-
Notifications
You must be signed in to change notification settings - Fork 11
/
core.py
560 lines (450 loc) · 23.1 KB
/
core.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
import dspy
import pandas as pd
import re
import datetime
import os
import json
import numpy as np
from openai import OpenAI
from typing import List, Dict, Any
from dspy.evaluate import Evaluate
from dspy.teleprompt import BootstrapFewShot, BootstrapFewShotWithRandomSearch, MIPRO, MIPROv2, COPRO, BootstrapFinetune
from pydantic import create_model
# List of supported Groq models
SUPPORTED_GROQ_MODELS = [
"mixtral-8x7b-32768",
"gemma-7b-it",
"llama3-70b-8192",
"llama3-8b-8192",
"gemma2-9b-it"
]
SUPPORTED_GOOGLE_MODELS = [
"gemini-1.5-flash-8b",
"gemini-1.5-flash",
"gemini-1.5-pro"
]
# when using MIPRO or BootstrapFewShotWithRandomSearch, we need to configure the LM globally or it gives us a 'No LM loaded' error
lm = dspy.LM('openai/gpt-4o-mini')
dspy.configure(lm=lm)
def create_custom_signature(input_fields: List[str], output_fields: List[str], instructions: str, input_descs: List[str], output_descs: List[str]):
fields = {}
for i, field in enumerate(input_fields):
if i < len(input_descs) and input_descs[i]:
fields[field] = (str, dspy.InputField(default=..., desc=input_descs[i], json_schema_extra={"__dspy_field_type": "input"}))
else:
fields[field] = (str, dspy.InputField(default=..., json_schema_extra={"__dspy_field_type": "input"}))
for i, field in enumerate(output_fields):
if i < len(output_descs) and output_descs[i]:
fields[field] = (str, dspy.OutputField(default=..., desc=output_descs[i], json_schema_extra={"__dspy_field_type": "output"}))
else:
fields[field] = (str, dspy.OutputField(default=..., json_schema_extra={"__dspy_field_type": "output"}))
CustomSignatureModel = create_model('CustomSignatureModel', **fields)
class CustomSignature(dspy.Signature, CustomSignatureModel):
"""
{instructions}
"""
CustomSignature.__doc__ = CustomSignature.__doc__.format(instructions=instructions)
return CustomSignature
def generate_human_readable_id(input_fields: List[str], output_fields: List[str], dspy_module: str, llm_model: str, teacher_model: str, optimizer: str, instructions: str) -> str:
# Create a signature-based name
signature = '_'.join(input_fields + [':'] + output_fields)
signature_pascal = ''.join(word.capitalize() for word in signature.split('_'))
# Combine relevant information
model_name = ''.join(word.capitalize() for word in llm_model.split('-'))
module_name = dspy_module
optimizer_name = ''.join(word.capitalize() for word in optimizer.replace('bootstrap', 'bs').replace('randomsearch', 'rs').split('_'))
# Get current date
current_date = datetime.date.today().strftime("%Y%m%d")
# Create a human-readable ID with date
unique_id = f"{signature_pascal}-{model_name}_{module_name}_{optimizer_name}-{current_date}"
return unique_id
def create_dspy_module(dspy_module: str, CustomSignature: type, hint: str = None) -> dspy.Module:
if dspy_module == "Predict":
class CustomPredictModule(dspy.Module):
def __init__(self):
super().__init__()
self.predictor = dspy.Predict(CustomSignature)
def forward(self, **kwargs):
result = self.predictor(**kwargs)
return result
return CustomPredictModule()
elif dspy_module == "ChainOfThought":
class CustomChainOfThoughtModule(dspy.Module):
def __init__(self):
super().__init__()
self.cot = dspy.ChainOfThought(CustomSignature)
def forward(self, **kwargs):
return self.cot(**kwargs)
return CustomChainOfThoughtModule()
elif dspy_module == "ChainOfThoughtWithHint":
class CustomChainOfThoughtWithHintModule(dspy.Module):
def __init__(self):
super().__init__()
