Hugging Face's logo

jeffwan
/
alpaca-lora

TensorBoard
Model card Files Metrics Community
jeffwan commited on
Commit
c3f935d
·
1 Parent(s): a7b0e49

Publish 4*A100-80G training manifests

Browse files
Files changed (17) hide show
  1. wandb/debug-cli.root.log +0 -0
  2. wandb/debug-internal.log +1 -0
  3. wandb/debug.log +1 -0
  4. wandb/latest-run +1 -0
  5. wandb/offline-run-20230328_173601-8ou0tfhe/files/requirements.txt +186 -0
  6. wandb/offline-run-20230328_173601-8ou0tfhe/files/wandb-metadata.json +708 -0
  7. wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug-internal.log +169 -0
  8. wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug.log +24 -0
  9. wandb/offline-run-20230328_173601-8ou0tfhe/run-8ou0tfhe.wandb +0 -0
  10. wandb/run-20230328_174559-mzmf7gh8/files/config.yaml +595 -0
  11. wandb/run-20230328_174559-mzmf7gh8/files/output.log +165 -0
  12. wandb/run-20230328_174559-mzmf7gh8/files/requirements.txt +186 -0
  13. wandb/run-20230328_174559-mzmf7gh8/files/wandb-metadata.json +708 -0
  14. wandb/run-20230328_174559-mzmf7gh8/files/wandb-summary.json +1 -0
  15. wandb/run-20230328_174559-mzmf7gh8/logs/debug-internal.log +0 -0
  16. wandb/run-20230328_174559-mzmf7gh8/logs/debug.log +27 -0
  17. wandb/run-20230328_174559-mzmf7gh8/run-mzmf7gh8.wandb +0 -0
wandb/debug-cli.root.log ADDED
File without changes
wandb/debug-internal.log ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20230328_174559-mzmf7gh8/logs/debug-internal.log
wandb/debug.log ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20230328_174559-mzmf7gh8/logs/debug.log
wandb/latest-run ADDED
@@ -0,0 +1 @@
 
 
1
+ run-20230328_174559-mzmf7gh8
wandb/offline-run-20230328_173601-8ou0tfhe/files/requirements.txt ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ absl-py==1.4.0
2
+ accelerate==0.18.0
3
+ aiofiles==22.1.0
4
+ aiohttp==3.8.4
5
+ aiosignal==1.3.1
6
+ aiosqlite==0.18.0
7
+ altair==4.2.2
8
+ anyio==3.6.2
9
+ appdirs==1.4.4
10
+ argon2-cffi-bindings==21.2.0
11
+ argon2-cffi==21.3.0
12
+ arrow==1.2.3
13
+ asttokens==2.2.1
14
+ async-timeout==4.0.2
15
+ attrs==22.2.0
16
+ babel==2.11.0
17
+ backcall==0.2.0
18
+ beautifulsoup4==4.11.2
19
+ bitsandbytes==0.37.2
20
+ black==23.1.0
21
+ bleach==6.0.0
22
+ certifi==2019.11.28
23
+ cffi==1.15.1
24
+ chardet==3.0.4
25
+ charset-normalizer==3.0.1
26
+ click==8.1.3
27
+ comm==0.1.2
28
+ contourpy==1.0.7
29
+ cycler==0.11.0
30
+ datasets==2.10.1
31
+ dbus-python==1.2.16
32
+ debugpy==1.6.6
33
+ decorator==5.1.1
34
+ defusedxml==0.7.1
35
+ dill==0.3.6
36
+ docker-pycreds==0.4.0
37
+ entrypoints==0.4
38
+ executing==1.2.0
39
+ fastapi==0.95.0
40
+ fastjsonschema==2.16.2
41
+ ffmpy==0.3.0
42
+ filelock==3.10.7
43
+ fire==0.5.0
44
+ fonttools==4.39.3
45
+ fqdn==1.5.1
46
+ frozenlist==1.3.3
47
+ fsspec==2023.3.0
48
+ gitdb==4.0.10
49
+ gitpython==3.1.31
50
+ gradio==3.23.0
51
+ h11==0.14.0
52
+ httpcore==0.16.3
53
+ httpx==0.23.3
54
+ huggingface-hub==0.13.3
55
+ idna==2.8
56
+ ipykernel==6.21.2
57
+ ipython-genutils==0.2.0
58
+ ipython==8.10.0
59
+ ipywidgets==8.0.4
60
+ isoduration==20.11.0
61
+ jedi==0.18.2
62
+ jinja2==3.1.2
63
+ joblib==1.2.0
64
+ json5==0.9.11
65
+ jsonpointer==2.3
66
+ jsonschema==4.17.3
67
+ jupyter-archive==3.3.4
68
+ jupyter-client==8.0.2
69
+ jupyter-core==5.2.0
70
+ jupyter-events==0.5.0
71
+ jupyter-server-fileid==0.6.0
72
+ jupyter-server-terminals==0.4.4
73
+ jupyter-server-ydoc==0.6.1
74
+ jupyter-server==2.3.0
75
+ jupyter-ydoc==0.2.2
76
+ jupyterlab-pygments==0.2.2
77
+ jupyterlab-server==2.19.0
78
+ jupyterlab-widgets==3.0.5
79
+ jupyterlab==3.6.1
80
+ kiwisolver==1.4.4
81
+ linkify-it-py==2.0.0
82
+ markdown-it-py==2.2.0
83
+ markupsafe==2.1.2
84
+ matplotlib-inline==0.1.6
85
+ matplotlib==3.7.1
86
+ mdit-py-plugins==0.3.3
87
+ mdurl==0.1.2
88
+ mistune==2.0.5
89
+ multidict==6.0.4
90
+ multiprocess==0.70.14
91
+ mypy-extensions==1.0.0
92
+ nbclassic==0.5.1
93
+ nbclient==0.7.2
94
+ nbconvert==7.2.9
95
+ nbformat==5.7.3
96
+ nest-asyncio==1.5.6
97
+ nltk==3.8.1
98
+ notebook-shim==0.2.2
99
+ notebook==6.5.2
100
+ numpy==1.24.2
101
+ openai==0.27.2
102
+ orjson==3.8.9
103
+ packaging==23.0
104
+ pandas==1.5.3
105
+ pandocfilters==1.5.0
106
+ parso==0.8.3
107
+ pathspec==0.11.1
108
+ pathtools==0.1.2
109
+ peft==0.3.0.dev0
110
+ pexpect==4.8.0
111
+ pickleshare==0.7.5
112
+ pillow==9.4.0
113
+ pip==23.0.1
114
+ platformdirs==3.0.0
115
+ prometheus-client==0.16.0
116
+ prompt-toolkit==3.0.36
117
+ protobuf==4.22.1
118
+ psutil==5.9.4
119
+ ptyprocess==0.7.0
120
+ pure-eval==0.2.2
121
+ pyarrow==11.0.0
122
+ pycparser==2.21
123
+ pydantic==1.10.7
124
+ pydub==0.25.1
125
+ pygments==2.14.0
126
+ pygobject==3.36.0
127
+ pyparsing==3.0.9
128
+ pyrsistent==0.19.3
129
+ python-apt==2.0.1
130
+ python-dateutil==2.8.2
131
+ python-json-logger==2.0.6
132
+ python-multipart==0.0.6
133
+ pytz==2022.7.1
134
+ pyyaml==6.0
135
+ pyzmq==25.0.0
136
+ regex==2023.3.23
137
+ requests-unixsocket==0.2.0
138
+ requests==2.28.2
139
+ responses==0.18.0
140
+ rfc3339-validator==0.1.4
141
+ rfc3986-validator==0.1.1
142
+ rfc3986==1.5.0
143
+ rouge-score==0.1.2
144
+ semantic-version==2.10.0
145
+ send2trash==1.8.0
146
+ sentencepiece==0.1.97
147
+ sentry-sdk==1.18.0
148
+ setproctitle==1.3.2
149
+ setuptools==67.6.1
150
+ six==1.14.0
151
+ smmap==5.0.0
152
+ sniffio==1.3.0
153
+ soupsieve==2.4
154
+ stack-data==0.6.2
155
+ starlette==0.26.1
156
+ termcolor==2.2.0
157
+ terminado==0.17.1
158
+ tinycss2==1.2.1
159
+ tokenize-rt==5.0.0
160
+ tokenizers==0.12.1
161
+ tomli==2.0.1
162
+ toolz==0.12.0
163
+ torch==1.13.1+cu116
164
+ torchaudio==0.13.1+cu116
165
+ torchvision==0.14.1+cu116
166
+ tornado==6.2
167
+ tqdm==4.65.0
168
+ traitlets==5.9.0
169
+ transformers==4.28.0.dev0
170
+ typing-extensions==4.5.0
171
+ uc-micro-py==1.0.1
172
+ uri-template==1.2.0
173
+ urllib3==1.26.15
174
+ uvicorn==0.21.1
175
+ wandb==0.14.0
176
+ wcwidth==0.2.6
177
+ webcolors==1.12
178
+ webencodings==0.5.1
179
+ websocket-client==1.5.1
180
+ websockets==10.4
181
+ wheel==0.38.4
182
+ widgetsnbextension==4.0.5
183
+ xxhash==3.2.0
184
+ y-py==0.5.5
185
+ yarl==1.8.2
186
+ ypy-websocket==0.8.2
wandb/offline-run-20230328_173601-8ou0tfhe/files/wandb-metadata.json ADDED
@@ -0,0 +1,708 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-5.4.0-113-generic-x86_64-with-glibc2.31",
3
+ "python": "3.10.10",
4
+ "heartbeatAt": "2023-03-28T17:36:01.807653",
5
+ "startedAt": "2023-03-28T17:36:01.715893",
6
+ "docker": null,
7
+ "cuda": null,
8
+ "args": [
9
+ "--base_model",
10
+ "/root/llama-7b-hf/",
11
+ "--data_path",
12
+ "./alpaca_data_cleaned.json",
13
+ "--output_dir",
14
+ "./lora-alpaca",
15
+ "--batch_size",
16
+ "1024",
17
+ "--micro_batch_size",
18
+ "128"
19
+ ],
20
+ "state": "running",
21
+ "program": "/root/alpaca-lora/finetune.py",
22
+ "codePath": "finetune.py",
23
+ "git": {
24
+ "remote": "https://github.com/tloen/alpaca-lora.git",
25
+ "commit": "345c8fbb7bec0eb05e81c13693d06465287a7e3d"
26
+ },
27
+ "email": null,
28
+ "root": "/root/alpaca-lora",
29
+ "host": "c35059eec1c7",
30
+ "username": "root",
31
+ "executable": "/usr/bin/python",
32
+ "cpu_count": 64,
33
+ "cpu_count_logical": 128,
34
+ "cpu_freq": {
35
+ "current": 1668.3349843750002,
36
+ "min": 1500.0,
37
+ "max": 2500.0
38
+ },
39
+ "cpu_freq_per_core": [
40
+ {
41
+ "current": 1499.835,
42
+ "min": 1500.0,
43
+ "max": 2500.0
44
+ },
45
+ {
46
+ "current": 1499.68,
47
+ "min": 1500.0,
48
+ "max": 2500.0
49
+ },
50
+ {
51
+ "current": 1499.109,
52
+ "min": 1500.0,
53
+ "max": 2500.0
54
+ },
55
+ {
56
+ "current": 1499.126,
57
+ "min": 1500.0,
58
+ "max": 2500.0
59
+ },
60
+ {
61
+ "current": 1499.324,
62
+ "min": 1500.0,
63
+ "max": 2500.0
64
+ },
65
+ {
66
+ "current": 1499.181,
67
+ "min": 1500.0,
68
+ "max": 2500.0
69
+ },
70
+ {
71
+ "current": 1499.484,
