Spaces:
Runtime error
Runtime error
Commit
Β·
8571d5a
1
Parent(s):
2eb6d1a
feat: small quality updates
Browse files
src/distilabel_dataset_generator/sft.py
CHANGED
|
@@ -8,14 +8,7 @@ from distilabel.llms import InferenceEndpointsLLM
|
|
| 8 |
from distilabel.pipeline import Pipeline
|
| 9 |
from distilabel.steps import KeepColumns
|
| 10 |
from distilabel.steps.tasks import MagpieGenerator, TextGeneration
|
| 11 |
-
|
| 12 |
-
from src.distilabel_dataset_generator.utils import (
|
| 13 |
-
OAuthToken,
|
| 14 |
-
get_duplicate_button,
|
| 15 |
-
get_login_button,
|
| 16 |
-
get_org_dropdown,
|
| 17 |
-
swap_visibilty,
|
| 18 |
-
)
|
| 19 |
|
| 20 |
INFORMATION_SEEKING_PROMPT = (
|
| 21 |
"You are an AI assistant designed to provide accurate and concise information on a wide"
|
|
@@ -180,7 +173,7 @@ def _run_pipeline(result_queue, num_turns, num_rows, system_prompt, token: str =
|
|
| 180 |
result_queue.put(distiset)
|
| 181 |
|
| 182 |
|
| 183 |
-
def generate_system_prompt(dataset_description,
|
| 184 |
progress(0.1, desc="Initializing text generation")
|
| 185 |
generate_description = TextGeneration(
|
| 186 |
llm=InferenceEndpointsLLM(
|
|
@@ -230,10 +223,13 @@ def generate_dataset(
|
|
| 230 |
if repo_id is not None:
|
| 231 |
if not repo_id:
|
| 232 |
raise gr.Error("Please provide a dataset name to push the dataset to.")
|
| 233 |
-
|
|
|
|
|
|
|
| 234 |
raise gr.Error(
|
| 235 |
-
"
|
| 236 |
)
|
|
|
|
| 237 |
if num_turns > 4:
|
| 238 |
raise gr.Info(
|
| 239 |
"You can only generate a dataset with 4 or fewer turns. Setting to 4."
|
|
@@ -263,20 +259,22 @@ def generate_dataset(
|
|
| 263 |
target=_run_pipeline,
|
| 264 |
args=(result_queue, num_turns, num_rows, system_prompt),
|
| 265 |
)
|
| 266 |
-
|
| 267 |
try:
|
| 268 |
p.start()
|
| 269 |
total_steps = 100
|
| 270 |
for step in range(total_steps):
|
| 271 |
if not p.is_alive():
|
| 272 |
break
|
| 273 |
-
progress(
|
|
|
|
|
|
|
|
|
|
| 274 |
time.sleep(0.5) # Adjust this value based on your needs
|
| 275 |
p.join()
|
| 276 |
except Exception as e:
|
| 277 |
raise gr.Error(f"An error occurred during dataset generation: {str(e)}")
|
| 278 |
-
|
| 279 |
-
|
| 280 |
distiset = result_queue.get()
|
| 281 |
|
| 282 |
if repo_id is not None:
|
|
@@ -290,20 +288,13 @@ def generate_dataset(
|
|
| 290 |
gr.Info(
|
| 291 |
f'Dataset pushed to Hugging Face Hub: <a href="https://huggingface.co/datasets/{repo_id}">https://huggingface.co/datasets/{repo_id}</a>'
|
| 292 |
)
|
| 293 |
-
|
| 294 |
# If not pushing to hub generate the dataset directly
|
| 295 |
distiset = distiset["default"]["train"]
|
| 296 |
if num_turns == 1:
|
| 297 |
outputs = distiset.to_pandas()[["prompt", "completion"]]
|
| 298 |
else:
|
| 299 |
outputs = distiset.to_pandas()[["messages"]]
|
| 300 |
-
# outputs = {"conversation_id": [], "role": [], "content": []}
|
| 301 |
-
# conversations = distiset["messages"]
|
| 302 |
-
# for idx, entry in enumerate(conversations):
|
| 303 |
-
# for message in entry["messages"]:
|
| 304 |
-
# outputs["conversation_id"].append(idx + 1)
|
| 305 |
-
# outputs["role"].append(message["role"])
|
| 306 |
-
# outputs["content"].append(message["content"])
|
| 307 |
|
| 308 |
progress(1.0, desc="Dataset generation completed")
