demo = gr.Interface(
|
|
| 255 |
gr.Textbox(label="User Question", lines=2, placeholder="What do you know about the city of New York?"),
|
| 256 |
gr.Slider(0, 1, value=0.4, step=0.05, label="↓ = longer answers (EOT weight)"),
|
| 257 |
gr.Slider(1, 512, value=64, step=1, label="↑ = more iterations"),
|
|
|
|
| 258 |
gr.Slider(1.0, 20.0, value=5.0, step=0.5, label="↓ = more noising (sharpness)"),
|
| 259 |
-
gr.Slider(0.01, 1.0, value=0.05, step=0.01, label="↓ = more confidence guidance (noise clipping)"),
|
| 260 |
-
gr.Checkbox(value=False, label="Use confidence-guided noising"),
|
| 261 |
gr.Slider(0.0, 1.0, value=0.5, step=0.05, label="↑ = more clustered noising (fewer, larger edits)")
|
|
|
|
|
|
|
|
|
|
| 262 |
],
|
| 263 |
outputs=[gr.HTML(label="Diffusion Output")],
|
| 264 |
title="Diffusion Language Model Chat",
|
|
|
|
| 144 |
return sampled, conf
|
| 145 |
|
| 146 |
# --- Inference Wrapper ---
|
| 147 |
+
def diffusion_chat(question, eot_weight, max_it, pause_length, sharpness, clustering, use_confidence_noising, noise_clipping):
|
| 148 |
placeholder = "What do you know about the city of New York?"
|
| 149 |
if question.strip() == "":
|
| 150 |
question = placeholder
|
|
|
|
| 193 |
|
| 194 |
prev_decoded_tokens = decoded_tokens
|
| 195 |
yield f"<b>Iteration {i+1}/{max_it} (after generation):</b><br>" + "".join(highlighted).replace('\n', '<br>')
|
| 196 |
+
time.sleep(pause_length)
|
| 197 |
|
| 198 |
# --- Early stopping ---
|
| 199 |
last_tokens.append(current_tokens)
|
|
|
|
| 235 |
highlighted.append(token_str)
|
| 236 |
|
| 237 |
yield f"<b>Iteration {i+1}/{max_it} (before noising):</b><br>" + "".join(highlighted).replace('\n', '<br>')
|
| 238 |
+
time.sleep(pause_length)
|
| 239 |
|
| 240 |
|
| 241 |
final_tokens = tokenizer.convert_ids_to_tokens(current_tokens[answer_start:])
|
|
|
|
| 255 |
gr.Textbox(label="User Question", lines=2, placeholder="What do you know about the city of New York?"),
|
| 256 |
gr.Slider(0, 1, value=0.4, step=0.05, label="↓ = longer answers (EOT weight)"),
|
| 257 |
gr.Slider(1, 512, value=64, step=1, label="↑ = more iterations"),
|
| 258 |
+
gr.Slider(0.01, 5, value=0.01, step=0.01, label="↑ = longer pause (for visualization)"),
|
| 259 |
gr.Slider(1.0, 20.0, value=5.0, step=0.5, label="↓ = more noising (sharpness)"),
|
|
|
|
|
|
|
| 260 |
gr.Slider(0.0, 1.0, value=0.5, step=0.05, label="↑ = more clustered noising (fewer, larger edits)")
|
| 261 |
+
gr.Checkbox(value=False, label="Use confidence-guided noising"),
|
| 262 |
+
gr.Slider(0.01, 1.0, value=0.05, step=0.01, label="↓ = more confidence guidance (noise clipping)"),
|
| 263 |
+
|
| 264 |
],
|
| 265 |
outputs=[gr.HTML(label="Diffusion Output")],
|
| 266 |
title="Diffusion Language Model Chat",
|