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
|
# ------------------------------------
# Copyright (c) Microsoft Corporation.
# Licensed under the MIT License.
# ------------------------------------
import pytest
from pathlib import Path
from test_base import TestBase, servicePreparer
from devtools_testutils import recorded_by_proxy, is_live_and_not_recording
@pytest.mark.skipif(
condition=(not is_live_and_not_recording()),
reason="Skipped because we cannot record network calls with AOAI client",
)
class TestFineTuning(TestBase):
def _create_sft_finetuning_job(
self, openai_client, train_file_id, validation_file_id, model_type="openai", training_type="Standard"
):
"""Helper method to create a supervised fine-tuning job."""
return openai_client.fine_tuning.jobs.create(
training_file=train_file_id,
validation_file=validation_file_id,
model=self.test_finetuning_params["sft"][model_type]["model_name"],
method={
"type": "supervised",
"supervised": {
"hyperparameters": {
"n_epochs": self.test_finetuning_params["n_epochs"],
"batch_size": self.test_finetuning_params["batch_size"],
"learning_rate_multiplier": self.test_finetuning_params["learning_rate_multiplier"],
}
},
},
extra_body={"trainingType": training_type},
)
def _create_dpo_finetuning_job(self, openai_client, train_file_id, validation_file_id):
"""Helper method to create a DPO fine-tuning job."""
return openai_client.fine_tuning.jobs.create(
training_file=train_file_id,
validation_file=validation_file_id,
model=self.test_finetuning_params["dpo"]["openai"]["model_name"],
method={
"type": "dpo",
"dpo": {
"hyperparameters": {
"n_epochs": self.test_finetuning_params["n_epochs"],
"batch_size": self.test_finetuning_params["batch_size"],
"learning_rate_multiplier": self.test_finetuning_params["learning_rate_multiplier"],
}
},
},
extra_body={"trainingType": "Standard"},
)
def _create_rft_finetuning_job(self, openai_client, train_file_id, validation_file_id):
"""Helper method to create an RFT fine-tuning job."""
grader = {
"name": "Response Quality Grader",
"type": "score_model",
"model": "o3-mini",
"input": [
{
"role": "user",
"content": "Evaluate the model's response based on correctness and quality. Rate from 0 to 10.",
}
],
"range": [0.0, 10.0],
}
return openai_client.fine_tuning.jobs.create(
training_file=train_file_id,
validation_file=validation_file_id,
model=self.test_finetuning_params["rft"]["openai"]["model_name"],
method={
"type": "reinforcement",
"reinforcement": {
"grader": grader,
"hyperparameters": {
"n_epochs": self.test_finetuning_params["n_epochs"],
"batch_size": self.test_finetuning_params["batch_size"],
"learning_rate_multiplier": self.test_finetuning_params["learning_rate_multiplier"],
"eval_interval": 5,
"eval_samples": 2,
"reasoning_effort": "medium",
},
},
},
extra_body={"trainingType": "Standard"},
)
def _upload_test_files(self, openai_client, job_type="sft"):
"""Helper method to upload training and validation files for fine-tuning tests."""
test_data_dir = Path(__file__).parent.parent / "test_data" / "finetuning"
training_file_path = test_data_dir / self.test_finetuning_params[job_type]["training_file_name"]
validation_file_path = test_data_dir / self.test_finetuning_params[job_type]["validation_file_name"]
with open(training_file_path, "rb") as f:
train_file = openai_client.files.create(file=f, purpose="fine-tune")
train_processed_file = openai_client.files.wait_for_processing(train_file.id)
assert train_processed_file is not None
assert train_processed_file.id is not None
TestBase.assert_equal_or_not_none(train_processed_file.status, "processed")
print(f"[test_finetuning_{job_type}] Uploaded training file: {train_processed_file.id}")
with open(validation_file_path, "rb") as f:
validation_file = openai_client.files.create(file=f, purpose="fine-tune")
validation_processed_file = openai_client.files.wait_for_processing(validation_file.id)
assert validation_processed_file is not None
assert validation_processed_file.id is not None
TestBase.assert_equal_or_not_none(validation_processed_file.status, "processed")
print(f"[test_finetuning_{job_type}] Uploaded validation file: {validation_processed_file.id}")
return train_processed_file, validation_processed_file
def _cleanup_test_files(self, openai_client, train_file, validation_file, job_type):
"""Helper method to clean up uploaded files after testing."""