72
+ "min": 1500.0,
73
+ "max": 2500.0
74
+ },
75
+ {
76
+ "current": 1499.859,
77
+ "min": 1500.0,
78
+ "max": 2500.0
79
+ },
80
+ {
81
+ "current": 1498.666,
82
+ "min": 1500.0,
83
+ "max": 2500.0
84
+ },
85
+ {
86
+ "current": 1499.584,
87
+ "min": 1500.0,
88
+ "max": 2500.0
89
+ },
90
+ {
91
+ "current": 1499.174,
92
+ "min": 1500.0,
93
+ "max": 2500.0
94
+ },
95
+ {
96
+ "current": 1499.429,
97
+ "min": 1500.0,
98
+ "max": 2500.0
99
+ },
100
+ {
101
+ "current": 1498.836,
102
+ "min": 1500.0,
103
+ "max": 2500.0
104
+ },
105
+ {
106
+ "current": 1499.832,
107
+ "min": 1500.0,
108
+ "max": 2500.0
109
+ },
110
+ {
111
+ "current": 1499.39,
112
+ "min": 1500.0,
113
+ "max": 2500.0
114
+ },
115
+ {
116
+ "current": 1499.6,
117
+ "min": 1500.0,
118
+ "max": 2500.0
119
+ },
120
+ {
121
+ "current": 3346.562,
122
+ "min": 1500.0,
123
+ "max": 2500.0
124
+ },
125
+ {
126
+ "current": 1488.085,
127
+ "min": 1500.0,
128
+ "max": 2500.0
129
+ },
130
+ {
131
+ "current": 1488.039,
132
+ "min": 1500.0,
133
+ "max": 2500.0
134
+ },
135
+ {
136
+ "current": 1486.592,
137
+ "min": 1500.0,
138
+ "max": 2500.0
139
+ },
140
+ {
141
+ "current": 1487.74,
142
+ "min": 1500.0,
143
+ "max": 2500.0
144
+ },
145
+ {
146
+ "current": 1488.482,
147
+ "min": 1500.0,
148
+ "max": 2500.0
149
+ },
150
+ {
151
+ "current": 3348.706,
152
+ "min": 1500.0,
153
+ "max": 2500.0
154
+ },
155
+ {
156
+ "current": 1488.13,
157
+ "min": 1500.0,
158
+ "max": 2500.0
159
+ },
160
+ {
161
+ "current": 1498.984,
162
+ "min": 1500.0,
163
+ "max": 2500.0
164
+ },
165
+ {
166
+ "current": 1498.395,
167
+ "min": 1500.0,
168
+ "max": 2500.0
169
+ },
170
+ {
171
+ "current": 1499.979,
172
+ "min": 1500.0,
173
+ "max": 2500.0
174
+ },
175
+ {
176
+ "current": 1498.956,
177
+ "min": 1500.0,
178
+ "max": 2500.0
179
+ },
180
+ {
181
+ "current": 1499.173,
182
+ "min": 1500.0,
183
+ "max": 2500.0
184
+ },
185
+ {
186
+ "current": 1498.148,
187
+ "min": 1500.0,
188
+ "max": 2500.0
189
+ },
190
+ {
191
+ "current": 1499.279,
192
+ "min": 1500.0,
193
+ "max": 2500.0
194
+ },
195
+ {
196
+ "current": 1499.542,
197
+ "min": 1500.0,
198
+ "max": 2500.0
199
+ },
200
+ {
201
+ "current": 1494.104,
202
+ "min": 1500.0,
203
+ "max": 2500.0
204
+ },
205
+ {
206
+ "current": 1499.647,
207
+ "min": 1500.0,
208
+ "max": 2500.0
209
+ },
210
+ {
211
+ "current": 1499.189,
212
+ "min": 1500.0,
213
+ "max": 2500.0
214
+ },
215
+ {
216
+ "current": 1499.489,
217
+ "min": 1500.0,
218
+ "max": 2500.0
219
+ },
220
+ {
221
+ "current": 1499.785,
222
+ "min": 1500.0,
223
+ "max": 2500.0
224
+ },
225
+ {
226
+ "current": 1499.44,
227
+ "min": 1500.0,
228
+ "max": 2500.0
229
+ },
230
+ {
231
+ "current": 1499.902,
232
+ "min": 1500.0,
233
+ "max": 2500.0
234
+ },
235
+ {
236
+ "current": 1498.876,
237
+ "min": 1500.0,
238
+ "max": 2500.0
239
+ },
240
+ {
241
+ "current": 1496.697,
242
+ "min": 1500.0,
243
+ "max": 2500.0
244
+ },
245
+ {
246
+ "current": 1799.225,
247
+ "min": 1500.0,
248
+ "max": 2500.0
249
+ },
250
+ {
251
+ "current": 1490.788,
252
+ "min": 1500.0,
253
+ "max": 2500.0
254
+ },
255
+ {
256
+ "current": 1494.987,
257
+ "min": 1500.0,
258
+ "max": 2500.0
259
+ },
260
+ {
261
+ "current": 1465.434,
262
+ "min": 1500.0,
263
+ "max": 2500.0
264
+ },
265
+ {
266
+ "current": 1465.45,
267
+ "min": 1500.0,
268
+ "max": 2500.0
269
+ },
270
+ {
271
+ "current": 2182.568,
272
+ "min": 1500.0,
273
+ "max": 2500.0
274
+ },
275
+ {
276
+ "current": 1465.131,
277
+ "min": 1500.0,
278
+ "max": 2500.0
279
+ },
280
+ {
281
+ "current": 3347.661,
282
+ "min": 1500.0,
283
+ "max": 2500.0
284
+ },
285
+ {
286
+ "current": 3332.143,
287
+ "min": 1500.0,
288
+ "max": 2500.0
289
+ },
290
+ {
291
+ "current": 1484.646,
292
+ "min": 1500.0,
293
+ "max": 2500.0
294
+ },
295
+ {
296
+ "current": 1487.67,
297
+ "min": 1500.0,
298
+ "max": 2500.0
299
+ },
300
+ {
301
+ "current": 1499.77,
302
+ "min": 1500.0,
303
+ "max": 2500.0
304
+ },
305
+ {
306
+ "current": 1499.811,
307
+ "min": 1500.0,
308
+ "max": 2500.0
309
+ },
310
+ {
311
+ "current": 1499.561,
312
+ "min": 1500.0,
313
+ "max": 2500.0
314
+ },
315
+ {
316
+ "current": 1498.909,
317
+ "min": 1500.0,
318
+ "max": 2500.0
319
+ },
320
+ {
321
+ "current": 3340.022,
322
+ "min": 1500.0,
323
+ "max": 2500.0
324
+ },
325
+ {
326
+ "current": 1488.267,
327
+ "min": 1500.0,
328
+ "max": 2500.0
329
+ },
330
+ {
331
+ "current": 1488.782,
332
+ "min": 1500.0,
333
+ "max": 2500.0
334
+ },
335
+ {
336
+ "current": 1488.692,
337
+ "min": 1500.0,
338
+ "max": 2500.0
339
+ },
340
+ {
341
+ "current": 1498.424,
342
+ "min": 1500.0,
343
+ "max": 2500.0
344
+ },
345
+ {
346
+ "current": 1499.438,
347
+ "min": 1500.0,
348
+ "max": 2500.0
349
+ },
350
+ {
351
+ "current": 1499.609,
352
+ "min": 1500.0,
353
+ "max": 2500.0
354
+ },
355
+ {
356
+ "current": 1499.979,
357
+ "min": 1500.0,
358
+ "max": 2500.0
359
+ },
360
+ {
361
+ "current": 1499.002,
362
+ "min": 1500.0,
363
+ "max": 2500.0
364
+ },
365
+ {
366
+ "current": 1498.842,
367
+ "min": 1500.0,
368
+ "max": 2500.0
369
+ },
370
+ {
371
+ "current": 1499.564,
372
+ "min": 1500.0,
373
+ "max": 2500.0
374
+ },
375
+ {
376
+ "current": 1500.207,
377
+ "min": 1500.0,
378
+ "max": 2500.0
379
+ },
380
+ {
381
+ "current": 1499.855,
382
+ "min": 1500.0,
383
+ "max": 2500.0
384
+ },
385
+ {
386
+ "current": 1498.791,
387
+ "min": 1500.0,
388
+ "max": 2500.0
389
+ },
390
+ {
391
+ "current": 1499.935,
392
+ "min": 1500.0,
393
+ "max": 2500.0
394
+ },
395
+ {
396
+ "current": 1499.886,
397
+ "min": 1500.0,
398
+ "max": 2500.0
399
+ },
400
+ {
401
+ "current": 1499.045,
402
+ "min": 1500.0,
403
+ "max": 2500.0
404
+ },
405
+ {
406
+ "current": 1499.729,
407
+ "min": 1500.0,
408
+ "max": 2500.0
409
+ },
410
+ {
411
+ "current": 1498.851,
412
+ "min": 1500.0,
413
+ "max": 2500.0
414
+ },
415
+ {
416
+ "current": 1499.735,
417
+ "min": 1500.0,
418
+ "max": 2500.0
419
+ },
420
+ {
421
+ "current": 1499.435,
422
+ "min": 1500.0,
423
+ "max": 2500.0
424
+ },
425
+ {
426
+ "current": 1498.651,
427
+ "min": 1500.0,
428
+ "max": 2500.0
429
+ },
430
+ {
431
+ "current": 1499.641,
432
+ "min": 1500.0,
433
+ "max": 2500.0
434
+ },
435
+ {
436
+ "current": 1498.522,
437
+ "min": 1500.0,
438
+ "max": 2500.0
439
+ },
440
+ {
441
+ "current": 3350.427,
442
+ "min": 1500.0,
443
+ "max": 2500.0
444
+ },
445
+ {
446
+ "current": 1488.653,
447
+ "min": 1500.0,
448
+ "max": 2500.0
449
+ },
450
+ {
451
+ "current": 1488.823,
452
+ "min": 1500.0,
453
+ "max": 2500.0
454
+ },
455
+ {
456
+ "current": 1487.087,
457
+ "min": 1500.0,
458
+ "max": 2500.0
459
+ },
460
+ {
461
+ "current": 1487.723,
462
+ "min": 1500.0,
463
+ "max": 2500.0
464
+ },
465
+ {
466
+ "current": 1488.362,
467
+ "min": 1500.0,
468
+ "max": 2500.0
469
+ },
470
+ {
471
+ "current": 3349.924,
472
+ "min": 1500.0,
473
+ "max": 2500.0
474
+ },
475
+ {
476
+ "current": 1487.337,
477
+ "min": 1500.0,
478
+ "max": 2500.0
479
+ },
480
+ {
481
+ "current": 1499.287,
482
+ "min": 1500.0,
483
+ "max": 2500.0
484
+ },
485
+ {
486
+ "current": 1499.769,
487
+ "min": 1500.0,
488
+ "max": 2500.0
489
+ },
490
+ {
491
+ "current": 1497.729,
492
+ "min": 1500.0,
493
+ "max": 2500.0
494
+ },
495
+ {
496
+ "current": 1498.076,
497
+ "min": 1500.0,
498
+ "max": 2500.0
499
+ },
500
+ {
501
+ "current": 1499.303,
502
+ "min": 1500.0,
503
+ "max": 2500.0
504
+ },
505
+ {
506
+ "current": 1499.284,
507
+ "min": 1500.0,
508
+ "max": 2500.0
509
+ },
510
+ {
511
+ "current": 1498.664,
512
+ "min": 1500.0,
513
+ "max": 2500.0
514
+ },
515
+ {
516
+ "current": 1499.834,
517
+ "min": 1500.0,
518
+ "max": 2500.0
519
+ },
520
+ {
521
+ "current": 2357.762,
522
+ "min": 1500.0,
523
+ "max": 2500.0
524
+ },
525
+ {
526
+ "current": 1498.952,
527
+ "min": 1500.0,
528
+ "max": 2500.0
529
+ },
530
+ {
531
+ "current": 1493.673,
