Commit
·
ded6d1c
1
Parent(s):
791a4a1
add system prompt rewriter for more dynamic generation SFT
Browse files
src/synthetic_dataset_generator/apps/sft.py
CHANGED
@@ -1,4 +1,5 @@
|
|
1 |
import ast
|
|
|
2 |
import uuid
|
3 |
from typing import Dict, List, Union
|
4 |
|
@@ -32,6 +33,7 @@ from synthetic_dataset_generator.pipelines.sft import (
|
|
32 |
generate_pipeline_code,
|
33 |
get_magpie_generator,
|
34 |
get_prompt_generator,
|
|
|
35 |
get_response_generator,
|
36 |
)
|
37 |
from synthetic_dataset_generator.utils import (
|
@@ -103,6 +105,7 @@ def generate_dataset(
|
|
103 |
) -> pd.DataFrame:
|
104 |
num_rows = test_max_num_rows(num_rows)
|
105 |
progress(0.0, desc="(1/2) Generating instructions")
|
|
|
106 |
magpie_generator = get_magpie_generator(
|
107 |
system_prompt, num_turns, temperature, is_sample
|
108 |
)
|
@@ -112,6 +115,16 @@ def generate_dataset(
|
|
112 |
total_steps: int = num_rows * 2
|
113 |
batch_size = DEFAULT_BATCH_SIZE
|
114 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
115 |
# create instructions
|
116 |
n_processed = 0
|
117 |
magpie_results = []
|
@@ -123,7 +136,8 @@ def generate_dataset(
|
|
123 |
)
|
124 |
remaining_rows = num_rows - n_processed
|
125 |
batch_size = min(batch_size, remaining_rows)
|
126 |
-
|
|
|
127 |
batch = list(magpie_generator.process(inputs=inputs))
|
128 |
magpie_results.extend(batch[0])
|
129 |
n_processed += batch_size
|
@@ -487,7 +501,7 @@ with gr.Blocks() as app:
|
|
487 |
with gr.Column(scale=3):
|
488 |
success_message = gr.Markdown(
|
489 |
visible=True,
|
490 |
-
|
491 |
)
|
492 |
with gr.Accordion(
|
493 |
"Customize your pipeline with distilabel",
|
@@ -543,6 +557,7 @@ with gr.Blocks() as app:
|
|
543 |
fn=hide_pipeline_code_visibility,
|
544 |
inputs=[],
|
545 |
outputs=[pipeline_code_ui],
|
|
|
546 |
).success(
|
547 |
fn=push_dataset,
|
548 |
inputs=[
|
|
|
1 |
import ast
|
2 |
+
import random
|
3 |
import uuid
|
4 |
from typing import Dict, List, Union
|
5 |
|
|
|
33 |
generate_pipeline_code,
|
34 |
get_magpie_generator,
|
35 |
get_prompt_generator,
|
36 |
+
get_prompt_rewriter,
|
37 |
get_response_generator,
|
38 |
)
|
39 |
from synthetic_dataset_generator.utils import (
|
|
|
105 |
) -> pd.DataFrame:
|
106 |
num_rows = test_max_num_rows(num_rows)
|
107 |
progress(0.0, desc="(1/2) Generating instructions")
|
108 |
+
prompt_rewriter = get_prompt_rewriter()
|
109 |
magpie_generator = get_magpie_generator(
|
110 |
system_prompt, num_turns, temperature, is_sample
|
111 |
)
|
|
|
115 |
total_steps: int = num_rows * 2
|
116 |
batch_size = DEFAULT_BATCH_SIZE
|
117 |
|
118 |
+
# create prompt rewrites
|
119 |
+
inputs = [
|
120 |
+
{
|
121 |
+
"instruction": f"Rewrite this prompt keeping the same structure but highlighting different aspects of the original without adding anything new. Original prompt: {system_prompt} Rewritten prompt: "
|
122 |
+
}
|
123 |
+
for i in range(int(num_rows / 50))
|
124 |
+
]
|
125 |
+
batch = list(prompt_rewriter.process(inputs=inputs))
|
126 |
+
prompt_rewrites = [entry["generation"] for entry in batch[0]] + [system_prompt]
|
127 |
+
|
128 |
# create instructions
|
129 |
n_processed = 0
|
130 |
magpie_results = []
|
|
|
136 |
)
|
137 |
remaining_rows = num_rows - n_processed
|
138 |
batch_size = min(batch_size, remaining_rows)
|
139 |
+
rewritten_system_prompt = random.choice(prompt_rewrites)
|
140 |
+
inputs = [{"system_prompt": rewritten_system_prompt} for _ in range(batch_size)]
|
141 |
batch = list(magpie_generator.process(inputs=inputs))
|
142 |
magpie_results.extend(batch[0])
|
143 |
n_processed += batch_size
|
|
|
501 |
with gr.Column(scale=3):
|
502 |
