lhoestq HF staff commited on
Commit
fbe940a
1 Parent(s): 6b97460

run on examples click

Browse files
Files changed (2) hide show
  1. generate.py +5 -1
  2. gradio_app.py +38 -19
generate.py CHANGED
@@ -1,6 +1,7 @@
1
 
2
  import json
3
  import logging
 
4
  import time
5
  from pathlib import Path
6
  from typing import Annotated, Iterator
@@ -33,8 +34,11 @@ else:
33
  tokenizer = AutoTokenizer.from_pretrained(model_id)
34
  sampler = PenalizedMultinomialSampler()
35
  low_temperature_sampler = PenalizedMultinomialSampler(temperature=0.3)
36
- empty_tokens = [token_id for token_id in range(tokenizer.vocab_size) if not tokenizer.decode([token_id]).strip()]
37
  sampler.set_max_repeats(empty_tokens, 1)
 
 
 
38
 
39
  # This Sample & Dataset models ztr just templated with placeholder fields
40
 
 
1
 
2
  import json
3
  import logging
4
+ import regex
5
  import time
6
  from pathlib import Path
7
  from typing import Annotated, Iterator
 
34
  tokenizer = AutoTokenizer.from_pretrained(model_id)
35
  sampler = PenalizedMultinomialSampler()
36
  low_temperature_sampler = PenalizedMultinomialSampler(temperature=0.3)
37
+ empty_tokens = [token_id for token_id in range(tokenizer.vocab_size) if not tokenizer.decode([token_id], skip_special_tokens=True).strip()]
38
  sampler.set_max_repeats(empty_tokens, 1)
39
+ disallowed_patterns = [regex.compile(r"\p{Han}")] # focus on english for now
40
+ disallowed_tokens = [token_id for token_id in range(tokenizer.vocab_size) if any(pattern.match(tokenizer.decode([token_id], skip_special_tokens=True)) for pattern in disallowed_patterns)]
41
+ sampler.set_max_repeats(disallowed_tokens, 0)
42
 
43
  # This Sample & Dataset models ztr just templated with placeholder fields
44
 
gradio_app.py CHANGED
@@ -1,4 +1,4 @@
1
- import time
2
  from urllib.parse import urlparse, parse_qs
3
 
4
  import gradio as gr
@@ -13,8 +13,9 @@ DEFAULT_SEED = 42
13
  DEFAULT_SIZE = 3
14
 
15
  @spaces.GPU(duration=120)
16
- def stream_output(filename: str):
17
- parsed_filename = urlparse(filename)
 
18
  filename = parsed_filename.path
19
  params = parse_qs(parsed_filename.query)
20
  prompt = params["prompt"][0] if "prompt" in params else ""
@@ -22,30 +23,44 @@ def stream_output(filename: str):
22
  size = int(params["size"][0]) if "size" in params else DEFAULT_SIZE
23
  seed = int(params["seed"][0]) if "seed" in params else DEFAULT_SEED
24
  if size > MAX_SIZE:
25
- yield None, None, "Error: Maximum size is 20"
26
- content = ""
27
- start_time = time.time()
 
 
 
 
 
 
 
28
  for i, chunk in enumerate(stream_jsonl_file(
29
  filename=filename,
30
  prompt=prompt,
31
  columns=columns,
32
- seed=seed,
33
  size=size,
34
  )):
35
  content += chunk
36
  df = pd.read_json(io.StringIO(content), lines=True)
37
- state_msg = (
38
- f" Done generating {size} samples in {time.time() - start_time:.2f}s"
39
- if i + 1 == size else
40
- f"⚙️ Generating... [{i + 1}/{size}]"
41
- )
42
- yield df, "```json\n" + content + "\n```", state_msg
 
 
 
 
 
 
 
43
 
44
  title = "LLM DataGen"
45
  description = "Generate and stream synthetic dataset files in JSON Lines format"
46
  examples = [
47
  "movies_data.jsonl",
48
- "dungeon_and_dragon_characters.jsonl"
49
  "bad_amazon_reviews_on_defunct_products_that_people_hate.jsonl",
50
  "common_first_names.jsonl?columns=first_name,popularity&size=10",
51
  ]
@@ -53,16 +68,20 @@ examples = [
53
  with gr.Blocks() as demo:
54
  gr.Markdown(f"# {title}")
55
  gr.Markdown(description)
56
- filename_comp = gr.Textbox(examples[0], placeholder=examples[0])
57
- gr.Examples(examples, filename_comp)
58
  generate_button = gr.Button("Generate dataset")
59
- state_msg_comp = gr.Markdown("🔥 Ready to generate")
60
  with gr.Tab("Dataset"):
61
  dataframe_comp = gr.DataFrame()
62
  with gr.Tab("File content"):
63
  file_content_comp = gr.Markdown()
64
-
65
- generate_button.click(stream_output, filename_comp, [dataframe_comp, file_content_comp, state_msg_comp])
 