532
+ "min": 1500.0,
533
+ "max": 2500.0
534
+ },
535
+ {
536
+ "current": 1493.531,
537
+ "min": 1500.0,
538
+ "max": 2500.0
539
+ },
540
+ {
541
+ "current": 1498.266,
542
+ "min": 1500.0,
543
+ "max": 2500.0
544
+ },
545
+ {
546
+ "current": 1499.609,
547
+ "min": 1500.0,
548
+ "max": 2500.0
549
+ },
550
+ {
551
+ "current": 1498.552,
552
+ "min": 1500.0,
553
+ "max": 2500.0
554
+ },
555
+ {
556
+ "current": 1499.463,
557
+ "min": 1500.0,
558
+ "max": 2500.0
559
+ },
560
+ {
561
+ "current": 1499.305,
562
+ "min": 1500.0,
563
+ "max": 2500.0
564
+ },
565
+ {
566
+ "current": 2425.015,
567
+ "min": 1500.0,
568
+ "max": 2500.0
569
+ },
570
+ {
571
+ "current": 1492.407,
572
+ "min": 1500.0,
573
+ "max": 2500.0
574
+ },
575
+ {
576
+ "current": 1498.813,
577
+ "min": 1500.0,
578
+ "max": 2500.0
579
+ },
580
+ {
581
+ "current": 1465.671,
582
+ "min": 1500.0,
583
+ "max": 2500.0
584
+ },
585
+ {
586
+ "current": 1465.829,
587
+ "min": 1500.0,
588
+ "max": 2500.0
589
+ },
590
+ {
591
+ "current": 2202.008,
592
+ "min": 1500.0,
593
+ "max": 2500.0
594
+ },
595
+ {
596
+ "current": 1466.306,
597
+ "min": 1500.0,
598
+ "max": 2500.0
599
+ },
600
+ {
601
+ "current": 3340.321,
602
+ "min": 1500.0,
603
+ "max": 2500.0
604
+ },
605
+ {
606
+ "current": 3348.163,
607
+ "min": 1500.0,
608
+ "max": 2500.0
609
+ },
610
+ {
611
+ "current": 1488.452,
612
+ "min": 1500.0,
613
+ "max": 2500.0
614
+ },
615
+ {
616
+ "current": 1488.348,
617
+ "min": 1500.0,
618
+ "max": 2500.0
619
+ },
620
+ {
621
+ "current": 1482.09,
622
+ "min": 1500.0,
623
+ "max": 2500.0
624
+ },
625
+ {
626
+ "current": 1499.852,
627
+ "min": 1500.0,
628
+ "max": 2500.0
629
+ },
630
+ {
631
+ "current": 1498.955,
632
+ "min": 1500.0,
633
+ "max": 2500.0
634
+ },
635
+ {
636
+ "current": 1499.312,
637
+ "min": 1500.0,
638
+ "max": 2500.0
639
+ },
640
+ {
641
+ "current": 3352.095,
642
+ "min": 1500.0,
643
+ "max": 2500.0
644
+ },
645
+ {
646
+ "current": 1487.438,
647
+ "min": 1500.0,
648
+ "max": 2500.0
649
+ },
650
+ {
651
+ "current": 1486.989,
652
+ "min": 1500.0,
653
+ "max": 2500.0
654
+ },
655
+ {
656
+ "current": 1488.066,
657
+ "min": 1500.0,
658
+ "max": 2500.0
659
+ },
660
+ {
661
+ "current": 1498.481,
662
+ "min": 1500.0,
663
+ "max": 2500.0
664
+ },
665
+ {
666
+ "current": 1499.505,
667
+ "min": 1500.0,
668
+ "max": 2500.0
669
+ },
670
+ {
671
+ "current": 1499.061,
672
+ "min": 1500.0,
673
+ "max": 2500.0
674
+ },
675
+ {
676
+ "current": 1498.834,
677
+ "min": 1500.0,
678
+ "max": 2500.0
679
+ }
680
+ ],
681
+ "disk": {
682
+ "total": 100.0,
683
+ "used": 27.925994873046875
684
+ },
685
+ "gpu": "NVIDIA A100 80GB PCIe",
686
+ "gpu_count": 4,
687
+ "gpu_devices": [
688
+ {
689
+ "name": "NVIDIA A100 80GB PCIe",
690
+ "memory_total": 85899345920
691
+ },
692
+ {
693
+ "name": "NVIDIA A100 80GB PCIe",
694
+ "memory_total": 85899345920
695
+ },
696
+ {
697
+ "name": "NVIDIA A100 80GB PCIe",
698
+ "memory_total": 85899345920
699
+ },
700
+ {
701
+ "name": "NVIDIA A100 80GB PCIe",
702
+ "memory_total": 85899345920
703
+ }
704
+ ],
705
+ "memory": {
706
+ "total": 1007.7458572387695
707
+ }
708
+ }
wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug-internal.log ADDED
@@ -0,0 +1,169 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-03-28 17:36:01,725 INFO StreamThr :3984 [internal.py:wandb_internal():87] W&B internal server running at pid: 3984, started at: 2023-03-28 17:36:01.724535
2
+ 2023-03-28 17:36:01,729 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status
3
+ 2023-03-28 17:36:01,731 INFO WriterThread:3984 [datastore.py:open_for_write():85] open: /root/alpaca-lora/wandb/offline-run-20230328_173601-8ou0tfhe/run-8ou0tfhe.wandb
4
+ 2023-03-28 17:36:01,733 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: run_start
5
+ 2023-03-28 17:36:01,744 DEBUG HandlerThread:3984 [system_info.py:__init__():31] System info init
6
+ 2023-03-28 17:36:01,744 DEBUG HandlerThread:3984 [system_info.py:__init__():46] System info init done
7
+ 2023-03-28 17:36:01,745 INFO HandlerThread:3984 [system_monitor.py:start():183] Starting system monitor
8
+ 2023-03-28 17:36:01,745 INFO SystemMonitor:3984 [system_monitor.py:_start():147] Starting system asset monitoring threads
9
+ 2023-03-28 17:36:01,745 INFO HandlerThread:3984 [system_monitor.py:probe():204] Collecting system info
10
+ 2023-03-28 17:36:01,746 INFO SystemMonitor:3984 [interfaces.py:start():187] Started cpu monitoring
11
+ 2023-03-28 17:36:01,748 INFO SystemMonitor:3984 [interfaces.py:start():187] Started disk monitoring
12
+ 2023-03-28 17:36:01,749 INFO SystemMonitor:3984 [interfaces.py:start():187] Started gpu monitoring
13
+ 2023-03-28 17:36:01,751 INFO SystemMonitor:3984 [interfaces.py:start():187] Started memory monitoring
14
+ 2023-03-28 17:36:01,760 INFO SystemMonitor:3984 [interfaces.py:start():187] Started network monitoring
15
+ 2023-03-28 17:36:01,807 DEBUG HandlerThread:3984 [system_info.py:probe():195] Probing system
16
+ 2023-03-28 17:36:01,810 DEBUG HandlerThread:3984 [system_info.py:_probe_git():180] Probing git
17
+ 2023-03-28 17:36:01,818 DEBUG HandlerThread:3984 [system_info.py:_probe_git():188] Probing git done
18
+ 2023-03-28 17:36:01,818 DEBUG HandlerThread:3984 [system_info.py:probe():240] Probing system done
19
+ 2023-03-28 17:36:01,818 DEBUG HandlerThread:3984 [system_monitor.py:probe():213] {'os': 'Linux-5.4.0-113-generic-x86_64-with-glibc2.31', 'python': '3.10.10', 'heartbeatAt': '2023-03-28T17:36:01.807653', 'startedAt': '2023-03-28T17:36:01.715893', 'docker': None, 'cuda': None, 'args': ('--base_model', '/root/llama-7b-hf/', '--data_path', './alpaca_data_cleaned.json', '--output_dir', './lora-alpaca', '--batch_size', '1024', '--micro_batch_size', '128'), 'state': 'running', 'program': '/root/alpaca-lora/finetune.py', 'codePath': 'finetune.py', 'git': {'remote': 'https://github.com/tloen/alpaca-lora.git', 'commit': '345c8fbb7bec0eb05e81c13693d06465287a7e3d'}, 'email': None, 'root': '/root/alpaca-lora', 'host': 'c35059eec1c7', 'username': 'root', 'executable': '/usr/bin/python', 'cpu_count': 64, 'cpu_count_logical': 128, 'cpu_freq': {'current': 1668.3349843750002, 'min': 1500.0, 'max': 2500.0}, 'cpu_freq_per_core': [{'current': 1499.835, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.68, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.109, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.126, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.324, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.181, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.484, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.859, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.666, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.584, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.174, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.429, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.836, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.832, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.39, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.6, 'min': 1500.0, 'max': 2500.0}, {'current': 3346.562, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.085, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.039, 'min': 1500.0, 'max': 2500.0}, {'current': 1486.592, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.74, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.482, 'min': 1500.0, 'max': 2500.0}, {'current': 3348.706, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.13, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.984, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.395, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.979, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.956, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.173, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.148, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.279, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.542, 