Spaces:
Running
on
Zero
Running
on
Zero
Commit
·
f2f927b
1
Parent(s):
ad774a9
cleaning things up via gemini 2.5 pro
Browse files
app.py
CHANGED
@@ -3,420 +3,510 @@ import gradio as gr
|
|
3 |
import torch
|
4 |
import itertools # For color cycling
|
5 |
import tiktoken # For GPT-4 tokenizer
|
6 |
-
from transformers import AutoTokenizer # For Llama3 tokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
7 |
|
8 |
-
# Bytelatent
|
|
|
|
|
9 |
try:
|
10 |
from bytelatent.data.file_util import get_fs
|
11 |
from bytelatent.generate_patcher import patcher_nocache
|
12 |
from bytelatent.tokenizers.blt_tokenizer import BltTokenizer
|
13 |
from bytelatent.plotting.entropy_figure_via_matplot_lib import plot_entropies
|
14 |
from bytelatent.args import TrainArgs
|
15 |
-
from download_blt_weights import main as ensure_present
|
16 |
-
|
|
|
17 |
except ImportError as e:
|
18 |
-
|
19 |
-
|
20 |
# Define dummy classes/functions if BLT is not available to avoid NameErrors later
|
21 |
class BltTokenizer: pass
|
22 |
class TrainArgs: pass
|
23 |
def patcher_nocache(*args, **kwargs): return None
|
24 |
def plot_entropies(*args, **kwargs): return None
|
25 |
def ensure_present(*args, **kwargs): pass
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
|
|
|
27 |
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
VIZ_COLORS = [
|
32 |
-
"#a6cee3", "#1f78b4", "#b2df8a", "#33a02c", "#fb9a99", "#e31a1c",
|
33 |
-
"#fdbf6f", "#ff7f00", "#cab2d6", "#6a3d9a", "#ffff99", "#b15928"
|
34 |
-
] # Add more if you expect many segments
|
35 |
-
|
36 |
-
LLAMA3_MODEL_NAME = "meta-llama/Meta-Llama-3-8B" # Or choose another variant like Instruct
|
37 |
-
|
38 |
-
# --- Helper Functions ---
|
39 |
-
|
40 |
-
def create_bytelatent_highlight_data(tokenizer, patch_lengths_tensor, tokens_tensor, colors):
|
41 |
-
"""Generates data for gr.HighlightedText based on bytelatent patches."""
|
42 |
-
if not BLT_AVAILABLE:
|
43 |
-
return [("Bytelatent library not available.", "Error")]
|
44 |
-
if patch_lengths_tensor is None or tokens_tensor is None or patch_lengths_tensor.numel() == 0:
|
45 |
-
return None
|
46 |
-
patch_lengths = patch_lengths_tensor.tolist()
|
47 |
-
all_tokens = tokens_tensor.tolist()
|
48 |
-
highlighted_data = []
|
49 |
-
current_token_index = 0
|
50 |
-
patch_count = 0 # Initialize patch count
|
51 |
-
# color_cycler = itertools.cycle(colors) # Moved inside loop if needed per-patch
|
52 |
-
for i, length in enumerate(patch_lengths):
|
53 |
-
if length <= 0: continue
|
54 |
-
patch_token_ids = all_tokens[current_token_index : current_token_index + length]
|
55 |
-
if not patch_token_ids: continue
|
56 |
-
try: patch_text = tokenizer.decode(patch_token_ids)
|
57 |
-
except Exception as decode_err:
|
58 |
-
print(f"Warning: Bytelatent patch decoding failed: {decode_err}")
|
59 |
-
patch_text = f"[Decode Error: {len(patch_token_ids)} tokens]"
|
60 |
-
patch_label = f"BL Patch {i+1}"
|
61 |
-
highlighted_data.append((patch_text, patch_label))
|
62 |
-
patch_count += 1 # Increment count for each valid patch added
|
63 |
-
current_token_index += length
|
64 |
-
|
65 |
-
# Handle remainder separately, don't count it as a 'patch'
|
66 |
-
if current_token_index != len(all_tokens):
|
67 |
-
print(f"Warning: Bytelatent token mismatch. Consumed {current_token_index}, total {len(all_tokens)}")
|
68 |
-
remaining_tokens = all_tokens[current_token_index:]
|
69 |
-
if remaining_tokens:
|
70 |
-
try: remaining_text = tokenizer.decode(remaining_tokens)
|
71 |
-
except Exception: remaining_text = f"[Decode Error: {len(remaining_tokens)} remaining tokens]"
|
72 |
-
highlighted_data.append((remaining_text, "BL Remainder"))
|
73 |
-
|
74 |
-
# Return both highlighted data and the calculated patch count
|
75 |
-
return highlighted_data, patch_count
|
76 |
-
|
77 |
-
|
78 |
-
def create_tiktoken_highlight_data(prompt, colors):
|
79 |
-
"""Generates data for gr.HighlightedText based on tiktoken (gpt-4) tokens."""
|
80 |
try:
|
81 |
-
|
82 |
-
tiktoken_ids = enc.encode(prompt)
|
83 |
highlighted_data = []
|
84 |
-
# color_cycler = itertools.cycle(colors) # Moved inside loop if needed per-token
|
85 |
for i, token_id in enumerate(tiktoken_ids):
|
86 |
-
try:
|
87 |
-
|
|
|
88 |
try:
|
89 |
-
token_bytes =
|
90 |
token_text = f"[Bytes: {token_bytes.hex()}]"
|
91 |
except Exception: token_text = "[Decode Error]"
|
92 |
except Exception as e:
|
93 |
-
|
94 |
-
|
|
|
95 |
token_label = f"GPT4 Tk {i+1}"
|
96 |
highlighted_data.append((token_text, token_label))
|
|
|
97 |
token_count = len(tiktoken_ids)
|
98 |
-
|
99 |
-
|
100 |
-
|
101 |
-
|
102 |
-
|
103 |
-
|
104 |
-
|
105 |
-
|
|
|
106 |
|
107 |
|
108 |
-
def create_llama3_highlight_data(prompt,
|
109 |
"""Generates data for gr.HighlightedText based on Llama 3 tokenizer."""
|
|
|
110 |
try:
|
111 |
-
# Load Llama 3 tokenizer from Hugging Face Hub
|
112 |
-
print(f"Loading Llama 3 tokenizer: {model_name}")
|
113 |
-
# Use trust_remote_code=True if required by the specific model revision
|
114 |
-
tokenizer = AutoTokenizer.from_pretrained(model_name) #, trust_remote_code=True)
|
115 |
-
print("Llama 3 tokenizer loaded.")