|
| 309 |
return pd.DataFrame(outputs)
|
|
@@ -320,9 +311,7 @@ with gr.Blocks(
|
|
| 320 |
)
|
| 321 |
with gr.Row():
|
| 322 |
gr.Column(scale=1)
|
| 323 |
-
btn_generate_system_prompt = gr.Button(
|
| 324 |
-
value="Generate sample dataset"
|
| 325 |
-
)
|
| 326 |
gr.Column(scale=1)
|
| 327 |
|
| 328 |
system_prompt = gr.TextArea(
|
|
@@ -337,12 +326,12 @@ with gr.Blocks(
|
|
| 337 |
)
|
| 338 |
gr.Column(scale=1)
|
| 339 |
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
|
| 343 |
-
|
| 344 |
-
|
| 345 |
-
|
| 346 |
|
| 347 |
result = btn_generate_system_prompt.click(
|
| 348 |
fn=generate_system_prompt,
|
|
@@ -362,17 +351,21 @@ with gr.Blocks(
|
|
| 362 |
outputs=[table],
|
| 363 |
show_progress=True,
|
| 364 |
)
|
| 365 |
-
|
| 366 |
-
|
| 367 |
gr.Markdown("## Generate full dataset")
|
| 368 |
-
gr.Markdown(
|
|
|
|
|
|
|
| 369 |
with gr.Column() as push_to_hub_ui:
|
| 370 |
with gr.Row(variant="panel"):
|
| 371 |
num_turns = gr.Number(
|
| 372 |
value=1,
|
| 373 |
label="Number of turns in the conversation",
|
|
|
|
| 374 |
maximum=4,
|
| 375 |
-
|
|
|
|
| 376 |
)
|
| 377 |
num_rows = gr.Number(
|
| 378 |
value=100,
|
|
@@ -381,10 +374,9 @@ with gr.Blocks(
|
|
| 381 |
maximum=5000,
|
| 382 |
info="The number of rows in the dataset. Note that you are able to generate more rows at once but that this will take time.",
|
| 383 |
)
|
| 384 |
-
|
| 385 |
|
| 386 |
with gr.Row(variant="panel"):
|
| 387 |
-
hf_token = gr.Textbox(label="HF token")
|
| 388 |
repo_id = gr.Textbox(label="HF repo ID", placeholder="owner/dataset_name")
|
| 389 |
private = gr.Checkbox(label="Private dataset", value=True, interactive=True)
|
| 390 |
|
|
@@ -394,13 +386,7 @@ with gr.Blocks(
|
|
| 394 |
|
| 395 |
btn_generate_full_dataset.click(
|
| 396 |
fn=generate_dataset,
|
| 397 |
-
inputs=[
|
| 398 |
-
system_prompt,
|
| 399 |
-
num_turns,
|
| 400 |
-
num_rows,
|
| 401 |
-
private,
|
| 402 |
-
repo_id,
|
| 403 |
-
],
|
| 404 |
outputs=[table],
|
| 405 |
show_progress=True,
|
| 406 |
)
|
|
|
|
| 8 |
from distilabel.pipeline import Pipeline
|
| 9 |
from distilabel.steps import KeepColumns
|
| 10 |
from distilabel.steps.tasks import MagpieGenerator, TextGeneration
|
| 11 |
+
from huggingface_hub import whoami
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 12 |
|
| 13 |
INFORMATION_SEEKING_PROMPT = (
|
| 14 |
"You are an AI assistant designed to provide accurate and concise information on a wide"
|
|
|
|
| 173 |
result_queue.put(distiset)
|
| 174 |
|
| 175 |
|
| 176 |
+
def generate_system_prompt(dataset_description, progress=gr.Progress()):
|
| 177 |
progress(0.1, desc="Initializing text generation")
|
| 178 |
generate_description = TextGeneration(
|
| 179 |
llm=InferenceEndpointsLLM(
|
|
|
|
| 223 |
if repo_id is not None:
|
| 224 |
if not repo_id:
|
| 225 |
raise gr.Error("Please provide a dataset name to push the dataset to.")
|
| 226 |
+
try:
|
| 227 |
+
whoami(token=token)
|
| 228 |
+
except Exception:
|
| 229 |
raise gr.Error(
|
| 230 |
+
"Provide a Hugging Face to be able to push the dataset to the Hub."
|
| 231 |
)
|
| 232 |
+
|
| 233 |
if num_turns > 4:
|
| 234 |
raise gr.Info(
|
| 235 |
"You can only generate a dataset with 4 or fewer turns. Setting to 4."