self.cot_with_hint = dspy.ChainOfThought(CustomSignature)
self.hint = hint
def forward(self, **kwargs):
# Inject the hint into the kwargs
kwargs['hint'] = self.hint
return self.cot_with_hint(**kwargs)
return CustomChainOfThoughtWithHintModule()
else:
raise ValueError(f"Unsupported DSPy module: {dspy_module}")
def compile_program(input_fields: List[str], output_fields: List[str], dspy_module: str, llm_model: str, teacher_model: str, example_data: List[Dict[Any, Any]], optimizer: str, instructions: str, metric_type: str, judge_prompt_id=None, input_descs: List[str] = None, output_descs: List[str] = None, hint: str = None) -> str:
# Set up the LLM model
if llm_model.startswith("gpt-"):
lm = dspy.LM(f'openai/{llm_model}')
elif llm_model.startswith("claude-"):
lm = dspy.LM(f'anthropic/{llm_model}')
elif llm_model in SUPPORTED_GROQ_MODELS:
lm = dspy.LM(f'groq/{llm_model}', api_key=os.environ.get("GROQ_API_KEY"))
elif llm_model in SUPPORTED_GOOGLE_MODELS:
lm = dspy.LM(f'google/{llm_model}', api_key=os.environ.get("GOOGLE_API_KEY"))
else:
raise ValueError(f"Unsupported LLM model: {llm_model}")
# Configure DSPy with the LM
dspy.configure(lm=lm)
# Verify that the LM is configured
assert dspy.settings.lm is not None, "Failed to configure LM"
# Set up the teacher model
if teacher_model.startswith("gpt-"):
teacher_lm = dspy.LM(f'openai/{teacher_model}')
elif teacher_model.startswith("claude-"):
teacher_lm = dspy.LM(f'anthropic/{teacher_model}')
elif teacher_model in SUPPORTED_GROQ_MODELS:
teacher_lm = dspy.LM(f'groq/{teacher_model}', api_key=os.environ.get("GROQ_API_KEY"))
elif teacher_model in SUPPORTED_GOOGLE_MODELS:
teacher_lm = dspy.LM(f'google/{teacher_model}', api_key=os.environ.get("GOOGLE_API_KEY"))
else:
raise ValueError(f"Unsupported teacher model: {teacher_model}")
# Create the custom signature
CustomSignature = create_custom_signature(input_fields, output_fields, instructions, input_descs or [], output_descs or [])
# Create the DSPy module using the new function
module = create_dspy_module(dspy_module, CustomSignature, hint)
# Convert DataFrame to list of dictionaries
example_data_list = example_data.to_dict('records')
# Check if there are at least two examples
if len(example_data_list) < 2:
raise ValueError("At least two examples are required for compilation.")
# Create dataset with correct field names and convert 'funny' to string
dataset = [dspy.Example(**{input_fields[i]: example[input_fields[i]] for i in range(len(input_fields))},
**{output_fields[i]: str(example[output_fields[i]]) for i in range(len(output_fields))}).with_inputs(*input_fields)
for example in example_data_list]
# Split the dataset
split_index = int(0.8 * len(dataset))
trainset, devset = dataset[:split_index], dataset[split_index:]
# Set up the evaluation metric
if metric_type == "Exact Match":
def metric(gold, pred, trace=None):
print("Gold:", gold)
print("Pred:", pred)
print("Pred type:", type(pred))
print("Pred attributes:", dir(pred))
if isinstance(pred, dspy.Prediction):
print("Prediction fields:", pred.__dict__)
# Check if pred is empty or None
if not pred or (isinstance(pred, dspy.Prediction) and not pred.__dict__):
print("Warning: Prediction is empty or None")
return 0
try:
return int(all(gold[field] == getattr(pred, field) for field in output_fields))
except AttributeError as e:
print(f"AttributeError: {e}")
return 0
elif metric_type == "Cosine Similarity":
# Initialize the OpenAI client
client = OpenAI()
def get_embedding(text):
response = client.embeddings.create(
model="text-embedding-3-small",
input=text
)
return response.data[0].embedding
def cosine_similarity(a, b):
return np.dot(a, b) / (np.linalg.norm(a) * np.linalg.norm(b))