|
116 |
-
|
117 |
-
# Encode the prompt
|
118 |
llama_token_ids = tokenizer.encode(prompt)
|
119 |
-
|
120 |
highlighted_data = []
|
121 |
-
# color_cycler = itertools.cycle(colors) # Moved inside loop if needed per-token
|
122 |
-
|
123 |
for i, token_id in enumerate(llama_token_ids):
|
124 |
try:
|
125 |
-
# Decode individual token.
|
126 |
token_text = tokenizer.decode([token_id])
|
127 |
except Exception as e:
|
128 |
-
|
129 |
token_text = "[Decode Error]"
|
130 |
|
131 |
-
token_label = f"Llama3 Tk {i+1}"
|
132 |
highlighted_data.append((token_text, token_label))
|
133 |
|
134 |
token_count = len(llama_token_ids)
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
153 |
|
154 |
|
155 |
# --- Main Processing Function ---
|
156 |
|
157 |
-
def process_text(prompt: str
|
|
|
|
|
|
|
|
|
|
|
158 |
"""
|
159 |
Processes the input prompt using ByteLatent, Tiktoken, and Llama 3,
|
160 |
returning visualizations, counts, and status.
|
161 |
-
|
162 |
-
Args:
|
163 |
-
prompt: The input text string from the Gradio interface.
|
164 |
-
model_name: The name of the bytelatent model to use.
|
165 |
-
|
166 |
-
Returns:
|
167 |
-
A tuple containing:
|
168 |
-
- Matplotlib Figure for the entropy plot (or None).
|
169 |
-
- List of tuples for bytelatent gr.HighlightedText (or None).
|
170 |
-
- Integer count of bytelatent patches.
|
171 |
-
- List of tuples for tiktoken gr.HighlightedText (or None).
|
172 |
-
- Integer count of tiktoken tokens.
|
173 |
-
- List of tuples for Llama 3 gr.HighlightedText (or None).
|
174 |
-
- Integer count of Llama 3 tokens.
|
175 |
-
- Status/Error message string.
|
176 |
"""
|
|
|
177 |
fig = None
|
178 |
-
bl_highlighted_data =
|
179 |
-
tk_highlighted_data =
|
180 |
-
llama_highlighted_data =
|
181 |
-
|
182 |
-
|
183 |
-
|
184 |
-
|
185 |
-
|
186 |
-
# --- 1. Tiktoken Processing (Independent) ---
|
187 |
-
status_message += "\nProcessing with Tiktoken (gpt-4)..."
|
188 |
-
tk_highlighted_data, tk_count_calc = create_tiktoken_highlight_data(prompt, VIZ_COLORS)
|
189 |
-
if tk_highlighted_data and tk_highlighted_data[0][1] == "Error":
|
190 |
-
status_message += f"\nTiktoken Error: {tk_highlighted_data[0][0]}"
|
191 |
-
tk_count = 0 # Ensure count is 0 on error
|
192 |
-
else:
|
193 |
-
tk_count = tk_count_calc # Assign calculated count
|
194 |
-
status_message += f"\nTiktoken processing successful ({tk_count} tokens)."
|
195 |
-
|
196 |
-
# --- 2. Llama 3 Processing (Independent) ---
|
197 |
-
status_message += "\nProcessing with Llama 3 tokenizer..."
|
198 |
-
llama_highlighted_data, llama_count_calc = create_llama3_highlight_data(prompt, VIZ_COLORS)
|
199 |
-
if llama_highlighted_data and llama_highlighted_data[0][1] == "Error":
|
200 |
-
status_message += f"\nLlama 3 Error: {llama_highlighted_data[0][0]}"
|
201 |
-
llama_count = 0 # Ensure count is 0 on error
|
202 |
else:
|
203 |
-
|
204 |
-
status_message += f"\nLlama 3 processing successful ({llama_count} tokens)."
|
205 |
-
|
206 |
-
# --- 3. Bytelatent Processing ---
|
207 |
-
if BLT_AVAILABLE:
|
208 |
-
try:
|
209 |
-
status_message += f"\nLoading Bytelatent entropy model for '{model_name}'..."
|
210 |
-
# (Bytelatent loading code remains the same)
|
211 |
-
consolidated_path = os.path.join("hf-weights", model_name)
|
212 |
-
train_args_path = os.path.join(consolidated_path, "params.json")
|
213 |
-
if not os.path.exists(train_args_path): raise FileNotFoundError(f"BLT training args not found at {train_args_path}.")
|
214 |
-
fs = get_fs(train_args_path); train_args = TrainArgs.model_validate_json(fs.read_text(train_args_path))
|
215 |
-
bl_tokenizer = train_args.data.tokenizer_args.build(); assert isinstance(bl_tokenizer, BltTokenizer)
|
216 |
-
patcher_args = train_args.data.patcher_args.model_copy(deep=True); patcher_args.realtime_patching = True
|
217 |
-
device = "cuda" if torch.cuda.is_available() else "cpu"; print(f"Using BLT device: {device}")
|
218 |
-
patcher_args.patching_device = device; patcher_args.device = device
|
219 |
-
entropy_model_dir = os.path.join(consolidated_path, "entropy_model")
|
220 |
-
if not os.path.exists(entropy_model_dir): raise FileNotFoundError(f"Entropy model directory not found at {entropy_model_dir}.")
|
221 |
-
patcher_args.entropy_model_checkpoint_dir = entropy_model_dir; bl_patcher = patcher_args.build()
|
222 |
-
status_message += "\nBytelatent entropy model loaded."
|
223 |
-
|
224 |
-
# --- Processing ---
|
225 |
-
status_message += "\nRunning Bytelatent entropy model patching..."
|
226 |
-
print(f"Processing prompt with entropy model: '{prompt}'")
|
227 |
-
prompt_bytes = prompt.encode('utf-8')
|
228 |
-
max_bytes = 512 # Define max bytes
|
229 |
-
if len(prompt_bytes) > max_bytes:
|
230 |
-
print(f"Warning: Prompt exceeds {max_bytes} bytes ({len(prompt_bytes)}). Truncating for entropy model.")
|
231 |
-
# Find the byte position that corresponds to the last full character within the limit
|
232 |
-
# This avoids splitting a multi-byte character
|
233 |
-
try:
|
234 |
-
last_char_pos = prompt_bytes[:max_bytes].rfind(b' ') # Simple whitespace split point find, might not be perfect
|
235 |
-
if last_char_pos == -1: # If no space, truncate hard (less ideal)
|
236 |
-
prompt_bl = prompt_bytes[:max_bytes].decode('utf-8', errors='ignore')
|
237 |
-
else:
|
238 |
-
prompt_bl = prompt_bytes[:last_char_pos].decode('utf-8', errors='ignore')
|
239 |
-
|
240 |
-
except Exception: # Fallback to simple truncation on decode errors
|
241 |
-
prompt_bl = prompt_bytes[:max_bytes].decode('utf-8', errors='ignore')
|
242 |
-
|
243 |
-
status_message += f"\nWarning: Prompt truncated to approx {len(prompt_bl.encode('utf-8'))} bytes for Bytelatent entropy model."
