Commit
Β·
d034c0d
1
Parent(s):
52dc8a2
Add plain text prompt support and sample limiting to generate-responses.py
Browse files- Add --prompt-column option to accept plain text prompts, automatically converting to chat format
- Add --max-samples option to limit dataset processing for testing and development
- Update README.md with examples showing both chat message and plain text prompt usage
- Enhance dataset card generation to reflect input column type (chat vs plain text)
- Improve validation logic to handle both input modes
π€ Generated with [Claude Code](https://claude.ai/code)
Co-Authored-By: Claude <[email protected]>
- README.md +13 -2
- generate-responses.py +63 -15
README.md
CHANGED
@@ -52,32 +52,43 @@ hf jobs uv run \
|
|
52 |
|
53 |
### generate-responses.py
|
54 |
|
55 |
-
Generate responses for
|
56 |
|
57 |
**Features:**
|
58 |
|
59 |
- π¬ Automatic chat template application
|
|
|
60 |
- π Multi-GPU tensor parallelism support
|
61 |
- π Smart filtering for prompts exceeding context length
|
62 |
- π Comprehensive dataset cards with generation metadata
|
63 |
- β‘ HF Transfer enabled for fast model downloads
|
64 |
- ποΈ Full control over sampling parameters
|
|
|
65 |
|
66 |
**Usage:**
|
67 |
|
68 |
```bash
|
69 |
-
#
|
70 |
uv run generate-responses.py \
|
71 |
username/input-dataset \
|
72 |
username/output-dataset \
|
73 |
--messages-column messages \
|
74 |
--max-tokens 1024
|
75 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
76 |
# With custom model and parameters
|
77 |
uv run generate-responses.py \
|
78 |
username/input-dataset \
|
79 |
username/output-dataset \
|
80 |
--model-id meta-llama/Llama-3.1-8B-Instruct \
|
|
|
81 |
--temperature 0.9 \
|
82 |
--top-p 0.95 \
|
83 |
--max-model-len 8192
|
|
|
52 |
|
53 |
### generate-responses.py
|
54 |
|
55 |
+
Generate responses for prompts using generative LLMs (e.g., Llama, Qwen, Mistral) with vLLM's high-performance inference engine.