success_message = gr.Markdown(
|
503 |
visible=True,
|
504 |
+
min_height=100, # don't remove this otherwise progress is not visible
|
505 |
)
|
506 |
with gr.Accordion(
|
507 |
"Customize your pipeline with distilabel",
|
|
|
557 |
fn=hide_pipeline_code_visibility,
|
558 |
inputs=[],
|
559 |
outputs=[pipeline_code_ui],
|
560 |
+
show_progress=True,
|
561 |
).success(
|
562 |
fn=push_dataset,
|
563 |
inputs=[
|
src/synthetic_dataset_generator/pipelines/sft.py
CHANGED
@@ -175,12 +175,11 @@ def get_magpie_generator(system_prompt, num_turns, temperature, is_sample):
|
|
175 |
generation_kwargs={
|
176 |
"temperature": temperature,
|
177 |
"do_sample": True,
|
178 |
-
"max_new_tokens": 256 if is_sample else MAX_NUM_TOKENS,
|
179 |
"stop_sequences": _STOP_SEQUENCES,
|
180 |
},
|
181 |
),
|
182 |
n_turns=num_turns,
|
183 |
-
system_prompt=system_prompt,
|
184 |
output_mappings=output_mappings,
|
185 |
only_instruction=True,
|
186 |
)
|
@@ -195,19 +194,34 @@ def get_magpie_generator(system_prompt, num_turns, temperature, is_sample):
|
|
195 |
generation_kwargs={
|
196 |
"temperature": temperature,
|
197 |
"do_sample": True,
|
198 |
-
"max_new_tokens": 256 if is_sample else MAX_NUM_TOKENS,
|
199 |
"stop_sequences": _STOP_SEQUENCES,
|
200 |
},
|
201 |
),
|
202 |
end_with_user=True,
|
203 |
n_turns=num_turns,
|
204 |
-
system_prompt=system_prompt,
|
205 |
output_mappings=output_mappings,
|
206 |
)
|
207 |
magpie_generator.load()
|
208 |
return magpie_generator
|
209 |
|
210 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
211 |
def get_response_generator(system_prompt, num_turns, temperature, is_sample):
|
212 |
if num_turns == 1:
|
213 |
response_generator = TextGeneration(
|
@@ -218,7 +232,7 @@ def get_response_generator(system_prompt, num_turns, temperature, is_sample):
|
|
218 |
api_key=_get_next_api_key(),
|
219 |
generation_kwargs={
|
220 |
"temperature": temperature,
|
221 |
-
"max_new_tokens": 256 if is_sample else MAX_NUM_TOKENS,
|
222 |
},
|
223 |
),
|
224 |
system_prompt=system_prompt,
|
|
|
175 |
generation_kwargs={
|
176 |
"temperature": temperature,
|
177 |
"do_sample": True,
|
178 |
+
"max_new_tokens": 256 if is_sample else int(MAX_NUM_TOKENS * 0.25),
|
179 |
"stop_sequences": _STOP_SEQUENCES,
|
180 |
},
|
181 |
),
|
182 |
n_turns=num_turns,
|
|
|
183 |
output_mappings=output_mappings,
|
184 |
only_instruction=True,
|
185 |
)
|
|
|
194 |
generation_kwargs={
|
195 |
"temperature": temperature,
|
196 |
"do_sample": True,
|
197 |
+
"max_new_tokens": 256 if is_sample else int(MAX_NUM_TOKENS * 0.5),
|
198 |
"stop_sequences": _STOP_SEQUENCES,
|
199 |
},
|
200 |
),
|
201 |
end_with_user=True,
|
202 |
n_turns=num_turns,
|
|
|
203 |
output_mappings=output_mappings,
|
204 |
)
|
205 |
magpie_generator.load()
|
206 |
return magpie_generator
|
207 |
|
208 |
|
209 |
+
def get_prompt_rewriter():
|
210 |
+
prompt_rewriter = TextGeneration(
|
211 |
+
llm=InferenceEndpointsLLM(
|
212 |
+
model_id=MODEL,
|
213 |
+
tokenizer_id=MODEL,
|
214 |
+
base_url=BASE_URL,
|
215 |
+
api_key=_get_next_api_key(),
|
216 |
+
generation_kwargs={
|
217 |
+
"temperature": 1,
|
218 |
+
},
|
219 |
+
),
|
220 |
+
)
|
221 |
+
prompt_rewriter.load()
|
222 |
+
return prompt_rewriter
|
223 |
+
|
224 |
+
|
225 |
def get_response_generator(system_prompt, num_turns, temperature, is_sample):
|
226 |
if num_turns == 1:
|
227 |
response_generator = TextGeneration(
|
|
|
232 |
api_key=_get_next_api_key(),
|
233 |
generation_kwargs={
|
234 |
"temperature": temperature,
|
235 |
+
"max_new_tokens": 256 if is_sample else int(MAX_NUM_TOKENS * 0.5),
|
236 |
},
|
237 |
),
|
238 |
system_prompt=system_prompt,
|