|
244 |
-
else:
|
245 |
-
prompt_bl = prompt
|
246 |
-
|
247 |
-
results = patcher_nocache([prompt_bl], tokenizer=bl_tokenizer, patcher=bl_patcher)
|
248 |
|
249 |
-
|
250 |
-
|
251 |
-
|
252 |
-
|
253 |
-
bl_count = 0
|
254 |
-
else:
|
255 |
-
batch_patch_lengths, batch_scores, batch_tokens = results
|
256 |
-
patch_lengths, scores, tokens = batch_patch_lengths[0], batch_scores[0], batch_tokens[0]
|
257 |
-
# --- Visualization Data Generation ---
|
258 |
-
try: decoded_output_for_plot = bl_tokenizer.decode(tokens.tolist())
|
259 |
-
except Exception as decode_err:
|
260 |
-
print(f"Warning: Error decoding full sequence for plot: {decode_err}")
|
261 |
-
decoded_output_for_plot = prompt_bl # Use truncated prompt for plot if decode fails
|
262 |
-
|
263 |
-
fig = plot_entropies(patch_lengths, scores, decoded_output_for_plot, threshold=bl_patcher.threshold)
|
264 |
-
bl_highlighted_data, bl_count_calc = create_bytelatent_highlight_data(bl_tokenizer, patch_lengths, tokens, VIZ_COLORS)
|
265 |
-
bl_count = bl_count_calc # Assign calculated count
|
266 |
-
|
267 |
-
status_message += f"\nBytelatent entropy model processing and visualization successful ({bl_count} patches)."
|
268 |
-
print("Bytelatent Entropy model processing and decoding complete.")
|
269 |
-
|
270 |
-
except FileNotFoundError as e:
|
271 |
-
print(f"Bytelatent Error: {e}")
|
272 |
-
status_message += f"\nBytelatent FileNotFoundError: {str(e)}"
|
273 |
-
bl_highlighted_data = [(f"Bytelatent Error: {e}", "Error")]
|
274 |
-
bl_count = 0
|
275 |
-
except Exception as e:
|
276 |
-
print(f"An unexpected Bytelatent error occurred: {e}")
|
277 |
-
import traceback
|
278 |
-
traceback.print_exc()
|
279 |
-
status_message += f"\nBytelatent Unexpected Error: {str(e)}"
|
280 |
-
bl_highlighted_data = [(f"Bytelatent Error: {e}", "Error")]
|
281 |
-
bl_count = 0
|
282 |
else:
|
283 |
-
|
284 |
-
bl_highlighted_data = [("Bytelatent library not available.", "Error")]
|
285 |
-
bl_count = 0
|
286 |
-
fig = None # Ensure fig is None if BLT is skipped
|
287 |
|
288 |
-
#
|
289 |
-
|
|
|
|
|
|
|
|
|
290 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
291 |
|
292 |
# --- Gradio Interface ---
|
293 |
|
294 |
-
|
295 |
-
|
296 |
-
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
|
301 |
-
|
302 |
-
tiktoken_color_map.
|
|
|
303 |
|
304 |
-
|
305 |
-
|
306 |
-
|
307 |
-
|
308 |
-
with gr.Blocks(theme=gr.themes.Origin()) as iface:
|
309 |
-
gr.Markdown("# BLT's Entropy-based Patcher vs. Tokenizer Visualisation")
|
310 |
-
gr.Markdown(
|
311 |
-
"Enter text to visualize its segmentation according to different methods:\n"
|
312 |
-
"1. **Byte Latent Transformer (BLT):** Entropy-based patching plot and patched text (_for this space ONLY_ - limited to ~512 bytes).\n"
|
313 |
-
"2. **Tiktoken (GPT-4):** Text segmented by `cl100k_base` tokens.\n"
|
314 |
-
f"3. **Llama 3:** Text segmented by the `{LLAMA3_MODEL_NAME}` tokenizer."
|
315 |
-
)
|
316 |
|
317 |
with gr.Row():
|
318 |
with gr.Column(scale=1): # Input Column
|
319 |
prompt_input = gr.Textbox(
|
320 |
label="Input Prompt",
|
321 |
-
value=
|
322 |
placeholder="Enter text here...",
|
323 |
-
|
324 |
lines=5,
|
325 |
-
info="
|
326 |
)
|
327 |
submit_button = gr.Button("Generate Visualizations", variant="primary")
|
328 |
-
status_output = gr.Textbox(label="Processing Status", interactive=False, lines=
|
329 |
|
330 |
with gr.Column(scale=2): # Output Column
|
331 |
-
|
332 |
# --- Bytelatent Output Area ---
|
333 |
-
|
334 |
-
gr.
|
335 |
-
|
336 |
-
|
337 |
-
|
338 |
-
|
339 |
-
|
340 |
-
|
341 |
-
|
342 |
-
|
343 |
-
|
344 |
-
|
345 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
346 |
|
347 |
# --- Tiktoken Output Area ---
|
348 |
-
|
349 |
-
gr.
|
350 |
-
|
351 |
-
|
352 |
-
|
353 |
-
|
354 |
-
|
355 |
-
|
356 |
-
|
357 |
-
|
358 |
-
|
359 |
-
|
|
|
|
|
|
|
|
|
360 |
|
361 |
# --- Llama 3 Output Area ---
|
362 |
-
|
363 |
-
gr.
|
364 |
-
|
365 |
-
|
366 |
-
|
367 |
-
|
368 |
-
|
369 |
-
|
370 |
-
|
371 |
-
|
372 |
-
|
373 |
-
|
|
|
|
|
|
|
|
|
|
|
374 |
|
375 |
# Define the action for the button click
|
376 |
submit_button.click(
|
377 |
fn=process_text,
|
378 |
inputs=prompt_input,
|
379 |
-
# Ensure order matches the
|
380 |
outputs=[
|
381 |
-
|
382 |
-
|
383 |
-
|
384 |
-
|
385 |
-
|
386 |
-
|
387 |
-
|
388 |
-
|
389 |
-
|
|
|
|
|
|
|
|
|
390 |
)
|
391 |
|
392 |
# --- Launch the Gradio App ---
|
393 |
if __name__ == "__main__":
|
394 |
-
|
395 |
-
|
396 |
-
|
397 |
-
|
398 |
-
|
399 |
-
|
400 |
-
try:
|
401 |
-
import transformers
|
402 |
-
import sentencepiece
|
403 |
-
print("- transformers found.")
|
404 |
-
print("- sentencepiece found.")
|
405 |
-
except ImportError:
|
406 |
-
print("WARNING: 'transformers' or 'sentencepiece' not found. Llama 3 visualization will fail. Install with: pip install transformers sentencepiece")
|
407 |
-
|
408 |
-
if BLT_AVAILABLE:
|
409 |
-
print("- Bytelatent libraries found.")