|
56 |
|
57 |
**Features:**
|
58 |
|
59 |
- π¬ Automatic chat template application
|
60 |
+
- π Support for both chat messages and plain text prompts
|
61 |
- π Multi-GPU tensor parallelism support
|
62 |
- π Smart filtering for prompts exceeding context length
|
63 |
- π Comprehensive dataset cards with generation metadata
|
64 |
- β‘ HF Transfer enabled for fast model downloads
|
65 |
- ποΈ Full control over sampling parameters
|
66 |
+
- π― Sample limiting with `--max-samples` for testing
|
67 |
|
68 |
**Usage:**
|
69 |
|
70 |
```bash
|
71 |
+
# With chat-formatted messages (default)
|
72 |
uv run generate-responses.py \
|
73 |
username/input-dataset \
|
74 |
username/output-dataset \
|
75 |
--messages-column messages \
|
76 |
--max-tokens 1024
|
77 |
|
78 |
+
# With plain text prompts (NEW!)
|
79 |
+
uv run generate-responses.py \
|
80 |
+
username/input-dataset \
|
81 |
+
username/output-dataset \
|
82 |
+
--prompt-column question \
|
83 |
+
--max-tokens 1024 \
|
84 |
+
--max-samples 100
|
85 |
+
|
86 |
# With custom model and parameters
|
87 |
uv run generate-responses.py \
|
88 |
username/input-dataset \
|
89 |
username/output-dataset \
|
90 |
--model-id meta-llama/Llama-3.1-8B-Instruct \
|
91 |
+
--prompt-column text \
|
92 |
--temperature 0.9 \
|
93 |
--top-p 0.95 \
|
94 |
--max-model-len 8192
|
generate-responses.py
CHANGED
@@ -82,6 +82,7 @@ def create_dataset_card(
|
|
82 |
source_dataset: str,
|
83 |
model_id: str,
|
84 |
messages_column: str,
|
|
|
85 |
sampling_params: SamplingParams,
|
86 |
tensor_parallel_size: int,
|
87 |
num_examples: int,
|
@@ -119,7 +120,7 @@ This dataset contains generated responses for prompts from [{source_dataset}](ht
|
|
119 |
## Generation Details
|
120 |
|
121 |
- **Source Dataset**: [{source_dataset}](https://huggingface.co/datasets/{source_dataset})
|
122 |
-
- **
|
123 |
- **Model**: [{model_id}](https://huggingface.co/{model_id})
|
124 |
- **Number of Examples**: {num_examples:,}
|
125 |
- **Generation Date**: {generation_time}{filtering_section}
|
@@ -154,7 +155,7 @@ uv run https://huggingface.co/datasets/uv-scripts/vllm/raw/main/generate-respons
|
|
154 |
{source_dataset} \\
|
155 |
<output-dataset> \\
|
156 |
--model-id {model_id} \\
|
157 |
-
--messages-column
|
158 |
--temperature {sampling_params.temperature} \\
|
159 |
--top-p {sampling_params.top_p} \\
|
160 |
--top-k {sampling_params.top_k} \\
|
@@ -168,6 +169,7 @@ def main(
|
|
168 |
output_dataset_hub_id: str,
|
169 |
model_id: str = "Qwen/Qwen3-30B-A3B-Instruct-2507",
|
170 |
messages_column: str = "messages",
|
|
|
171 |
output_column: str = "response",
|
172 |
temperature: float = 0.7,
|
173 |
top_p: float = 0.8,
|
@@ -179,6 +181,7 @@ def main(
|
|
179 |
max_model_len: Optional[int] = None,
|
180 |
tensor_parallel_size: Optional[int] = None,
|
181 |
skip_long_prompts: bool = True,
|
|
|
182 |
hf_token: Optional[str] = None,
|
183 |
):
|
184 |
"""
|
@@ -189,6 +192,7 @@ def main(
|
|
189 |
output_dataset_hub_id: Where to save results on Hugging Face Hub
|
190 |
model_id: Hugging Face model ID for generation
|
191 |
messages_column: Column name containing chat messages
|
|
|
192 |
output_column: Column name for generated responses
|
193 |
temperature: Sampling temperature
|
194 |
top_p: Top-p sampling parameter
|
@@ -200,6 +204,7 @@ def main(
|
|
200 |
max_model_len: Maximum model context length (None uses model default)
|
201 |
tensor_parallel_size: Number of GPUs to use (auto-detect if None)
|
202 |
skip_long_prompts: Skip prompts exceeding max_model_len instead of failing
|
|
|
203 |
hf_token: Hugging Face authentication token
|
204 |
"""
|
205 |
generation_start_time = datetime.now().isoformat()
|
@@ -261,15 +266,34 @@ def main(
|
|
261 |
# Load dataset
|
262 |
logger.info(f"Loading dataset: {src_dataset_hub_id}")
|
263 |
dataset = load_dataset(src_dataset_hub_id, split="train")
|
|
|
|
|
|
|
|
|
|
|
|
|
264 |
total_examples = len(dataset)
|
265 |
logger.info(f"Dataset loaded with {total_examples:,} examples")
|
266 |
|
267 |
-
#
|
268 |
-
if
|
269 |
-
|
270 |
-
|
271 |
-
|
272 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
273 |
|
274 |
# Get effective max length for filtering
|
275 |