'min': 1500.0, 'max': 2500.0}, {'current': 1494.104, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.647, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.189, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.489, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.785, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.44, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.902, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.876, 'min': 1500.0, 'max': 2500.0}, {'current': 1496.697, 'min': 1500.0, 'max': 2500.0}, {'current': 1799.225, 'min': 1500.0, 'max': 2500.0}, {'current': 1490.788, 'min': 1500.0, 'max': 2500.0}, {'current': 1494.987, 'min': 1500.0, 'max': 2500.0}, {'current': 1465.434, 'min': 1500.0, 'max': 2500.0}, {'current': 1465.45, 'min': 1500.0, 'max': 2500.0}, {'current': 2182.568, 'min': 1500.0, 'max': 2500.0}, {'current': 1465.131, 'min': 1500.0, 'max': 2500.0}, {'current': 3347.661, 'min': 1500.0, 'max': 2500.0}, {'current': 3332.143, 'min': 1500.0, 'max': 2500.0}, {'current': 1484.646, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.67, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.77, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.811, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.561, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.909, 'min': 1500.0, 'max': 2500.0}, {'current': 3340.022, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.267, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.782, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.692, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.424, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.438, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.609, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.979, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.002, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.842, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.564, 'min': 1500.0, 'max': 2500.0}, {'current': 1500.207, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.855, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.791, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.935, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.886, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.045, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.729, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.851, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.735, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.435, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.651, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.641, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.522, 'min': 1500.0, 'max': 2500.0}, {'current': 3350.427, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.653, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.823, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.087, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.723, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.362, 'min': 1500.0, 'max': 2500.0}, {'current': 3349.924, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.337, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.287, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.769, 'min': 1500.0, 'max': 2500.0}, {'current': 1497.729, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.076, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.303, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.284, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.664, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.834, 'min': 1500.0, 'max': 2500.0}, {'current': 2357.762, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.952, 'min': 1500.0, 'max': 2500.0}, {'current': 1493.673, 'min': 1500.0, 'max': 2500.0}, {'current': 1493.531, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.266, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.609, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.552, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.463, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.305, 'min': 1500.0, 'max': 2500.0}, {'current': 2425.015, 'min': 1500.0, 'max': 2500.0}, {'current': 1492.407, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.813, 'min': 1500.0, 'max': 2500.0}, {'current': 1465.671, 'min': 1500.0, 'max': 2500.0}, {'current': 1465.829, 'min': 1500.0, 'max': 2500.0}, {'current': 2202.008, 'min': 1500.0, 'max': 2500.0}, {'current': 1466.306, 'min': 1500.0, 'max': 2500.0}, {'current': 3340.321, 'min': 1500.0, 'max': 2500.0}, {'current': 3348.163, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.452, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.348, 'min': 1500.0, 'max': 2500.0}, {'current': 1482.09, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.852, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.955, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.312, 'min': 1500.0, 'max': 2500.0}, {'current': 3352.095, 'min': 1500.0, 'max': 2500.0}, {'current': 1487.438, 'min': 1500.0, 'max': 2500.0}, {'current': 1486.989, 'min': 1500.0, 'max': 2500.0}, {'current': 1488.066, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.481, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.505, 'min': 1500.0, 'max': 2500.0}, {'current': 1499.061, 'min': 1500.0, 'max': 2500.0}, {'current': 1498.834, 'min': 1500.0, 'max': 2500.0}], 'disk': {'total': 100.0, 'used': 27.925994873046875}, 'gpu': 'NVIDIA A100 80GB PCIe', 'gpu_count': 4, 'gpu_devices': [{'name': 'NVIDIA A100 80GB PCIe', 'memory_total': 85899345920}, {'name': 'NVIDIA A100 80GB PCIe', 'memory_total': 85899345920}, {'name': 'NVIDIA A100 80GB PCIe', 'memory_total': 85899345920}, {'name': 'NVIDIA A100 80GB PCIe', 'memory_total': 85899345920}], 'memory': {'total': 1007.7458572387695}}
20
+ 2023-03-28 17:36:01,819 INFO HandlerThread:3984 [system_monitor.py:probe():214] Finished collecting system info
21
+ 2023-03-28 17:36:01,819 INFO HandlerThread:3984 [system_monitor.py:probe():217] Publishing system info
22
+ 2023-03-28 17:36:01,819 DEBUG HandlerThread:3984 [system_info.py:_save_pip():51] Saving list of pip packages installed into the current environment
23
+ 2023-03-28 17:36:01,819 DEBUG HandlerThread:3984 [system_info.py:_save_pip():67] Saving pip packages done
24
+ 2023-03-28 17:36:01,821 INFO HandlerThread:3984 [system_monitor.py:probe():219] Finished publishing system info
25
+ 2023-03-28 17:36:06,731 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
26
+ 2023-03-28 17:36:06,731 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
27
+ 2023-03-28 17:36:11,732 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
28
+ 2023-03-28 17:36:11,733 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
29
+ 2023-03-28 17:36:16,734 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
30
+ 2023-03-28 17:36:16,734 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
31
+ 2023-03-28 17:36:21,735 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
32
+ 2023-03-28 17:36:21,736 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
33
+ 2023-03-28 17:36:26,737 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
34
+ 2023-03-28 17:36:26,738 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
35
+ 2023-03-28 17:36:31,739 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
36
+ 2023-03-28 17:36:31,740 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
37
+ 2023-03-28 17:36:36,741 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
38
+ 2023-03-28 17:36:36,742 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