openai_client.files.delete(train_file.id)
print(f"[test_finetuning_{job_type}] Deleted training file: {train_file.id}")
openai_client.files.delete(validation_file.id)
print(f"[test_finetuning_{job_type}] Deleted validation file: {validation_file.id}")
@servicePreparer()
@recorded_by_proxy
def test_sft_finetuning_create_job(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_sft] Created fine-tuning job: {fine_tuning_job.id}")
TestBase.validate_fine_tuning_job(fine_tuning_job)
TestBase.assert_equal_or_not_none(fine_tuning_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(fine_tuning_job.validation_file, validation_file.id)
assert fine_tuning_job.method is not None, "Method should not be None for SFT job"
TestBase.assert_equal_or_not_none(fine_tuning_job.method.type, "supervised")
print(f"[test_finetuning_sft] SFT method validation passed - type: {fine_tuning_job.method.type}")
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "sft")
@servicePreparer()
@recorded_by_proxy
def test_finetuning_retrieve_job(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_sft] Created job: {fine_tuning_job.id}")
retrieved_job = openai_client.fine_tuning.jobs.retrieve(fine_tuning_job.id)
print(f"[test_finetuning_sft] Retrieved job: {retrieved_job.id}")
TestBase.validate_fine_tuning_job(retrieved_job, expected_job_id=fine_tuning_job.id)
TestBase.assert_equal_or_not_none(retrieved_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(retrieved_job.validation_file, validation_file.id)
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "sft")
@servicePreparer()
@recorded_by_proxy
def test_finetuning_list_jobs(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_sft] Created job: {fine_tuning_job.id}")
jobs_list = list(openai_client.fine_tuning.jobs.list())
print(f"[test_finetuning_sft] Listed {len(jobs_list)} jobs")
assert len(jobs_list) > 0
job_ids = [job.id for job in jobs_list]
assert fine_tuning_job.id in job_ids
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "sft")
@servicePreparer()
@recorded_by_proxy
def test_finetuning_cancel_job(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_sft] Created job: {fine_tuning_job.id}")
cancelled_job = openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft] Cancelled job: {cancelled_job.id}")
TestBase.validate_fine_tuning_job(cancelled_job, expected_job_id=fine_tuning_job.id)
TestBase.assert_equal_or_not_none(cancelled_job.status, "cancelled")
retrieved_job = openai_client.fine_tuning.jobs.retrieve(fine_tuning_job.id)
print(f"[test_finetuning_sft] Verified cancellation persisted for job: {retrieved_job.id}")
TestBase.validate_fine_tuning_job(
retrieved_job, expected_job_id=fine_tuning_job.id, expected_status="cancelled"
)
self._cleanup_test_files(openai_client, train_file, validation_file, "sft")
@servicePreparer()
@recorded_by_proxy
def test_dpo_finetuning_create_job(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "dpo")
fine_tuning_job = self._create_dpo_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_dpo] Created DPO fine-tuning job: {fine_tuning_job.id}")
print(fine_tuning_job)
TestBase.validate_fine_tuning_job(fine_tuning_job)
TestBase.assert_equal_or_not_none(fine_tuning_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(fine_tuning_job.validation_file, validation_file.id)
assert fine_tuning_job.method is not None, "Method should not be None for DPO job"
TestBase.assert_equal_or_not_none(fine_tuning_job.method.type, "dpo")
print(f"[test_finetuning_dpo] DPO method validation passed - type: {fine_tuning_job.method.type}")
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_dpo] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "dpo")
@servicePreparer()
@recorded_by_proxy
def test_rft_finetuning_create_job(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "rft")
fine_tuning_job = self._create_rft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_rft] Created RFT fine-tuning job: {fine_tuning_job.id}")
TestBase.validate_fine_tuning_job(fine_tuning_job)
TestBase.assert_equal_or_not_none(fine_tuning_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(fine_tuning_job.validation_file, validation_file.id)
assert fine_tuning_job.method is not None, "Method should not be None for RFT job"
TestBase.assert_equal_or_not_none(fine_tuning_job.method.type, "reinforcement")
print(f"[test_finetuning_rft] RFT method validation passed - type: {fine_tuning_job.method.type}")
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_rft] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "rft")
@servicePreparer()
@recorded_by_proxy
def test_finetuning_list_events(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(openai_client, train_file.id, validation_file.id)
print(f"[test_finetuning_sft] Created job: {fine_tuning_job.id}")
TestBase.validate_fine_tuning_job(fine_tuning_job)
TestBase.assert_equal_or_not_none(fine_tuning_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(fine_tuning_job.validation_file, validation_file.id)
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft] Cancelled job: {fine_tuning_job.id}")
events_list = list(openai_client.fine_tuning.jobs.list_events(fine_tuning_job.id))
print(f"[test_finetuning_sft] Listed {len(events_list)} events for job: {fine_tuning_job.id}")
# Verify that events exist (at minimum, job creation event should be present)
assert len(events_list) > 0, "Fine-tuning job should have at least one event"
# Verify events have required attributes
for event in events_list:
assert event.id is not None, "Event should have an ID"
assert event.object is not None, "Event should have an object type"
assert event.created_at is not None, "Event should have a creation timestamp"
assert event.level is not None, "Event should have a level"
assert event.message is not None, "Event should have a message"
assert event.type is not None, "Event should have a type"
print(f"[test_finetuning_sft] Successfully validated {len(events_list)} events")
self._cleanup_test_files(openai_client, train_file, validation_file, "sft")
@servicePreparer()
@recorded_by_proxy
def test_sft_finetuning_create_job_oss_model(self, **kwargs):
with self.create_client(**kwargs) as project_client:
with project_client.get_openai_client() as openai_client:
train_file, validation_file = self._upload_test_files(openai_client, "sft")
fine_tuning_job = self._create_sft_finetuning_job(
openai_client, train_file.id, validation_file.id, "oss", "GlobalStandard"
)
print(f"[test_finetuning_sft_oss] Created fine-tuning job: {fine_tuning_job.id}")
TestBase.validate_fine_tuning_job(
fine_tuning_job, expected_model=self.test_finetuning_params["sft"]["oss"]["model_name"]
)
TestBase.assert_equal_or_not_none(fine_tuning_job.training_file, train_file.id)
TestBase.assert_equal_or_not_none(fine_tuning_job.validation_file, validation_file.id)
assert fine_tuning_job.method is not None, "Method should not be None for SFT job"
TestBase.assert_equal_or_not_none(fine_tuning_job.method.type, "supervised")
print(f"[test_finetuning_sft_oss] SFT method validation passed - type: {fine_tuning_job.method.type}")
openai_client.fine_tuning.jobs.cancel(fine_tuning_job.id)
print(f"[test_finetuning_sft_oss] Cancelled job: {fine_tuning_job.id}")
self._cleanup_test_files(openai_client, train_file, validation_file, "sft_oss")
|