|
410 |
-
# Ensure bytelatent model is present only if library is available
|
411 |
-
try:
|
412 |
-
print("Ensuring Bytelatent model 'blt-1b' weights are present...")
|
413 |
-
ensure_present(["blt-1b"])
|
414 |
-
print("Bytelatent model check complete.")
|
415 |
-
except Exception as blt_dl_err:
|
416 |
-
print(f"WARNING: Failed to ensure Bytelatent model presence: {blt_dl_err}")
|
417 |
-
else:
|
418 |
-
print("INFO: Bytelatent libraries not found, skipping related functionality.")
|
419 |
-
|
420 |
-
print(f"Attempting to use Llama 3 Tokenizer: {LLAMA3_MODEL_NAME}. Ensure you have access (e.g., via `huggingface-cli login` if needed).")
|
421 |
-
print("Launching Gradio interface...")
|
422 |
iface.launch()
|
|
|
3 |
import torch
|
4 |
import itertools # For color cycling
|
5 |
import tiktoken # For GPT-4 tokenizer
|
6 |
+
from transformers import AutoTokenizer, HfArgumentParser # For Llama3 tokenizer & args potentially
|
7 |
+
import traceback # For detailed error logging
|
8 |
+
import logging # For better logging practices
|
9 |
+
from typing import Optional, Tuple, List, Dict, Any
|
10 |
+
import matplotlib.figure # For type hinting
|
11 |
+
import matplotlib.pyplot as plt
|
12 |
+
|
13 |
+
# --- Configuration ---
|
14 |
+
|
15 |
+
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
16 |
+
|
17 |
+
class Config:
|
18 |
+
# Visualization
|
19 |
+
VIZ_COLORS: List[str] = [
|
20 |
+
"#a6cee3", "#1f78b4", "#b2df8a", "#33a02c", "#fb9a99", "#e31a1c",
|
21 |
+
"#fdbf6f", "#ff7f00", "#cab2d6", "#6a3d9a", "#ffff99", "#b15928"
|
22 |
+
]
|
23 |
+
MAX_EXPECTED_SEGMENTS: int = 1 # Max segments for color map generation
|
24 |
+
|
25 |
+
# Model/Tokenizer Names
|
26 |
+
LLAMA3_MODEL_NAME: str = "meta-llama/Meta-Llama-3-8B" # Or choose another variant like Instruct
|
27 |
+
TIKTOKEN_ENCODING_NAME: str = "cl100k_base"
|
28 |
+
BLT_MODEL_NAME: str = "blt-1b" # Default Bytelatent model
|
29 |
+
|
30 |
+
# Bytelatent Specific
|
31 |
+
BLT_WEIGHTS_DIR: str = "hf-weights"
|
32 |
+
BLT_MAX_BYTES_FOR_DEMO: int = 512 # Limit for this specific demo's entropy model
|
33 |
+
|
34 |
+
# Gradio
|
35 |
+
DEFAULT_PROMPT: str = "Daenerys Targaryen is in Game of Thrones, a fantasy epic by George R.R. Martin."
|
36 |
+
GRADIO_THEME = gr.themes.Origin()
|
37 |
+
GRADIO_TITLE: str = "BLT's Entropy-based Patcher vs. Tokenizer Visualisation"
|
38 |
+
GRADIO_DESC: str = (
|
39 |
+
"Enter text to visualize its segmentation according to different methods:\n"
|
40 |
+
f"1. **Byte Latent Transformer (BLT):** Entropy-based patching plot and patched text (_for this space ONLY_ - limited to ~{BLT_MAX_BYTES_FOR_DEMO} bytes using `blt_main_entropy_100m_512w`).\n"
|
41 |
+
f"2. **Tiktoken (GPT-4):** Text segmented by `{TIKTOKEN_ENCODING_NAME}` tokens.\n"
|
42 |
+
f"3. **Llama 3:** Text segmented by the `{LLAMA3_MODEL_NAME}` tokenizer."
|
43 |
+
)
|
44 |
|
45 |
+
# --- Bytelatent Processor ---
|
46 |
+
|
47 |
+
# Attempt to import Bytelatent libraries
|
48 |
try:
|
49 |
from bytelatent.data.file_util import get_fs
|
50 |
from bytelatent.generate_patcher import patcher_nocache
|
51 |
from bytelatent.tokenizers.blt_tokenizer import BltTokenizer
|
52 |
from bytelatent.plotting.entropy_figure_via_matplot_lib import plot_entropies
|
53 |
from bytelatent.args import TrainArgs
|
54 |
+
from download_blt_weights import main as ensure_present # Assuming this downloads weights
|
55 |
+
_BLT_AVAILABLE = True
|
56 |
+
logging.info("Bytelatent libraries found.")
|
57 |
except ImportError as e:
|
58 |
+
logging.warning(f"Bytelatent libraries not found. Bytelatent functionality will be disabled. Error: {e}")
|
59 |
+
_BLT_AVAILABLE = False
|
60 |
# Define dummy classes/functions if BLT is not available to avoid NameErrors later
|
61 |
class BltTokenizer: pass
|
62 |
class TrainArgs: pass
|
63 |
def patcher_nocache(*args, **kwargs): return None
|
64 |
def plot_entropies(*args, **kwargs): return None
|
65 |
def ensure_present(*args, **kwargs): pass
|
66 |
+
matplotlib = None # No plotting if BLT isn't there
|
67 |
+
|
68 |
+
class BytelatentProcessor:
|
69 |
+
"""Handles loading and running the Bytelatent entropy model."""
|
70 |
+
def __init__(self, model_name: str, weights_dir: str):
|
71 |
+
self.model_name = model_name
|
72 |
+
self.weights_dir = weights_dir
|
73 |
+
self.is_available: bool = False
|
74 |
+
self.tokenizer: Optional[BltTokenizer] = None
|
75 |
+
self.patcher: Optional[Any] = None # Type depends on bytelatent implementation
|
76 |
+
self.device: str = "cuda" if torch.cuda.is_available() else "cpu"
|
77 |
+
|
78 |
+
if _BLT_AVAILABLE:
|
79 |
+
try:
|
80 |
+
# 1. Ensure weights are present
|
81 |
+
logging.info(f"Ensuring Bytelatent model '{model_name}' weights are present...")
|
82 |
+
ensure_present([model_name]) # Call the download script
|
83 |
+
logging.info("Bytelatent model check complete.")
|
84 |
+
|
85 |
+
# 2. Load Bytelatent model components
|
86 |
+
consolidated_path = os.path.join(self.weights_dir, model_name)
|
87 |
+
train_args_path = os.path.join(consolidated_path, "params.json")
|
88 |
+
entropy_model_dir = os.path.join(consolidated_path, "entropy_model")
|
89 |
+
|
90 |
+
if not os.path.exists(train_args_path):
|
91 |
+
raise FileNotFoundError(f"BLT training args not found at {train_args_path}.")