 
 
 
 
66
 
67
 
68
  demo.launch()
 
1
+ from pathlib import Path
2
  from urllib.parse import urlparse, parse_qs
3
 
4
  import gradio as gr
 
13
  DEFAULT_SIZE = 3
14
 
15
  @spaces.GPU(duration=120)
16
+ def stream_output(query: str, continue_content: str = ""):
17
+ query = Path(query).name
18
+ parsed_filename = urlparse(query)
19
  filename = parsed_filename.path
20
  params = parse_qs(parsed_filename.query)
21
  prompt = params["prompt"][0] if "prompt" in params else ""
 
23
  size = int(params["size"][0]) if "size" in params else DEFAULT_SIZE
24
  seed = int(params["seed"][0]) if "seed" in params else DEFAULT_SEED
25
  if size > MAX_SIZE:
26
+ raise gr.Error(f"Maximum size is {MAX_SIZE}. Duplicate this Space to remove this limit.")
27
+ content = continue_content
28
+ df = pd.read_json(io.StringIO(content), lines=True)
29
+ continue_content_size = len(df)
30
+ state_msg = f"⚙️ Generating... [{continue_content_size + 1}/{continue_content_size + size}]"
31
+ if list(df.columns):
32
+ columns = list(df.columns)
33
+ else:
34
+ df = pd.DataFrame({"1": [], "2": [], "3": []})
35
+ yield df, "```json\n" + content + "\n```", gr.Button(state_msg), gr.Button("Generate one more batch", interactive=False), gr.DownloadButton("⬇️ Download", interactive=False)
36
  for i, chunk in enumerate(stream_jsonl_file(
37
  filename=filename,
38
  prompt=prompt,
39
  columns=columns,
40
+ seed=seed + (continue_content_size // size),
41
  size=size,
42
  )):
43
  content += chunk
44
  df = pd.read_json(io.StringIO(content), lines=True)
45
+ state_msg = f"⚙️ Generating... [{continue_content_size + i + 1}/{continue_content_size + size}]"
46
+ yield df, "```json\n" + content + "\n```", gr.Button(state_msg), gr.Button("Generate one more batch", interactive=False), gr.DownloadButton("⬇️ Download", interactive=False)
47
+ with open(query, "w", encoding="utf-8") as f:
48
+ f.write(content)
49
+ yield df, "```json\n" + content + "\n```", gr.Button("Generate dataset"), gr.Button("Generate one more batch", visible=True, interactive=True), gr.DownloadButton("⬇️ Download", value=query, visible=True, interactive=True)
50
+
51
+
52
+ def stream_more_output(query: str):
53
+ query = Path(query).name
54
+ with open(query, "r", encoding="utf-8") as f:
55
+ continue_content = f.read()
56
+ yield from stream_output(query=query, continue_content=continue_content)
57
+
58
 
59
  title = "LLM DataGen"
60
  description = "Generate and stream synthetic dataset files in JSON Lines format"
61
  examples = [
62
  "movies_data.jsonl",
63
+ "dungeon_and_dragon_characters.jsonl",
64
  "bad_amazon_reviews_on_defunct_products_that_people_hate.jsonl",
65
  "common_first_names.jsonl?columns=first_name,popularity&size=10",
66
  ]
 
68
  with gr.Blocks() as demo:
69
  gr.Markdown(f"# {title}")
70
  gr.Markdown(description)
71
+ filename_comp = gr.Textbox(examples[0], placeholder=examples[0], label="File name to generate")
72
+ outputs = []
73
  generate_button = gr.Button("Generate dataset")
 
74
  with gr.Tab("Dataset"):
75
  dataframe_comp = gr.DataFrame()
76
  with gr.Tab("File content"):
77
  file_content_comp = gr.Markdown()
78
+ with gr.Row():
79
+ generate_more_button = gr.Button("Generate one more batch", visible=False, interactive=False, scale=3)
80
+ download_button = gr.DownloadButton("⬇️ Download", visible=False, interactive=False, scale=1)
81
+ outputs = [dataframe_comp, file_content_comp, generate_button, generate_more_button, download_button]
82
+ examples = gr.Examples(examples, filename_comp, outputs, fn=stream_output, run_on_click=True)
83
+ generate_button.click(stream_output, filename_comp, outputs)
84
+ generate_more_button.click(stream_more_output, filename_comp, outputs)
85
 
86
 
87
  demo.launch()