39
+ 2023-03-28 17:36:41,743 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
40
+ 2023-03-28 17:36:41,743 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
41
+ 2023-03-28 17:36:46,745 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
42
+ 2023-03-28 17:36:46,745 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
43
+ 2023-03-28 17:36:51,747 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
44
+ 2023-03-28 17:36:51,748 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
45
+ 2023-03-28 17:36:56,749 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
46
+ 2023-03-28 17:36:56,749 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
47
+ 2023-03-28 17:37:01,750 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
48
+ 2023-03-28 17:37:01,751 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
49
+ 2023-03-28 17:37:01,760 DEBUG SystemMonitor:3984 [system_monitor.py:_start():161] Starting system metrics aggregation loop
50
+ 2023-03-28 17:37:06,752 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
51
+ 2023-03-28 17:37:06,753 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
52
+ 2023-03-28 17:37:11,754 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
53
+ 2023-03-28 17:37:11,754 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
54
+ 2023-03-28 17:37:16,755 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
55
+ 2023-03-28 17:37:16,756 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
56
+ 2023-03-28 17:37:21,757 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
57
+ 2023-03-28 17:37:21,758 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
58
+ 2023-03-28 17:37:26,759 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
59
+ 2023-03-28 17:37:26,759 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
60
+ 2023-03-28 17:37:31,760 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
61
+ 2023-03-28 17:37:31,760 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
62
+ 2023-03-28 17:37:36,762 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
63
+ 2023-03-28 17:37:36,762 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
64
+ 2023-03-28 17:37:41,763 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
65
+ 2023-03-28 17:37:41,764 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
66
+ 2023-03-28 17:37:46,765 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
67
+ 2023-03-28 17:37:46,765 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
68
+ 2023-03-28 17:37:51,766 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
69
+ 2023-03-28 17:37:51,766 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
70
+ 2023-03-28 17:37:56,767 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
71
+ 2023-03-28 17:37:56,768 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
72
+ 2023-03-28 17:38:01,770 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
73
+ 2023-03-28 17:38:01,770 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
74
+ 2023-03-28 17:38:06,771 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
75
+ 2023-03-28 17:38:06,772 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
76
+ 2023-03-28 17:38:11,773 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
77
+ 2023-03-28 17:38:11,774 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
78
+ 2023-03-28 17:38:16,775 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
79
+ 2023-03-28 17:38:16,776 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
80
+ 2023-03-28 17:38:21,777 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
81
+ 2023-03-28 17:38:21,777 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
82
+ 2023-03-28 17:38:26,778 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
83
+ 2023-03-28 17:38:26,779 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
84
+ 2023-03-28 17:38:31,780 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
85
+ 2023-03-28 17:38:31,781 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
86
+ 2023-03-28 17:38:36,782 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
87
+ 2023-03-28 17:38:36,783 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
88
+ 2023-03-28 17:38:41,784 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
89
+ 2023-03-28 17:38:41,784 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
90
+ 2023-03-28 17:38:46,785 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
91
+ 2023-03-28 17:38:46,786 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
92
+ 2023-03-28 17:38:51,787 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
93
+ 2023-03-28 17:38:51,787 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
94
+ 2023-03-28 17:38:56,788 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
95
+ 2023-03-28 17:38:56,789 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
96
+ 2023-03-28 17:39:01,789 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
97
+ 2023-03-28 17:39:01,790 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
98
+ 2023-03-28 17:39:06,791 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
99
+ 2023-03-28 17:39:06,791 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
100
+ 2023-03-28 17:39:11,793 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
101
+ 2023-03-28 17:39:11,793 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
102
+ 2023-03-28 17:39:16,794 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
103
+ 2023-03-28 17:39:16,795 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
104
+ 2023-03-28 17:39:21,799 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
105
+ 2023-03-28 17:39:21,800 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
106
+ 2023-03-28 17:39:26,801 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
107
+ 2023-03-28 17:39:26,801 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
108
+ 2023-03-28 17:39:31,803 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
109
+ 2023-03-28 17:39:31,803 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
110
+ 2023-03-28 17:39:36,805 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
111
+ 2023-03-28 17:39:36,805 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
112
+ 2023-03-28 17:39:41,327 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: partial_history
113
+ 2023-03-28 17:39:41,806 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
114
+ 2023-03-28 17:39:41,807 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
115
+ 2023-03-28 17:39:46,808 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
116
+ 2023-03-28 17:39:46,809 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
117
+ 2023-03-28 17:39:51,810 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
118
+ 2023-03-28 17:39:51,810 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
119
+ 2023-03-28 17:39:56,811 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
120
+ 2023-03-28 17:39:56,812 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
121
+ 2023-03-28 17:40:01,813 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
122
+ 2023-03-28 17:40:01,813 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
123
+ 2023-03-28 17:40:06,815 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
124
+ 2023-03-28 17:40:06,815 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
125
+ 2023-03-28 17:40:11,817 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
126
+ 2023-03-28 17:40:11,817 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
127
+ 2023-03-28 17:40:16,818 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
128
+ 2023-03-28 17:40:16,819 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
129
+ 2023-03-28 17:40:21,820 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
130
+ 2023-03-28 17:40:21,821 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
131
+ 2023-03-28 17:40:26,822 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