|
92 |
+
if not os.path.exists(entropy_model_dir):
|
93 |
+
raise FileNotFoundError(f"BLT Entropy model directory not found at {entropy_model_dir}.")
|
94 |
+
|
95 |
+
fs = get_fs(train_args_path)
|
96 |
+
train_args = TrainArgs.model_validate_json(fs.read_text(train_args_path))
|
97 |
+
|
98 |
+
self.tokenizer = train_args.data.tokenizer_args.build()
|
99 |
+
assert isinstance(self.tokenizer, BltTokenizer), "Failed to build Bytelatent Tokenizer"
|
100 |
+
|
101 |
+
patcher_args = train_args.data.patcher_args.model_copy(deep=True)
|
102 |
+
patcher_args.realtime_patching = True
|
103 |
+
patcher_args.patching_device = self.device
|
104 |
+
patcher_args.device = self.device
|
105 |
+
patcher_args.entropy_model_checkpoint_dir = entropy_model_dir
|
106 |
+
self.patcher = patcher_args.build()
|
107 |
+
|
108 |
+
self.is_available = True
|
109 |
+
logging.info(f"Bytelatent processor for '{model_name}' loaded successfully on device '{self.device}'.")
|
110 |
+
|
111 |
+
except FileNotFoundError as e:
|
112 |
+
logging.error(f"Bytelatent setup failed: Required file/directory not found. {e}")
|
113 |
+
except Exception as e:
|
114 |
+
logging.error(f"An unexpected error occurred during Bytelatent setup: {e}")
|
115 |
+
logging.error(traceback.format_exc())
|
116 |
+
else:
|
117 |
+
logging.warning("Skipping Bytelatent setup as libraries are unavailable.")
|
118 |
+
|
119 |
+
def _create_highlight_data(self, patch_lengths: torch.Tensor, tokens: torch.Tensor) -> Tuple[List[Tuple[str, str]], int]:
|
120 |
+
"""Generates data for gr.HighlightedText based on bytelatent patches."""
|
121 |
+
if not self.is_available or self.tokenizer is None or patch_lengths.numel() == 0:
|
122 |
+
return [("Bytelatent processing failed or produced no patches.", "Error")], 0
|
123 |
+
|
124 |
+
patch_lengths_list = patch_lengths.tolist()
|
125 |
+
all_token_ids = tokens.tolist()
|
126 |
+
highlighted_data = []
|
127 |
+
current_token_index = 0
|
128 |
+
patch_count = 0
|
129 |
+
|
130 |
+
for i, length in enumerate(patch_lengths_list):
|
131 |
+
if length <= 0: continue
|
132 |
+
patch_token_ids = all_token_ids[current_token_index : current_token_index + length]
|
133 |
+
if not patch_token_ids: continue
|
134 |
+
|
135 |
+
try:
|
136 |
+
patch_text = self.tokenizer.decode(patch_token_ids)
|
137 |
+
except Exception as decode_err:
|
138 |
+
logging.warning(f"Bytelatent patch decoding failed: {decode_err}")
|
139 |
+
patch_text = f"[Decode Error: {len(patch_token_ids)} tokens]"
|
140 |
+
|
141 |
+
patch_label = f"BL Patch {i+1}"
|
142 |
+
highlighted_data.append((patch_text, patch_label))
|
143 |
+
patch_count += 1
|
144 |
+
current_token_index += length
|
145 |
+
|
146 |
+
# Handle remainder tokens if any
|
147 |
+
if current_token_index < len(all_token_ids):
|
148 |
+
remaining_tokens = all_token_ids[current_token_index:]
|
149 |
+
try:
|
150 |
+
remaining_text = self.tokenizer.decode(remaining_tokens)
|
151 |
+
label = "BL Remainder"
|
152 |
+
except Exception:
|
153 |
+
remaining_text = f"[Decode Error: {len(remaining_tokens)} remaining tokens]"
|
154 |
+
label = "Error"
|
155 |
+
highlighted_data.append((remaining_text, label))
|
156 |
+
logging.warning(f"Bytelatent token mismatch. Consumed {current_token_index}, total {len(all_token_ids)}. Remainder added.")
|
157 |
+
|
158 |
+
return highlighted_data, patch_count
|
159 |
+
|
160 |
+
def process(self, prompt: str, max_bytes: int) -> Tuple[Optional[matplotlib.figure.Figure], List[Tuple[str, str]], int, str]:
|
161 |
+
"""Processes the prompt using the loaded Bytelatent model."""
|
162 |
+
status = ""
|
163 |
+
if not self.is_available or self.tokenizer is None or self.patcher is None:
|
164 |
+
status = "Bytelatent processor not available."
|
165 |
+
return None, [("Bytelatent not available.", "Error")], 0, status
|
166 |
+
|
167 |
+
# Truncate prompt if necessary for this demo's model
|
168 |
+
prompt_bytes = prompt.encode('utf-8')
|
169 |
+
prompt_bl = prompt
|
170 |
+
if len(prompt_bytes) > max_bytes:
|
171 |
+
try:
|
172 |
+
# Find last full character within limit (simple space split fallback)
|
173 |
+
try:
|
174 |
+
prompt_bl = prompt_bytes[:max_bytes].decode('utf-8', errors='strict')
|
175 |
+
# If successful, find last space to avoid cutting mid-word visually
|
176 |
+
last_space = prompt_bl.rfind(' ')
|
177 |
+
if last_space != -1:
|
178 |
+
prompt_bl = prompt_bl[:last_space]
|
179 |
+
except UnicodeDecodeError:
|
180 |
+
# If strict fails, find last valid byte sequence start before max_bytes
|
181 |
+
i = max_bytes
|
182 |
+
while i > 0:
|
183 |
+
try:
|
184 |
+
prompt_bytes[:i].decode('utf-8', errors='strict')
|
185 |
+
break # Found valid end point
|
186 |
+
except UnicodeDecodeError:
|
187 |
+
i -= 1
|
188 |
+
prompt_bl = prompt_bytes[:i].decode('utf-8', errors='ignore') # Decode ignoring errors now
|
189 |
+
|
190 |
+
|
191 |
+
trunc_len = len(prompt_bl.encode('utf-8'))
|
192 |
+
status = f"Warning: Prompt truncated to {trunc_len} bytes for Bytelatent entropy model.\n"
|
193 |
+
logging.warning(status.strip())
|
194 |
+
except Exception as trunc_err:
|
195 |
+
# Fallback if complex truncation fails
|
196 |
+
prompt_bl = prompt_bytes[:max_bytes].decode('utf-8', errors='ignore')
|
197 |
+
trunc_len = len(prompt_bl.encode('utf-8'))
|
198 |
+
status = f"Warning: Prompt aggressively truncated to ~{trunc_len} bytes due to encoding issue. Error: {trunc_err}\n"
|
199 |
+
logging.warning(status.strip())
|
200 |
+
|
201 |
+
|
202 |
+
# Run Bytelatent patching
|
203 |
+
try:
|
204 |
+
logging.info(f"Running Bytelatent entropy model patching on {len(prompt_bl.encode('utf-8'))} bytes...")