def metric(gold, pred, trace=None):
gold_vector = np.concatenate([get_embedding(str(gold[field])) for field in output_fields])
pred_vector = np.concatenate([get_embedding(str(pred[field])) for field in output_fields])
similarity = cosine_similarity(gold_vector, pred_vector)
return similarity
elif metric_type == "LLM-as-a-Judge":
if judge_prompt_id is None:
raise ValueError("Judge prompt ID is required for LLM-as-a-Judge metric")
example2_id = "JokeTopic:Funny-Gpt4oMini_ChainOfThought_Bootstrapfewshotwithrandomsearch-20241003.json"
# Load the judge prompt details
if judge_prompt_id == example2_id:
judge_prompt_path = f"example_data/{judge_prompt_id}"
else:
judge_prompt_path = f"prompts/{judge_prompt_id}.json"
if not os.path.exists(judge_prompt_path):
raise ValueError(f"Judge prompt not found: {judge_prompt_path}")
with open(judge_prompt_path, 'r') as f:
judge_prompt_details = json.load(f)
print("Judge Prompt Path:", judge_prompt_path)
print("Judge Prompt Details:", judge_prompt_details)
judge_input_fields = judge_prompt_details.get('input_fields', [])
judge_input_descs = judge_prompt_details.get('input_descs', [])
judge_output_fields = judge_prompt_details.get('output_fields', [])
judge_output_descs = judge_prompt_details.get('output_descs', [])
judge_module = judge_prompt_details.get('dspy_module', 'Predict')
judge_instructions = judge_prompt_details.get('instructions', '')
judge_human_readable_id = judge_prompt_details.get('human_readable_id')
print("Judge Prompt Details:")
print(json.dumps(judge_prompt_details, indent=2))
# Create the custom signature for the judge program
JudgeSignature = create_custom_signature(judge_input_fields, judge_output_fields, judge_instructions, judge_input_descs, judge_output_descs)
print("\nJudge Signature:")
print(JudgeSignature)
# Create the judge program
judge_program = create_dspy_module(judge_module, JudgeSignature)
print("\nJudge Program:")
print(judge_program)
# Load the compiled judge program
if judge_prompt_id == example2_id:
judge_program_path = f"example_data/{judge_human_readable_id}-program.json"
else:
judge_program_path = f"programs/{judge_human_readable_id}.json"
if not os.path.exists(judge_program_path):
raise ValueError(f"Compiled judge program not found: {judge_program_path}")
with open(judge_program_path, 'r') as f:
judge_program_content = json.load(f)
print("\nCompiled Judge Program Content:")
print(json.dumps(judge_program_content, indent=2))
judge_program.load(judge_program_path)
def metric(gold, pred, trace=None):
try:
# Prepare input for the judge program based on judge_input_fields
judge_input = {}
for field in judge_input_fields:
if field in gold:
judge_input[field] = gold[field]
elif field in pred:
judge_input[field] = pred[field]
else:
print(f"Warning: Required judge input field '{field}' not found in gold or pred")
judge_input[field] = "" # or some default value
print("Judge Input:")
print(json.dumps(judge_input, indent=2))
# Run the judge program
result = judge_program(**judge_input)
print("Judge Program Result:")
print(result)
print("Result type:", type(result))
print("Result attributes:", dir(result))
if hasattr(result, 'toDict'):
print("Result as dict:", result.toDict())
# Extract the score from the judge output
if len(judge_output_fields) == 1:
score_field = judge_output_fields[0]
if hasattr(result, score_field):
score = getattr(result, score_field)
print(f"Score: {score}")
return float(score)
else:
# If the score field is not directly accessible, try to access it from the result dictionary
result_dict = result.toDict() if hasattr(result, 'toDict') else {}
if score_field in result_dict:
score = result_dict[score_field]
print(f"Score: {score}")
return float(score)
else:
print(f"Error: Judge program did not return expected field '{score_field}'")
print(f"Available fields: {result_dict.keys() if result_dict else dir(result)}")
return 0.0
else:
print(f"Error: Expected 1 output field, got {len(judge_output_fields)}")
print(f"Output fields: {judge_output_fields}")
return 0.0
except Exception as e:
print(f"Error in metric function: {str(e)}")
return 0.0 # Return a default score in case of error
else:
raise ValueError(f"Unknown metric type: {metric_type}")
# Use a single thread for evaluation
kwargs = dict(num_threads=1, display_progress=True, display_table=1)
# Evaluate the module to establish a baseline
baseline_evaluate = Evaluate(metric=metric, devset=devset, num_threads=1)
baseline_score = baseline_evaluate(module)
# Set up the optimizer
if optimizer == "BootstrapFewShot":
teleprompter = BootstrapFewShot(metric=metric, teacher_settings=dict(lm=teacher_lm))
compiled_program = teleprompter.compile(module, trainset=trainset)
elif optimizer == "BootstrapFewShotWithRandomSearch":
teleprompter = BootstrapFewShotWithRandomSearch(metric=metric, teacher_settings=dict(lm=teacher_lm), num_threads=1)
compiled_program = teleprompter.compile(module, trainset=trainset, valset=devset)
elif optimizer == "COPRO":
teleprompter = COPRO(metric=metric, teacher_settings=dict(lm=teacher_lm))
compiled_program = teleprompter.compile(module, trainset=trainset, eval_kwargs=kwargs)
elif optimizer == "MIPRO":
teleprompter = MIPRO(metric=metric, teacher_settings=dict(lm=teacher_lm), prompt_model=teacher_lm, task_model=lm)
num_trials = 10 # Adjust this value as needed
max_bootstrapped_demos = 5 # Adjust this value as needed
max_labeled_demos = 5 # Adjust this value as needed
compiled_program = teleprompter.compile(module, trainset=trainset, num_trials=num_trials,
max_bootstrapped_demos=max_bootstrapped_demos,
max_labeled_demos=max_labeled_demos,
eval_kwargs=kwargs, requires_permission_to_run=False)
elif optimizer == "MIPROv2":
teleprompter = MIPROv2(metric=metric, prompt_model=lm, task_model=teacher_lm, num_candidates=10, init_temperature=1.0)
num_batches = 30
max_bootstrapped_demos = 8
max_labeled_demos = 16
compiled_program = teleprompter.compile(
module,
trainset=trainset,
valset=devset,
num_batches=num_batches,
max_bootstrapped_demos=max_bootstrapped_demos,
max_labeled_demos=max_labeled_demos,
eval_kwargs=kwargs,
requires_permission_to_run=False
)
else:
raise ValueError(f"Unsupported optimizer: {optimizer}")
# Evaluate the compiled program
evaluate = Evaluate(metric=metric, devset=devset, num_threads=1)
score = evaluate(compiled_program)
print("Evaluation Score:")
print(score)
# Generate a human-readable ID for the compiled program
human_readable_id = generate_human_readable_id(input_fields, output_fields, dspy_module, llm_model, teacher_model, optimizer, instructions)
# Create datasets folder if it doesn't exist
os.makedirs('datasets', exist_ok=True)
# Save the dataframe to the datasets folder
dataset_path = os.path.join('datasets', f"{human_readable_id}.csv")
example_data.to_csv(dataset_path, index=False)
print(f"Dataset saved to {dataset_path}")
# Create 'programs' folder if it doesn't exist
os.makedirs('programs', exist_ok=True)
# Save the compiled program
compiled_program.save(f"programs/{human_readable_id}.json")
print(f"Compiled program saved to programs/{human_readable_id}.json")
usage_instructions = f"""Program compiled successfully!
Evaluation score: {score}
Baseline score: {baseline_score}
The compiled program has been saved as 'programs/{human_readable_id}.json'.