132
+ 2023-03-28 17:40:26,822 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
133
+ 2023-03-28 17:40:31,824 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
134
+ 2023-03-28 17:40:31,824 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
135
+ 2023-03-28 17:40:36,825 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
136
+ 2023-03-28 17:40:36,826 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
137
+ 2023-03-28 17:40:41,827 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
138
+ 2023-03-28 17:40:41,828 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
139
+ 2023-03-28 17:40:46,829 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
140
+ 2023-03-28 17:40:46,829 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
141
+ 2023-03-28 17:40:51,830 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
142
+ 2023-03-28 17:40:51,831 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
143
+ 2023-03-28 17:40:56,832 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
144
+ 2023-03-28 17:40:56,833 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
145
+ 2023-03-28 17:41:01,834 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
146
+ 2023-03-28 17:41:01,835 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
147
+ 2023-03-28 17:41:06,836 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
148
+ 2023-03-28 17:41:06,836 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
149
+ 2023-03-28 17:41:11,837 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
150
+ 2023-03-28 17:41:11,837 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
151
+ 2023-03-28 17:41:16,838 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
152
+ 2023-03-28 17:41:16,839 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
153
+ 2023-03-28 17:41:21,840 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
154
+ 2023-03-28 17:41:21,840 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
155
+ 2023-03-28 17:41:26,841 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
156
+ 2023-03-28 17:41:26,842 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
157
+ 2023-03-28 17:41:31,843 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
158
+ 2023-03-28 17:41:31,843 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
159
+ 2023-03-28 17:41:36,844 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
160
+ 2023-03-28 17:41:36,844 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
161
+ 2023-03-28 17:41:40,407 ERROR cpu :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
162
+ 2023-03-28 17:41:40,410 ERROR cpu :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
163
+ 2023-03-28 17:41:41,845 DEBUG HandlerThread:3984 [handler.py:handle_request():144] handle_request: status_report
164
+ 2023-03-28 17:41:41,846 DEBUG SenderThread:3984 [sender.py:send_request():363] send_request: status_report
165
+ 2023-03-28 17:41:41,860 ERROR memory :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
166
+ 2023-03-28 17:41:41,860 ERROR memory :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
167
+ 2023-03-28 17:41:42,410 ERROR cpu :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
168
+ 2023-03-28 17:41:42,412 ERROR cpu :3984 [interfaces.py:monitor():141] Failed to sample metric: process no longer exists (pid=3898)
169
+ 2023-03-28 17:41:42,823 INFO MainThread:3984 [internal.py:handle_exit():77] Internal process exited
wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug.log ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-03-28 17:36:01,717 INFO MainThread:3898 [wandb_setup.py:_flush():76] Configure stats pid to 3898
2
+ 2023-03-28 17:36:01,717 INFO MainThread:3898 [wandb_setup.py:_flush():76] Loading settings from /root/.config/wandb/settings
3
+ 2023-03-28 17:36:01,717 INFO MainThread:3898 [wandb_setup.py:_flush():76] Loading settings from /root/alpaca-lora/wandb/settings
4
+ 2023-03-28 17:36:01,717 INFO MainThread:3898 [wandb_setup.py:_flush():76] Loading settings from environment variables: {}
5
+ 2023-03-28 17:36:01,717 INFO MainThread:3898 [wandb_setup.py:_flush():76] Applying setup settings: {'_disable_service': False}
6
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_setup.py:_flush():76] Inferring run settings from compute environment: {'program_relpath': 'finetune.py', 'program': '/root/alpaca-lora/finetune.py'}
7
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_setup.py:_flush():76] Applying login settings: {'mode': 'offline'}
8
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:_log_setup():506] Logging user logs to /root/alpaca-lora/wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug.log
9
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:_log_setup():507] Logging internal logs to /root/alpaca-lora/wandb/offline-run-20230328_173601-8ou0tfhe/logs/debug-internal.log
10
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:init():546] calling init triggers
11
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:init():552] wandb.init called with sweep_config: {}
12
+ config: {}
13
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:init():602] starting backend
14
+ 2023-03-28 17:36:01,718 INFO MainThread:3898 [wandb_init.py:init():606] setting up manager
15
+ 2023-03-28 17:36:01,720 INFO MainThread:3898 [backend.py:_multiprocessing_setup():106] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
16
+ 2023-03-28 17:36:01,721 INFO MainThread:3898 [wandb_init.py:init():613] backend started and connected
17
+ 2023-03-28 17:36:01,724 INFO MainThread:3898 [wandb_init.py:init():701] updated telemetry
18
+ 2023-03-28 17:36:01,730 INFO MainThread:3898 [wandb_init.py:init():789] starting run threads in backend
19
+ 2023-03-28 17:36:01,825 INFO MainThread:3898 [wandb_run.py:_console_start():2114] atexit reg
20
+ 2023-03-28 17:36:01,825 INFO MainThread:3898 [wandb_run.py:_redirect():1969] redirect: SettingsConsole.WRAP_RAW
21
+ 2023-03-28 17:36:01,825 INFO MainThread:3898 [wandb_run.py:_redirect():2034] Wrapping output streams.
22
+ 2023-03-28 17:36:01,825 INFO MainThread:3898 [wandb_run.py:_redirect():2059] Redirects installed.
23
+ 2023-03-28 17:36:01,826 INFO MainThread:3898 [wandb_init.py:init():831] run started, returning control to user process
24
+ 2023-03-28 17:36:01,830 INFO MainThread:3898 [wandb_run.py:_config_callback():1251] config_cb None None {'vocab_size': 32000, 'hidden_size': 4096, 'intermediate_size': 11008, 'num_hidden_layers': 32, 'num_attention_heads': 32, 'hidden_act': 'silu', 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'return_dict': True, 'output_hidden_states': False, 'output_attentions': False, 'torchscript': False, 'torch_dtype': 'float16', 'use_bfloat16': False, 'tf_legacy_loss': False, 'pruned_heads': {}, 'tie_word_embeddings': False, 'is_encoder_decoder': False, 'is_decoder': False, 'cross_attention_hidden_size': None, 'add_cross_attention': False, 'tie_encoder_decoder': False, 'max_length': 20, 'min_length': 0, 'do_sample': False, 'early_stopping': False, 'num_beams': 1, 'num_beam_groups': 1, 'diversity_penalty': 0.0, 'temperature': 1.0, 'top_k': 50, 'top_p': 1.0, 'typical_p': 1.0, 'repetition_penalty': 1.0, 'length_penalty': 1.0, 'no_repeat_ngram_size': 0, 'encoder_no_repeat_ngram_size': 0, 'bad_words_ids': None, 'num_return_sequences': 1, 'chunk_size_feed_forward': 0, 'output_scores': False, 'return_dict_in_generate': False, 'forced_bos_token_id': None, 'forced_eos_token_id': None, 'remove_invalid_values': False, 'exponential_decay_length_penalty': None, 'suppress_tokens': None, 'begin_suppress_tokens': None, 'architectures': ['LlamaForCausalLM'], 'finetuning_task': None, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'tokenizer_class': None, 'prefix': None, 'bos_token_id': 1, 'pad_token_id': 0, 'eos_token_id': 2, 