|
205 |
+
results = patcher_nocache([prompt_bl], tokenizer=self.tokenizer, patcher=self.patcher)
|
206 |
+
status += "Bytelatent patching executed.\n"
|
207 |
+
|
208 |
+
if not results:
|
209 |
+
logging.warning("Bytelatent entropy processing returned no results.")
|
210 |
+
status += "Warning: Bytelatent generated no patches."
|
211 |
+
return None, [("No patches generated by Bytelatent.", "Info")], 0, status
|
212 |
+
|
213 |
+
batch_patch_lengths, batch_scores, batch_tokens = results
|
214 |
+
patch_lengths, scores, tokens = batch_patch_lengths[0], batch_scores[0], batch_tokens[0]
|
215 |
+
|
216 |
+
# Create highlighted text data
|
217 |
+
highlighted_data, patch_count = self._create_highlight_data(patch_lengths, tokens)
|
218 |
+
|
219 |
+
# Create plot
|
220 |
+
fig = None
|
221 |
+
if plot_entropies is not None: # Check if plotting function is available
|
222 |
+
try:
|
223 |
+
# Use the potentially truncated prompt_bl for the plot text axis if full decode fails
|
224 |
+
decoded_output_for_plot = self.tokenizer.decode(tokens.tolist())
|
225 |
+
except Exception as decode_err:
|
226 |
+
logging.warning(f"Error decoding full BLT token sequence for plot: {decode_err}. Using (truncated) input prompt for plot axis.")
|
227 |
+
decoded_output_for_plot = prompt_bl
|
228 |
+
|
229 |
+
fig = plot_entropies(patch_lengths, scores, decoded_output_for_plot, threshold=self.patcher.threshold)
|
230 |
+
status += f"Bytelatent plot generated. Found {patch_count} patches.\n"
|
231 |
+
else:
|
232 |
+
status += "Plotting unavailable.\n"
|
233 |
+
|
234 |
+
logging.info(f"Bytelatent processing complete. Patches: {patch_count}")
|
235 |
+
return fig, highlighted_data, patch_count, status.strip()
|
236 |
+
|
237 |
+
except Exception as e:
|
238 |
+
logging.error(f"An error occurred during Bytelatent processing: {e}")
|
239 |
+
logging.error(traceback.format_exc())
|
240 |
+
status += f"Error during Bytelatent processing: {e}"
|
241 |
+
return None, [(f"Bytelatent Error: {e}", "Error")], 0, status.strip()
|
242 |
+
|
243 |
|
244 |
+
# --- Tokenizer Helpers ---
|
245 |
|
246 |
+
def create_tiktoken_highlight_data(prompt: str, encoding: tiktoken.Encoding) -> Tuple[List[Tuple[str, str]], int, str]:
|
247 |
+
"""Generates data for gr.HighlightedText based on tiktoken."""
|
248 |
+
status = "Processing with Tiktoken...\n"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
249 |
try:
|
250 |
+
tiktoken_ids = encoding.encode(prompt)
|
|
|
251 |
highlighted_data = []
|
|
|
252 |
for i, token_id in enumerate(tiktoken_ids):
|
253 |
+
try:
|
254 |
+
token_text = encoding.decode([token_id])
|
255 |
+
except (UnicodeDecodeError, TypeError): # Handle bytes that don't form valid unicode
|
256 |
try:
|
257 |
+
token_bytes = encoding.decode_single_token_bytes(token_id)
|
258 |
token_text = f"[Bytes: {token_bytes.hex()}]"
|
259 |
except Exception: token_text = "[Decode Error]"
|
260 |
except Exception as e:
|
261 |
+
logging.warning(f"Unexpected tiktoken decode error for token ID {token_id}: {e}")
|
262 |
+
token_text = "[Decode Error]"
|
263 |
+
|
264 |
token_label = f"GPT4 Tk {i+1}"
|
265 |
highlighted_data.append((token_text, token_label))
|
266 |
+
|
267 |
token_count = len(tiktoken_ids)
|
268 |
+
status += f"Tiktoken processing successful ({token_count} tokens)."
|
269 |
+
logging.info(f"Tiktoken processing complete. Found {token_count} tokens.")
|
270 |
+
return highlighted_data, token_count, status.strip()
|
271 |
+
|
272 |
+
except Exception as e:
|
273 |
+
logging.error(f"Error during tiktoken processing: {e}")
|
274 |
+
logging.error(traceback.format_exc())
|
275 |
+
status += f"Error during Tiktoken processing: {e}"
|
276 |
+
return [(f"Error processing with tiktoken: {e}", "Error")], 0, status.strip()
|
277 |
|
278 |
|
279 |
+
def create_llama3_highlight_data(prompt: str, tokenizer: AutoTokenizer) -> Tuple[List[Tuple[str, str]], int, str]:
|
280 |
"""Generates data for gr.HighlightedText based on Llama 3 tokenizer."""
|
281 |
+
status = f"Processing with Llama 3 ({tokenizer.name_or_path})...\n"
|
282 |
try:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
283 |
llama_token_ids = tokenizer.encode(prompt)
|
|
|
284 |
highlighted_data = []
|
|
|
|
|
285 |
for i, token_id in enumerate(llama_token_ids):
|
286 |
try:
|
287 |
+
# Decode individual token. Add special handling if needed for specific tokenizers.
|
288 |
token_text = tokenizer.decode([token_id])
|
289 |
except Exception as e:
|
290 |
+
logging.warning(f"Unexpected Llama 3 decode error for token ID {token_id}: {e}")
|
291 |
token_text = "[Decode Error]"
|
292 |
|
293 |
+
token_label = f"Llama3 Tk {i+1}"
|
294 |
highlighted_data.append((token_text, token_label))
|
295 |
|
296 |
token_count = len(llama_token_ids)
|
297 |
+
status += f"Llama 3 processing successful ({token_count} tokens)."
|
298 |
+
logging.info(f"Llama 3 processing complete. Found {token_count} tokens.")