if max_model_len is not None:
|
@@ -280,18 +304,29 @@ def main(
|
|
280 |
logger.info(f"Using effective max model length: {effective_max_len}")
|
281 |
|
282 |
# Process messages and apply chat template
|
283 |
-
logger.info("
|
284 |
all_prompts = []
|
285 |
valid_prompts = []
|
286 |
valid_indices = []
|
287 |
skipped_info = []
|
288 |
|
289 |
-
for i, example in enumerate(tqdm(dataset, desc="Processing
|
290 |
-
|
291 |
-
|
292 |
-
|
293 |
-
|
294 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
295 |
all_prompts.append(prompt)
|
296 |
|
297 |
# Count tokens if filtering is enabled
|
@@ -352,6 +387,7 @@ def main(
|
|
352 |
source_dataset=src_dataset_hub_id,
|
353 |
model_id=model_id,
|
354 |
messages_column=messages_column,
|
|
|
355 |
sampling_params=sampling_params,
|
356 |
tensor_parallel_size=tensor_parallel_size,
|
357 |
num_examples=total_examples,
|
@@ -419,12 +455,22 @@ Examples:
|
|
419 |
default="messages",
|
420 |
help="Column containing chat messages (default: messages)",
|
421 |
)
|
|
|
|
|
|
|
|
|
|
|
422 |
parser.add_argument(
|
423 |
"--output-column",
|
424 |
type=str,
|
425 |
default="response",
|
426 |
help="Column name for generated responses (default: response)",
|
427 |
)
|
|
|
|
|
|
|
|
|
|
|
428 |
parser.add_argument(
|
429 |
"--temperature",
|
430 |
type=float,
|
@@ -502,6 +548,7 @@ Examples:
|
|
502 |
output_dataset_hub_id=args.output_dataset_hub_id,
|
503 |
model_id=args.model_id,
|
504 |
messages_column=args.messages_column,
|
|
|
505 |
output_column=args.output_column,
|
506 |
temperature=args.temperature,
|
507 |
top_p=args.top_p,
|
@@ -513,6 +560,7 @@ Examples:
|
|
513 |
max_model_len=args.max_model_len,
|
514 |
tensor_parallel_size=args.tensor_parallel_size,
|
515 |
skip_long_prompts=args.skip_long_prompts,
|
|
|
516 |
hf_token=args.hf_token,
|
517 |
)
|
518 |
else:
|
|
|
82 |
source_dataset: str,
|
83 |
model_id: str,
|
84 |
messages_column: str,
|
85 |
+
prompt_column: Optional[str],
|
86 |
sampling_params: SamplingParams,
|
87 |
tensor_parallel_size: int,
|
88 |
num_examples: int,
|
|
|
120 |
## Generation Details
|
121 |
|
122 |
- **Source Dataset**: [{source_dataset}](https://huggingface.co/datasets/{source_dataset})
|
123 |
+
- **Input Column**: `{prompt_column if prompt_column else messages_column}` ({'plain text prompts' if prompt_column else 'chat messages'})
|
124 |
- **Model**: [{model_id}](https://huggingface.co/{model_id})
|
125 |
- **Number of Examples**: {num_examples:,}
|
126 |
- **Generation Date**: {generation_time}{filtering_section}
|
|
|
155 |
{source_dataset} \\
|
156 |
<output-dataset> \\
|
157 |
--model-id {model_id} \\
|
158 |
+
{'--prompt-column ' + prompt_column if prompt_column else '--messages-column ' + messages_column} \\
|
159 |
--temperature {sampling_params.temperature} \\
|
160 |
--top-p {sampling_params.top_p} \\
|
161 |
--top-k {sampling_params.top_k} \\
|
|
|
169 |
output_dataset_hub_id: str,
|
170 |
model_id: str = "Qwen/Qwen3-30B-A3B-Instruct-2507",
|
171 |
messages_column: str = "messages",
|
172 |
+
prompt_column: Optional[str] = None,
|
173 |
output_column: str = "response",
|
174 |
temperature: float = 0.7,
|
175 |
top_p: float = 0.8,
|
|
|
181 |
max_model_len: Optional[int] = None,
|
182 |
tensor_parallel_size: Optional[int] = None,
|
183 |
skip_long_prompts: bool = True,
|
184 |
+
max_samples: Optional[int] = None,
|
185 |
hf_token: Optional[str] = None,
|
186 |
):
|
187 |
"""
|
|
|
192 |
output_dataset_hub_id: Where to save results on Hugging Face Hub
|
193 |
model_id: Hugging Face model ID for generation
|
194 |
messages_column: Column name containing chat messages
|
195 |
+
prompt_column: Column name containing plain text prompts (alternative to messages_column)
|
196 |
output_column: Column name for generated responses
|
197 |
temperature: Sampling temperature
|
198 |
top_p: Top-p sampling parameter
|
|
|
204 |
max_model_len: Maximum model context length (None uses model default)
|
205 |
tensor_parallel_size: Number of GPUs to use (auto-detect if None)
|
206 |
skip_long_prompts: Skip prompts exceeding max_model_len instead of failing
|
207 |
+
max_samples: Maximum number of samples to process (None for all)
|
208 |