'sep_token_id': None, 'decoder_start_token_id': None, 'task_specific_params': None, 'problem_type': None, '_name_or_path': '/root/llama-7b-hf/', 'transformers_version': '4.28.0.dev0', 'model_type': 'llama', 'output_dir': './lora-alpaca', 'overwrite_output_dir': False, 'do_train': False, 'do_eval': True, 'do_predict': False, 'evaluation_strategy': 'steps', 'prediction_loss_only': False, 'per_device_train_batch_size': 128, 'per_device_eval_batch_size': 8, 'per_gpu_train_batch_size': 'None', 'per_gpu_eval_batch_size': 'None', 'gradient_accumulation_steps': 2, 'eval_accumulation_steps': 'None', 'eval_delay': 0, 'learning_rate': 0.0003, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'max_grad_norm': 1.0, 'num_train_epochs': 3, 'max_steps': -1, 'lr_scheduler_type': 'linear', 'warmup_ratio': 0.0, 'warmup_steps': 100, 'log_level': 'passive', 'log_level_replica': 'warning', 'log_on_each_node': True, 'logging_dir': './lora-alpaca/runs/Mar28_17-33-01_c35059eec1c7', 'logging_strategy': 'steps', 'logging_first_step': False, 'logging_steps': 10, 'logging_nan_inf_filter': True, 'save_strategy': 'steps', 'save_steps': 200, 'save_total_limit': 3, 'save_on_each_node': False, 'no_cuda': False, 'use_mps_device': False, 'seed': 42, 'data_seed': 'None', 'jit_mode_eval': False, 'use_ipex': False, 'bf16': False, 'fp16': True, 'fp16_opt_level': 'O1', 'half_precision_backend': 'cuda_amp', 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': 'None', 'local_rank': 0, 'xpu_backend': 'None', 'tpu_num_cores': 'None', 'tpu_metrics_debug': False, 'debug': '[]', 'dataloader_drop_last': False, 'eval_steps': 200, 'dataloader_num_workers': 0, 'past_index': -1, 'run_name': './lora-alpaca', 'disable_tqdm': False, 'remove_unused_columns': True, 'label_names': 'None', 'load_best_model_at_end': True, 'metric_for_best_model': 'loss', 'greater_is_better': False, 'ignore_data_skip': False, 'sharded_ddp': '[]', 'fsdp': '[]', 'fsdp_min_num_params': 0, 'fsdp_config': "{'fsdp_min_num_params': 0, 'xla': False, 'xla_fsdp_grad_ckpt': False}", 'fsdp_transformer_layer_cls_to_wrap': 'None', 'deepspeed': 'None', 'label_smoothing_factor': 0.0, 'optim': 'adamw_torch', 'optim_args': 'None', 'adafactor': False, 'group_by_length': False, 'length_column_name': 'length', 'report_to': "['wandb']", 'ddp_find_unused_parameters': False, 'ddp_bucket_cap_mb': 'None', 'dataloader_pin_memory': True, 'skip_memory_metrics': True, 'use_legacy_prediction_loop': False, 'push_to_hub': False, 'resume_from_checkpoint': 'None', 'hub_model_id': 'None', 'hub_strategy': 'every_save', 'hub_token': '<HUB_TOKEN>', 'hub_private_repo': False, 'gradient_checkpointing': False, 'include_inputs_for_metrics': False, 'fp16_backend': 'auto', 'push_to_hub_model_id': 'None', 'push_to_hub_organization': 'None', 'push_to_hub_token': '<PUSH_TO_HUB_TOKEN>', 'mp_parameters': '', 'auto_find_batch_size': False, 'full_determinism': False, 'torchdynamo': 'None', 'ray_scope': 'last', 'ddp_timeout': 1800, 'torch_compile': False, 'torch_compile_backend': 'None', 'torch_compile_mode': 'None', 'train_batch_size': 128, 'eval_batch_size': 8}
wandb/offline-run-20230328_173601-8ou0tfhe/run-8ou0tfhe.wandb ADDED
Binary file (58.8 kB). View file
 
wandb/run-20230328_174559-mzmf7gh8/files/config.yaml ADDED
@@ -0,0 +1,595 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ wandb_version: 1
2
+
3
+ _name_or_path:
4
+ desc: null
5
+ value: /root/llama-7b-hf/
6
+ _wandb:
7
+ desc: null
8
+ value:
9
+ cli_version: 0.14.0
10
+ framework: huggingface
11
+ huggingface_version: 4.28.0.dev0
12
+ is_jupyter_run: false
13
+ is_kaggle_kernel: false
14
+ m:
15
+ - 1: train/global_step
16
+ 6:
17
+ - 3
18
+ - 1: train/loss
19
+ 5: 1
20
+ 6:
21
+ - 1
22
+ - 1: train/learning_rate
23
+ 5: 1
24
+ 6:
25
+ - 1
26
+ - 1: train/epoch
27
+ 5: 1
28
+ 6:
29
+ - 1
30
+ - 1: train/train_runtime
31
+ 5: 1
32
+ 6:
33
+ - 1
34
+ - 1: train/train_samples_per_second
35
+ 5: 1
36
+ 6:
37
+ - 1
38
+ - 1: train/train_steps_per_second
39
+ 5: 1
40
+ 6:
41
+ - 1
42
+ - 1: train/total_flos
43
+ 5: 1
44
+ 6:
45
+ - 1
46
+ - 1: train/train_loss
47
+ 5: 1
48
+ 6:
49
+ - 1
50
+ python_version: 3.10.10
51
+ start_time: 1680025559.327599
52
+ t:
53
+ 1:
54
+ - 1
55
+ - 11
56
+ - 49
57
+ - 51
58
+ - 55
59
+ - 71
60
+ 2:
61
+ - 1
62
+ - 11
63
+ - 49
64
+ - 51
65
+ - 55
66
+ - 71
67
+ 3:
68
+ - 7
69
+ - 23
70
+ 4: 3.10.10
71
+ 5: 0.14.0
72
+ 6: 4.28.0.dev0
73
+ 8:
74
+ - 5
75
+ adafactor:
76
+ desc: null
77
+ value: false
78
+ adam_beta1:
79
+ desc: null
80
+ value: 0.9
81
+ adam_beta2:
82
+ desc: null
83
+ value: 0.999
84
+ adam_epsilon:
85
+ desc: null
86
+ value: 1.0e-08
87
+ add_cross_attention:
88
+ desc: null
89
+ value: false
90
+ architectures:
91
+ desc: null
92
+ value:
93
+ - LlamaForCausalLM
94
+ auto_find_batch_size:
95
+ desc: null
96
+ value: false
97
+ bad_words_ids:
98
+ desc: null
99
+ value: null
100
+ begin_suppress_tokens:
101
+ desc: null
102
+ value: null
103
+ bf16:
104
+ desc: null
105
+ value: false
106
+ bf16_full_eval:
107
+ desc: null
108
+ value: false
109
+ bos_token_id:
110
+ desc: null
111
+ value: 1
112
+ chunk_size_feed_forward:
113
+ desc: null
114
+ value: 0
115
+ cross_attention_hidden_size:
116
+ desc: null
117
+ value: null
118
+ data_seed:
119
+ desc: null
120
+ value: None
121
+ dataloader_drop_last:
122
+ desc: null
123
+ value: false
124
+ dataloader_num_workers:
125
+ desc: null
126
+ value: 0
127
+ dataloader_pin_memory:
128
+ desc: null
129
+ value: true
130
+ ddp_bucket_cap_mb:
131
+ desc: null
132
+ value: None
133
+ ddp_find_unused_parameters:
134
+ desc: null
135
+ value: false
136
+ ddp_timeout:
137
+ desc: null
138
+ value: 1800
139
+ debug:
140
+ desc: null
141
+ value: '[]'
142
+ decoder_start_token_id:
143
+ desc: null
144
+ value: null
145
+ deepspeed:
146
+ desc: null
147
+ value: None
148
+ disable_tqdm:
149
+ desc: null
150
+ value: false
151
+ diversity_penalty:
152
+ desc: null
153
+ value: 0.0
154
+ do_eval:
155
+ desc: null
156
+ value: true
157
+ do_predict:
158
+ desc: null
159
+ value: false
160
+ do_sample:
161
+ desc: null
162
+ value: false
163
+ do_train:
164
+ desc: null
165
+ value: false
166
+ early_stopping:
167
+ desc: null
168
+ value: false
169
+ encoder_no_repeat_ngram_size:
170
+ desc: null
171
+ value: 0
172
+ eos_token_id:
173
+ desc: null
174
+ value: 2
175
+ eval_accumulation_steps:
176
+ desc: null
177
+ value: None
178
+ eval_batch_size:
179
+ desc: null
180
+ value: 8
181
+ eval_delay:
182
+ desc: null
183
+ value: 0
184
+ eval_steps:
185
+ desc: null
186
+ value: 200
187
+ evaluation_strategy:
188
+ desc: null
189
+ value: steps
190
+ exponential_decay_length_penalty:
191
+ desc: null
192
+ value: null
193
+ finetuning_task:
194
+ desc: null
195
+ value: null
196
+ forced_bos_token_id:
197
+ desc: null
198
+ value: null
199
+ forced_eos_token_id:
200
+ desc: null
201
+ value: null
202
+ fp16:
203
+ desc: null
204
+ value: true
205
+ fp16_backend:
206
+ desc: null
207
+ value: auto
208
+ fp16_full_eval:
209
+ desc: null
210
+ value: false
211
+ fp16_opt_level:
212
+ desc: null
213
+ value: O1
214
+ fsdp:
215
+ desc: null
216
+ value: '[]'
217
+ fsdp_config:
218
+ desc: null
219
+ value: '{''fsdp_min_num_params'': 0, ''xla'': False, ''xla_fsdp_grad_ckpt'': False}'
220
+ fsdp_min_num_params:
221
+ desc: null
222
+ value: 0
223
+ fsdp_transformer_layer_cls_to_wrap:
224
+ desc: null
225
+ value: None
226
+ full_determinism:
227
+ desc: null
228
+ value: false
229
+ gradient_accumulation_steps:
230
+ desc: null
231
+ value: 2
232
+ gradient_checkpointing:
233
+ desc: null
234
+ value: false
235
+ greater_is_better:
236
+ desc: null
237
+ value: false
238
+ group_by_length:
239
+ desc: null
240
+ value: false
241
+ half_precision_backend:
242
+ desc: null
243
+ value: cuda_amp
244
+ hidden_act:
245
+ desc: null
246