|
299 |
+
return highlighted_data, token_count, status.strip()
|
300 |
+
|
301 |
+
except Exception as e:
|
302 |
+
logging.error(f"Error during Llama 3 processing: {e}")
|
303 |
+
logging.error(traceback.format_exc())
|
304 |
+
status += f"Error during Llama 3 processing: {e}"
|
305 |
+
return [(f"Error processing with Llama 3: {e}", "Error")], 0, status.strip()
|
306 |
+
|
307 |
+
# --- Global Initializations ---
|
308 |
+
|
309 |
+
# Initialize Bytelatent Processor (loads model if available)
|
310 |
+
blt_processor = BytelatentProcessor(Config.BLT_MODEL_NAME, Config.BLT_WEIGHTS_DIR)
|
311 |
+
|
312 |
+
# Load Tiktoken Encoding
|
313 |
+
try:
|
314 |
+
tiktoken_encoding = tiktoken.get_encoding(Config.TIKTOKEN_ENCODING_NAME)
|
315 |
+
logging.info(f"Tiktoken encoding '{Config.TIKTOKEN_ENCODING_NAME}' loaded.")
|
316 |
+
tiktoken_available = True
|
317 |
+
except Exception as e:
|
318 |
+
logging.error(f"Failed to load Tiktoken encoding '{Config.TIKTOKEN_ENCODING_NAME}': {e}")
|
319 |
+
tiktoken_encoding = None
|
320 |
+
tiktoken_available = False
|
321 |
+
|
322 |
+
# Load Llama 3 Tokenizer
|
323 |
+
try:
|
324 |
+
# Use trust_remote_code=True if required by the specific model revision
|
325 |
+
llama_tokenizer = AutoTokenizer.from_pretrained(Config.LLAMA3_MODEL_NAME) #, trust_remote_code=True)
|
326 |
+
logging.info(f"Llama 3 tokenizer '{Config.LLAMA3_MODEL_NAME}' loaded.")
|
327 |
+
llama_available = True
|
328 |
+
except ImportError:
|
329 |
+
logging.error("Transformers or SentencePiece library not found. Llama 3 functionality disabled. Install with: pip install transformers sentencepiece")
|
330 |
+
llama_tokenizer = None
|
331 |
+
llama_available = False
|
332 |
+
except OSError as e:
|
333 |
+
logging.error(f"Error loading Llama 3 tokenizer '{Config.LLAMA3_MODEL_NAME}': {e}")
|
334 |
+
error_msg = f"Could not load Llama 3 tokenizer '{Config.LLAMA3_MODEL_NAME}'. Check model name, network, and authentication (use `huggingface-cli login` if needed)."
|
335 |
+
logging.error(error_msg)
|
336 |
+
llama_tokenizer = None
|
337 |
+
llama_available = False
|
338 |
+
except Exception as e:
|
339 |
+
logging.error(f"An unexpected error occurred loading Llama 3 tokenizer: {e}")
|
340 |
+
logging.error(traceback.format_exc())
|
341 |
+
llama_tokenizer = None
|
342 |
+
llama_available = False
|
343 |
|
344 |
|
345 |
# --- Main Processing Function ---
|
346 |
|
347 |
+
def process_text(prompt: str) -> Tuple[
|
348 |
+
Optional[matplotlib.figure.Figure], List[Tuple[str, str]], int, # BLT
|
349 |
+
List[Tuple[str, str]], int, # Tiktoken
|
350 |
+
List[Tuple[str, str]], int, # Llama 3
|
351 |
+
str # Status
|
352 |
+
]:
|
353 |
"""
|
354 |
Processes the input prompt using ByteLatent, Tiktoken, and Llama 3,
|
355 |
returning visualizations, counts, and status.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
356 |
"""
|
357 |
+
status_messages = ["Processing started..."]
|
358 |
fig = None
|
359 |
+
bl_highlighted_data, bl_count = [("Bytelatent not available.", "Error")], 0
|
360 |
+
tk_highlighted_data, tk_count = [("Tiktoken not available.", "Error")], 0
|
361 |
+
llama_highlighted_data, llama_count = [("Llama 3 not available.", "Error")], 0
|
362 |
+
|
363 |
+
# 1. Bytelatent Processing
|
364 |
+
if blt_processor.is_available:
|
365 |
+
fig, bl_highlighted_data, bl_count, bl_status = blt_processor.process(prompt, Config.BLT_MAX_BYTES_FOR_DEMO)
|
366 |
+
status_messages.append(f"Bytelatent Status:\n{bl_status}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
367 |
else:
|
368 |
+
status_messages.append("Bytelatent Status: Skipped (processor unavailable).")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
369 |
|
370 |
+
# 2. Tiktoken Processing
|
371 |
+
if tiktoken_available and tiktoken_encoding:
|
372 |
+
tk_highlighted_data, tk_count, tk_status = create_tiktoken_highlight_data(prompt, tiktoken_encoding)
|
373 |
+
status_messages.append(f"Tiktoken Status:\n{tk_status}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
374 |
else:
|
375 |
+
status_messages.append("Tiktoken Status: Skipped (tokenizer unavailable).")
|
|
|
|
|
|
|
376 |
|
377 |
+
# 3. Llama 3 Processing
|
378 |
+
if llama_available and llama_tokenizer:
|
379 |
+
llama_highlighted_data, llama_count, llama_status = create_llama3_highlight_data(prompt, llama_tokenizer)
|
380 |
+
status_messages.append(f"Llama 3 Status:\n{llama_status}")
|
381 |
+
else:
|
382 |
+
status_messages.append("Llama 3 Status: Skipped (tokenizer unavailable).")
|
383 |
|
384 |
+
final_status = "\n---\n".join(status_messages)
|
385 |
+
if fig is not None and matplotlib is not None:
|
386 |
+
try:
|
387 |
+
plt.close(fig) # Close the specific figure
|
388 |
+
logging.debug("Closed Matplotlib figure.")
|
389 |
+
except Exception as close_err:
|
390 |
+
logging.warning(f"Could not close Matplotlib figure: {close_err}")
|
391 |
+
return fig, bl_highlighted_data, bl_count, tk_highlighted_data, tk_count, llama_highlighted_data, llama_count, final_status
|
392 |
|
393 |
# --- Gradio Interface ---
|
394 |
|
395 |
+
def create_color_map(label_prefix: str, colors: List[str], max_segments: int) -> Dict[str, str]:
|
396 |
+
"""Generates a color map dictionary for Gradio HighlightedText."""
|
397 |
+
color_cycler = itertools.cycle(colors)
|
398 |
+
color_map = {f"{label_prefix} {i+1}": next(color_cycler) for i in range(max_segments)}
|
399 |
+
color_map.update({"Error": "#FF0000", "Info": "#808080", "BL Remainder": "#AAAAAA"}) # Common labels
|
400 |
+
return color_map
|
401 |
|
402 |
+
bytelatent_color_map = create_color_map("BL Patch", Config.VIZ_COLORS, Config.MAX_EXPECTED_SEGMENTS)
|
403 |
+
tiktoken_color_map = create_color_map("GPT4 Tk", Config.VIZ_COLORS, Config.MAX_EXPECTED_SEGMENTS)
|
404 |
+
llama3_color_map = create_color_map("Llama3 Tk", Config.VIZ_COLORS, Config.MAX_EXPECTED_SEGMENTS)
|
405 |
|
406 |
+
with gr.Blocks(theme=Config.GRADIO_THEME) as iface:
|
407 |
+
gr.Markdown(f"# {Config.GRADIO_TITLE}")
|
408 |
+
gr.Markdown(Config.GRADIO_DESC)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
409 |
|
410 |
with gr.Row():
|
411 |
with gr.Column(scale=1): # Input Column
|
412 |
prompt_input = gr.Textbox(
|
413 |
label="Input Prompt",
|
414 |
+
value=Config.DEFAULT_PROMPT,
|
415 |
placeholder="Enter text here...",
|
416 |
+
# Max length is for UI input; Bytelatent truncation happens in backend
|
417 |
lines=5,
|
418 |
+
info=f"Note: Bytelatent processing is limited to ~{Config.BLT_MAX_BYTES_FOR_DEMO} bytes for this demo."