hf_token: Hugging Face authentication token
|
209 |
"""
|
210 |
generation_start_time = datetime.now().isoformat()
|
|
|
266 |
# Load dataset
|
267 |
logger.info(f"Loading dataset: {src_dataset_hub_id}")
|
268 |
dataset = load_dataset(src_dataset_hub_id, split="train")
|
269 |
+
|
270 |
+
# Apply max_samples if specified
|
271 |
+
if max_samples is not None and max_samples < len(dataset):
|
272 |
+
logger.info(f"Limiting dataset to {max_samples} samples")
|
273 |
+
dataset = dataset.select(range(max_samples))
|
274 |
+
|
275 |
total_examples = len(dataset)
|
276 |
logger.info(f"Dataset loaded with {total_examples:,} examples")
|
277 |
|
278 |
+
# Determine which column to use and validate
|
279 |
+
if prompt_column:
|
280 |
+
# Use prompt column mode
|
281 |
+
if prompt_column not in dataset.column_names:
|
282 |
+
logger.error(
|
283 |
+
f"Column '{prompt_column}' not found. Available columns: {dataset.column_names}"
|
284 |
+
)
|
285 |
+
sys.exit(1)
|
286 |
+
logger.info(f"Using prompt column mode with column: '{prompt_column}'")
|
287 |
+
use_messages = False
|
288 |
+
else:
|
289 |
+
# Use messages column mode
|
290 |
+
if messages_column not in dataset.column_names:
|
291 |
+
logger.error(
|
292 |
+
f"Column '{messages_column}' not found. Available columns: {dataset.column_names}"
|
293 |
+
)
|
294 |
+
sys.exit(1)
|
295 |
+
logger.info(f"Using messages column mode with column: '{messages_column}'")
|
296 |
+
use_messages = True
|
297 |
|
298 |
# Get effective max length for filtering
|
299 |
if max_model_len is not None:
|
|
|
304 |
logger.info(f"Using effective max model length: {effective_max_len}")
|
305 |
|
306 |
# Process messages and apply chat template
|
307 |
+
logger.info("Preparing prompts...")
|
308 |
all_prompts = []
|
309 |
valid_prompts = []
|
310 |
valid_indices = []
|
311 |
skipped_info = []
|
312 |
|
313 |
+
for i, example in enumerate(tqdm(dataset, desc="Processing prompts")):
|
314 |
+
if use_messages:
|
315 |
+
# Messages mode: use existing chat messages
|
316 |
+
messages = example[messages_column]
|
317 |
+
# Apply chat template
|
318 |
+
prompt = tokenizer.apply_chat_template(
|
319 |
+
messages, tokenize=False, add_generation_prompt=True
|
320 |
+
)
|
321 |
+
else:
|
322 |
+
# Prompt mode: convert plain text to messages format
|
323 |
+
user_prompt = example[prompt_column]
|
324 |
+
messages = [{"role": "user", "content": user_prompt}]
|
325 |
+
# Apply chat template
|
326 |
+
prompt = tokenizer.apply_chat_template(
|
327 |
+
messages, tokenize=False, add_generation_prompt=True
|
328 |
+
)
|
329 |
+
|
330 |
all_prompts.append(prompt)
|
331 |
|
332 |
# Count tokens if filtering is enabled
|
|
|
387 |
source_dataset=src_dataset_hub_id,
|
388 |
model_id=model_id,
|
389 |
messages_column=messages_column,
|
390 |
+
prompt_column=prompt_column,
|
391 |
sampling_params=sampling_params,
|
392 |
tensor_parallel_size=tensor_parallel_size,
|
393 |
num_examples=total_examples,
|
|
|
455 |
default="messages",
|
456 |
help="Column containing chat messages (default: messages)",
|
457 |
)
|
458 |
+
parser.add_argument(
|
459 |
+
"--prompt-column",
|
460 |
+
type=str,
|
461 |
+
help="Column containing plain text prompts (alternative to --messages-column)",
|
462 |
+
)
|
463 |
parser.add_argument(
|
464 |
"--output-column",
|
465 |
type=str,
|
466 |
default="response",
|
467 |
help="Column name for generated responses (default: response)",
|
468 |
)
|
469 |
+
parser.add_argument(
|
470 |
+
"--max-samples",
|
471 |
+
type=int,
|
472 |
+
help="Maximum number of samples to process (default: all)",
|
473 |
+
)
|
474 |
parser.add_argument(
|
475 |
"--temperature",
|
476 |
type=float,
|
|
|
548 |
output_dataset_hub_id=args.output_dataset_hub_id,
|
549 |
model_id=args.model_id,
|
550 |
messages_column=args.messages_column,
|
551 |
+
prompt_column=args.prompt_column,
|
552 |
output_column=args.output_column,
|
553 |
temperature=args.temperature,
|
554 |
top_p=args.top_p,
|
|
|
560 |
max_model_len=args.max_model_len,
|
561 |
tensor_parallel_size=args.tensor_parallel_size,
|
562 |
skip_long_prompts=args.skip_long_prompts,
|
563 |
+
max_samples=args.max_samples,
|
564 |
hf_token=args.hf_token,
|
565 |
)
|
566 |
else:
|