You can now use the compiled program as follows:
compiled_program = dspy.{dspy_module}(CustomSignature)
compiled_program.load('programs/{human_readable_id}.json')
result = compiled_program({', '.join(f'{field}=value' for field in input_fields)})
print({', '.join(f'result.{field}' for field in output_fields)})
"""
# Update the usage instructions to include the hint if applicable
if dspy_module == "ChainOfThoughtWithHint":
usage_instructions += f"\nHint: {hint}\n"
# Use the compiled program with the first row of example data
if len(example_data) > 0:
first_row = example_data.iloc[0]
input_data = {field: first_row[field] for field in input_fields}
result = compiled_program(**input_data)
messages = dspy.settings.lm.history[-1]['messages']
final_prompt = ""
for msg in messages:
final_prompt += f"{msg['content']}\n"
example_output = f"\nExample usage with first row of data:\n"
example_output += f"Input: {input_data}\n"
example_output += f"Output: {result}\n"
usage_instructions += example_output
return usage_instructions, final_prompt
# Function to list prompts
def list_prompts(signature_filter=None, output_filter=None):
if not os.path.exists('prompts'):
print("Prompts directory does not exist")
return []
files = os.listdir('prompts')
if not files:
print("No prompt files found in the prompts directory")
return []
prompt_details = []
for file in files:
if file.endswith('.json'):
with open(os.path.join('prompts', file), 'r') as f:
data = json.load(f)
prompt_id = file
signature = f"{', '.join(data['input_fields'])} -> {', '.join(data['output_fields'])}"
input_signature = f"{', '.join(data['input_fields'])}"
eval_score = data.get('evaluation_score', 'N/A')
# Exclude example data
details = {k: v for k, v in data.items() if k != 'example_data'}
# Check if signature_filter is provided and matches
if signature_filter and signature_filter.lower() not in signature.lower():
print(f"Skipping file {file} due to signature mismatch")
continue
# Check if output_filter is provided and matches
if output_filter:
if not all(filter_item.lower() in input_signature.lower() for filter_item in output_filter):
continue
prompt_details.append({
"ID": prompt_id,
"Signature": signature,
"Eval Score": eval_score,
"Details": json.dumps(details, indent=4) # Add full details as a JSON string
})
print(f"Found {len(prompt_details)} saved prompts")
return prompt_details # Return the list of prompts as dictionaries
def load_example_csv(example_name):
csv_path = f"example_data/{example_name}.csv"
try:
df = pd.read_csv(csv_path)
return df
except FileNotFoundError:
print(f"CSV file not found: {csv_path}")
return None
def export_to_csv(data):
df = pd.DataFrame(data)
filename = "exported_data.csv"
df.to_csv(filename, index=False)
return filename
# function to take a program from the program folder and run it on a row from the dataset
def generate_program_response(human_readable_id, row_data):
# Load the program details
program_path = f"programs/{human_readable_id}.json"
prompt_path = f"prompts/{human_readable_id}.json"
print("program_path:", program_path)
if not os.path.exists(program_path):
raise ValueError(f"Compiled program not found: {program_path}")
with open(prompt_path, 'r') as f:
program_details = json.load(f)
# Extract necessary information from program details
input_fields = program_details.get('input_fields', [])
input_descs = program_details.get('input_descs', [])
output_fields = program_details.get('output_fields', [])
output_descs = program_details.get('output_descs', [])
dspy_module = program_details.get('dspy_module', 'Predict')
instructions = program_details.get('instructions', '')
print("input_fields:", input_fields)
print("output_fields:", output_fields)
print("instructions:", instructions)
print("input_descs:", input_descs)
print("output_descs:", output_descs)
print("dspy_module:", dspy_module)
# Create the custom signature
CustomSignature = create_custom_signature(input_fields, output_fields, instructions, input_descs, output_descs)
print("CustomSignature:", CustomSignature)
compiled_program = create_dspy_module(dspy_module, CustomSignature)
print("compiled_program:", compiled_program)
compiled_program.load(program_path)
print("compiled_program after load:", compiled_program)
program_input = {}
for field in input_fields:
if field in row_data:
program_input[field] = row_data[field]
else:
print(f"Warning: Required input field '{field}' not found in row_data")
program_input[field] = "" # or some default value
# Run the program
try:
result = compiled_program(**program_input)
print("result:", result)
except Exception as e:
print(f"Error executing program: {str(e)}")
return f"Error: {str(e)}"
# Prepare the output
output = "Input:\n"
for field in input_fields:
output += f"{field}: {program_input[field]}\n"
print("result:", result)
output += "\nOutput:\n"
for field in output_fields:
output += f"{field}: {getattr(result, field)}\n"
print("output:", output)
return output