|
419 |
)
|
420 |
submit_button = gr.Button("Generate Visualizations", variant="primary")
|
421 |
+
status_output = gr.Textbox(label="Processing Status", interactive=False, lines=10) # More space for detailed status
|
422 |
|
423 |
with gr.Column(scale=2): # Output Column
|
|
|
424 |
# --- Bytelatent Output Area ---
|
425 |
+
if blt_processor.is_available: # Only show BLT section if it loaded
|
426 |
+
with gr.Accordion("BLT Entropy Patcher Output (`blt_main_entropy_100m_512w`)", open=True):
|
427 |
+
with gr.Row():
|
428 |
+
bl_count_output = gr.Number(label="Patch Count", value=0, interactive=False, step=1, scale=0)
|
429 |
+
highlighted_output_bl = gr.HighlightedText(
|
430 |
+
label="BLT Patches",
|
431 |
+
color_map=bytelatent_color_map,
|
432 |
+
show_legend=False,
|
433 |
+
show_inline_category=False,
|
434 |
+
container=False
|
435 |
+
)
|
436 |
+
plot_output = gr.Plot(label="Entropy vs. Token Index")
|
437 |
+
else:
|
438 |
+
gr.Markdown(f"### Bytelatent Output (`{Config.BLT_MODEL_NAME}`)")
|
439 |
+
gr.Markdown("_(Bytelatent processor failed to load or libraries are missing. Output unavailable.)_")
|
440 |
+
# Define dummy outputs if BLT is unavailable so the `outputs` list doesn't break
|
441 |
+
highlighted_output_bl = gr.HighlightedText(value=[("BLT Unavailable", "Error")], label="BLT Patches", visible=False)
|
442 |
+
bl_count_output = gr.Number(value=0, label="Patch Count", visible=False)
|
443 |
+
plot_output = gr.Plot(label="Entropy Plot", visible=False)
|
444 |
+
|
445 |
|
446 |
# --- Tiktoken Output Area ---
|
447 |
+
if tiktoken_available: # Only show Tiktoken section if it loaded
|
448 |
+
with gr.Accordion(f"Tiktoken Output (`{Config.TIKTOKEN_ENCODING_NAME}`)", open=True):
|
449 |
+
with gr.Row():
|
450 |
+
tk_count_output = gr.Number(label="Token Count", value=0, interactive=False, step=1, scale=0)
|
451 |
+
highlighted_output_tk = gr.HighlightedText(
|
452 |
+
label="Tiktoken Segments",
|
453 |
+
color_map=tiktoken_color_map,
|
454 |
+
show_legend=False,
|
455 |
+
show_inline_category=False,
|
456 |
+
container=False
|
457 |
+
)
|
458 |
+
else:
|
459 |
+
gr.Markdown(f"### Tiktoken Output (`{Config.TIKTOKEN_ENCODING_NAME}`)")
|
460 |
+
gr.Markdown("_(Tiktoken failed to load. Output unavailable.)_")
|
461 |
+
highlighted_output_tk = gr.HighlightedText(value=[("Tiktoken Unavailable", "Error")], label="Tiktoken Segments", visible=False)
|
462 |
+
tk_count_output = gr.Number(value=0, label="Token Count", visible=False)
|
463 |
|
464 |
# --- Llama 3 Output Area ---
|
465 |
+
if llama_available: # Only show Llama section if it loaded
|
466 |
+
with gr.Accordion(f"Llama 3 Output (`{Config.LLAMA3_MODEL_NAME}`)", open=True):
|
467 |
+
with gr.Row():
|
468 |
+
llama_count_output = gr.Number(label="Token Count", value=0, interactive=False, step=1, scale=0)
|
469 |
+
highlighted_output_llama = gr.HighlightedText(
|
470 |
+
label="Llama 3 Segments",
|
471 |
+
color_map=llama3_color_map,
|
472 |
+
show_legend=False,
|
473 |
+
show_inline_category=False,
|
474 |
+
container=False
|
475 |
+
)
|
476 |
+
else:
|
477 |
+
gr.Markdown(f"### Llama 3 Output (`{Config.LLAMA3_MODEL_NAME}`)")
|
478 |
+
gr.Markdown("_(Llama 3 tokenizer failed to load. Output unavailable.)_")
|
479 |
+
highlighted_output_llama = gr.HighlightedText(value=[("Llama 3 Unavailable", "Error")], label="Llama 3 Segments", visible=False)
|
480 |
+
llama_count_output = gr.Number(value=0, label="Token Count", visible=False)
|
481 |
+
|
482 |
|
483 |
# Define the action for the button click
|
484 |
submit_button.click(
|
485 |
fn=process_text,
|
486 |
inputs=prompt_input,
|
487 |
+
# Ensure order matches the return values of process_text
|
488 |
outputs=[
|
489 |
+
# Bytelatent outputs (even if dummy/hidden)
|
490 |
+
plot_output,
|
491 |
+
highlighted_output_bl,
|
492 |
+
bl_count_output,
|
493 |
+
# Tiktoken outputs (even if dummy/hidden)
|
494 |
+
highlighted_output_tk,
|
495 |
+
tk_count_output,
|
496 |
+
# Llama 3 outputs (even if dummy/hidden)
|
497 |
+
highlighted_output_llama,
|
498 |
+
llama_count_output,
|
499 |
+
# Status output
|
500 |
+
status_output
|
501 |
+
]
|
502 |
)
|
503 |
|
504 |
# --- Launch the Gradio App ---
|
505 |
if __name__ == "__main__":
|
506 |
+
logging.info("-----------------------------------------")
|
507 |
+
logging.info("Starting Gradio App...")
|
508 |
+
logging.info(f"Bytelatent Available: {blt_processor.is_available}")
|
509 |
+
logging.info(f"Tiktoken Available: {tiktoken_available}")
|
510 |
+
logging.info(f"Llama 3 Tokenizer Available: {llama_available}")
|
511 |
+
logging.info("-----------------------------------------")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
512 |
iface.launch()
|