|
|
|
|
| 259 |
target=_run_pipeline,
|
| 260 |
args=(result_queue, num_turns, num_rows, system_prompt),
|
| 261 |
)
|
| 262 |
+
|
| 263 |
try:
|
| 264 |
p.start()
|
| 265 |
total_steps = 100
|
| 266 |
for step in range(total_steps):
|
| 267 |
if not p.is_alive():
|
| 268 |
break
|
| 269 |
+
progress(
|
| 270 |
+
(step + 1) / total_steps,
|
| 271 |
+
desc=f"Generating dataset with {num_rows} rows",
|
| 272 |
+
)
|
| 273 |
time.sleep(0.5) # Adjust this value based on your needs
|
| 274 |
p.join()
|
| 275 |
except Exception as e:
|
| 276 |
raise gr.Error(f"An error occurred during dataset generation: {str(e)}")
|
| 277 |
+
|
|
|
|
| 278 |
distiset = result_queue.get()
|
| 279 |
|
| 280 |
if repo_id is not None:
|
|
|
|
| 288 |
gr.Info(
|
| 289 |
f'Dataset pushed to Hugging Face Hub: <a href="https://huggingface.co/datasets/{repo_id}">https://huggingface.co/datasets/{repo_id}</a>'
|
| 290 |
)
|
| 291 |
+
|
| 292 |
# If not pushing to hub generate the dataset directly
|
| 293 |
distiset = distiset["default"]["train"]
|
| 294 |
if num_turns == 1:
|
| 295 |
outputs = distiset.to_pandas()[["prompt", "completion"]]
|
| 296 |
else:
|
| 297 |
outputs = distiset.to_pandas()[["messages"]]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 298 |
|
| 299 |
progress(1.0, desc="Dataset generation completed")
|
| 300 |
return pd.DataFrame(outputs)
|
|
|
|
| 311 |
)
|
| 312 |
with gr.Row():
|
| 313 |
gr.Column(scale=1)
|
| 314 |
+
btn_generate_system_prompt = gr.Button(value="Generate sample dataset")
|
|
|
|
|
|
|
| 315 |
gr.Column(scale=1)
|
| 316 |
|
| 317 |
system_prompt = gr.TextArea(
|
|
|
|
| 326 |
)
|
| 327 |
gr.Column(scale=1)
|
| 328 |
|
| 329 |
+
with gr.Row():
|
| 330 |
+
table = gr.DataFrame(
|
| 331 |
+
value=DEFAULT_DATASET,
|
| 332 |
+
interactive=False,
|
| 333 |
+
wrap=True,
|
| 334 |
+
)
|
| 335 |
|
| 336 |
result = btn_generate_system_prompt.click(
|
| 337 |
fn=generate_system_prompt,
|
|
|
|
| 351 |
outputs=[table],
|
| 352 |
show_progress=True,
|
| 353 |
)
|
| 354 |
+
|
| 355 |
+
# Add a header for the full dataset generation section
|
| 356 |
gr.Markdown("## Generate full dataset")
|
| 357 |
+
gr.Markdown(
|
| 358 |
+
"Once you're satisfied with the sample, generate a larger dataset and push it to the hub. Get <a href='https://huggingface.co/settings/tokens' target='_blank'>a Hugging Face token</a> with write access to the organization you want to push the dataset to."
|
| 359 |
+
)
|
| 360 |
with gr.Column() as push_to_hub_ui:
|
| 361 |
with gr.Row(variant="panel"):
|
| 362 |
num_turns = gr.Number(
|
| 363 |
value=1,
|
| 364 |
label="Number of turns in the conversation",
|
| 365 |
+
minimum=1,
|
| 366 |
maximum=4,
|
| 367 |
+
step=1,
|
| 368 |
+
info="Choose between 1 (single turn with 'instruction-response' columns) and 2-4 (multi-turn conversation with a 'conversation' column).",
|
| 369 |
)
|
| 370 |
num_rows = gr.Number(
|
| 371 |
value=100,
|
|
|
|
| 374 |
maximum=5000,
|
| 375 |
info="The number of rows in the dataset. Note that you are able to generate more rows at once but that this will take time.",
|
| 376 |
)
|
|
|
|
| 377 |
|
| 378 |
with gr.Row(variant="panel"):
|
| 379 |
+
hf_token = gr.Textbox(label="HF token", type="password")
|
| 380 |
repo_id = gr.Textbox(label="HF repo ID", placeholder="owner/dataset_name")
|
| 381 |
private = gr.Checkbox(label="Private dataset", value=True, interactive=True)
|
| 382 |
|
|
|
|
| 386 |
|
| 387 |
btn_generate_full_dataset.click(
|
| 388 |
fn=generate_dataset,
|
| 389 |
+
inputs=[system_prompt, num_turns, num_rows, private, repo_id, hf_token],
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 390 |
outputs=[table],
|
| 391 |
show_progress=True,
|
| 392 |
)
|