+ value: silu
247
+ hidden_size:
248
+ desc: null
249
+ value: 4096
250
+ hub_model_id:
251
+ desc: null
252
+ value: None
253
+ hub_private_repo:
254
+ desc: null
255
+ value: false
256
+ hub_strategy:
257
+ desc: null
258
+ value: every_save
259
+ hub_token:
260
+ desc: null
261
+ value: <HUB_TOKEN>
262
+ id2label:
263
+ desc: null
264
+ value:
265
+ '0': LABEL_0
266
+ '1': LABEL_1
267
+ ignore_data_skip:
268
+ desc: null
269
+ value: false
270
+ include_inputs_for_metrics:
271
+ desc: null
272
+ value: false
273
+ initializer_range:
274
+ desc: null
275
+ value: 0.02
276
+ intermediate_size:
277
+ desc: null
278
+ value: 11008
279
+ is_decoder:
280
+ desc: null
281
+ value: false
282
+ is_encoder_decoder:
283
+ desc: null
284
+ value: false
285
+ jit_mode_eval:
286
+ desc: null
287
+ value: false
288
+ label2id:
289
+ desc: null
290
+ value:
291
+ LABEL_0: 0
292
+ LABEL_1: 1
293
+ label_names:
294
+ desc: null
295
+ value: None
296
+ label_smoothing_factor:
297
+ desc: null
298
+ value: 0.0
299
+ learning_rate:
300
+ desc: null
301
+ value: 0.0003
302
+ length_column_name:
303
+ desc: null
304
+ value: length
305
+ length_penalty:
306
+ desc: null
307
+ value: 1.0
308
+ load_best_model_at_end:
309
+ desc: null
310
+ value: true
311
+ local_rank:
312
+ desc: null
313
+ value: 0
314
+ log_level:
315
+ desc: null
316
+ value: passive
317
+ log_level_replica:
318
+ desc: null
319
+ value: warning
320
+ log_on_each_node:
321
+ desc: null
322
+ value: true
323
+ logging_dir:
324
+ desc: null
325
+ value: ./lora-alpaca/runs/Mar28_17-45-56_c35059eec1c7
326
+ logging_first_step:
327
+ desc: null
328
+ value: false
329
+ logging_nan_inf_filter:
330
+ desc: null
331
+ value: true
332
+ logging_steps:
333
+ desc: null
334
+ value: 10
335
+ logging_strategy:
336
+ desc: null
337
+ value: steps
338
+ lr_scheduler_type:
339
+ desc: null
340
+ value: linear
341
+ max_grad_norm:
342
+ desc: null
343
+ value: 1.0
344
+ max_length:
345
+ desc: null
346
+ value: 20
347
+ max_steps:
348
+ desc: null
349
+ value: -1
350
+ metric_for_best_model:
351
+ desc: null
352
+ value: loss
353
+ min_length:
354
+ desc: null
355
+ value: 0
356
+ model_type:
357
+ desc: null
358
+ value: llama
359
+ mp_parameters:
360
+ desc: null
361
+ value: ''
362
+ no_cuda:
363
+ desc: null
364
+ value: false
365
+ no_repeat_ngram_size:
366
+ desc: null
367
+ value: 0
368
+ num_attention_heads:
369
+ desc: null
370
+ value: 32
371
+ num_beam_groups:
372
+ desc: null
373
+ value: 1
374
+ num_beams:
375
+ desc: null
376
+ value: 1
377
+ num_hidden_layers:
378
+ desc: null
379
+ value: 32
380
+ num_return_sequences:
381
+ desc: null
382
+ value: 1
383
+ num_train_epochs:
384
+ desc: null
385
+ value: 3
386
+ optim:
387
+ desc: null
388
+ value: adamw_torch
389
+ optim_args:
390
+ desc: null
391
+ value: None
392
+ output_attentions:
393
+ desc: null
394
+ value: false
395
+ output_dir:
396
+ desc: null
397
+ value: ./lora-alpaca
398
+ output_hidden_states:
399
+ desc: null
400
+ value: false
401
+ output_scores:
402
+ desc: null
403
+ value: false
404
+ overwrite_output_dir:
405
+ desc: null
406
+ value: false
407
+ pad_token_id:
408
+ desc: null
409
+ value: 0
410
+ past_index:
411
+ desc: null
412
+ value: -1
413
+ per_device_eval_batch_size:
414
+ desc: null
415
+ value: 8
416
+ per_device_train_batch_size:
417
+ desc: null
418
+ value: 128
419
+ per_gpu_eval_batch_size:
420
+ desc: null
421
+ value: None
422
+ per_gpu_train_batch_size:
423
+ desc: null
424
+ value: None
425
+ prediction_loss_only:
426
+ desc: null
427
+ value: false
428
+ prefix:
429
+ desc: null
430
+ value: null
431
+ problem_type:
432
+ desc: null
433
+ value: null
434
+ pruned_heads:
435
+ desc: null
436
+ value: {}
437
+ push_to_hub:
438
+ desc: null
439
+ value: false
440
+ push_to_hub_model_id:
441
+ desc: null
442
+ value: None
443
+ push_to_hub_organization:
444
+ desc: null
445
+ value: None
446
+ push_to_hub_token:
447
+ desc: null
448
+ value: <PUSH_TO_HUB_TOKEN>
449
+ ray_scope:
450
+ desc: null
451
+ value: last
452
+ remove_invalid_values:
453
+ desc: null
454
+ value: false
455
+ remove_unused_columns:
456
+ desc: null
457
+ value: true
458
+ repetition_penalty:
459
+ desc: null
460
+ value: 1.0
461
+ report_to:
462
+ desc: null
463
+ value: '[''wandb'']'
464
+ resume_from_checkpoint:
465
+ desc: null
466
+ value: None
467
+ return_dict:
468
+ desc: null
469
+ value: true
470
+ return_dict_in_generate:
471
+ desc: null
472
+ value: false
473
+ rms_norm_eps:
474
+ desc: null
475
+ value: 1.0e-06
476
+ run_name:
477
+ desc: null
478
+ value: ''
479
+ save_on_each_node:
480
+ desc: null
481
+ value: false
482
+ save_steps:
483
+ desc: null
484
+ value: 200
485
+ save_strategy:
486
+ desc: null
487
+ value: steps
488
+ save_total_limit:
489
+ desc: null
490
+ value: 3
491
+ seed:
492
+ desc: null
493
+ value: 42
494
+ sep_token_id:
495
+ desc: null
496
+ value: null
497
+ sharded_ddp:
498
+ desc: null
499
+ value: '[]'
500
+ skip_memory_metrics:
501
+ desc: null
502
+ value: true
503
+ suppress_tokens:
504
+ desc: null
505
+ value: null
506
+ task_specific_params:
507
+ desc: null
508
+ value: null
509
+ temperature:
510
+ desc: null
511
+ value: 1.0
512
+ tf32:
513
+ desc: null
514
+ value: None
515
+ tf_legacy_loss:
516
+ desc: null
517
+ value: false
518
+ tie_encoder_decoder:
519
+ desc: null
520
+ value: false
521
+ tie_word_embeddings:
522
+ desc: null
523
+ value: false
524
+ tokenizer_class:
525
+ desc: null
526
+ value: null
527
+ top_k:
528
+ desc: null
529
+ value: 50
530
+ top_p:
531
+ desc: null
532
+ value: 1.0
533
+ torch_compile:
534
+ desc: null
535
+ value: false
536
+ torch_compile_backend:
537
+ desc: null
538
+ value: None
539
+ torch_compile_mode:
540
+ desc: null
541
+ value: None
542
+ torch_dtype:
543
+ desc: null
544
+ value: float16
545
+ torchdynamo:
546
+ desc: null
547
+ value: None
548
+ torchscript:
549
+ desc: null
550
+ value: false
551
+ tpu_metrics_debug:
552
+ desc: null
553
+ value: false
554
+ tpu_num_cores:
555
+ desc: null
556
+ value: None
557
+ train_batch_size:
558
+ desc: null
559
+ value: 128
560
+ transformers_version:
561
+ desc: null
562
+ value: 4.28.0.dev0
563
+ typical_p:
564
+ desc: null
565
+ value: 1.0
566
+ use_bfloat16:
567
+ desc: null
568
+ value: false
569
+ use_cache:
570
+ desc: null
571
+ value: false
572
+ use_ipex:
573
+ desc: null
574
+ value: false
575
+ use_legacy_prediction_loop:
576
+ desc: null
577
+ value: false
578
+ use_mps_device:
579
+ desc: null
580
+ value: false
581
+ vocab_size:
582
+ desc: null
583
+ value: 32000
584
+ warmup_ratio:
585
+ desc: null
586
+ value: 0.0
587
+ warmup_steps:
588
+ desc: null
589
+ value: 100
590
+ weight_decay:
591
+ desc: null
592
+ value: 0.0
593
+ xpu_backend:
594
+ desc: null
595
+ value: None
wandb/run-20230328_174559-mzmf7gh8/files/output.log ADDED
@@ -0,0 +1,165 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+
3
+
4
+
5
+
6
+
7
+
8
+
9
+
10
+
11
+
12
+ 7%|██████████████▌ | 10/147 [03:39<49:58, 21.89s/it]
13
+
14
+
15
+
16
+
17
+
18
+
19
+
20
+
21
+
22
+
23
+ 14%|█████████████████████████████ | 20/147 [07:18<46:22, 21.91s/it]
24
+
25
+
26
+
27
+
28
+
29
+
30
+
31
+
32
+
33
+
34
+ 20%|███████████████████████████████████████████▋ | 30/147 [10:57<42:44, 21.92s/it]
35
+
36
+
37
+
38
+
39
+
40
+
41
+
42
+
43
+
44
+
45
+ 27%|██████████████████████████████████████████████████████████▏ | 40/147 [14:36<39:03, 21.91s/it]
46
+
47
+
48
+
49
+
50
+
51
+
52
+
53
+
54
+
55
+
56
+ 34%|████████████████████████████████████████████████████████████████████████▊ | 50/147 [18:07<32:30, 20.10s/it]
57
+
58
+
59
+
60
+
61
+
62
+
63
+
64
+
65
+
66
+
67
+ 41%|███████████████████████████████████████████████████████████████████████████████████████▎ | 60/147 [21:46<31:45, 21.90s/it]
68
+
69
+
70
+
71
+
72
+
73
+
74
+
75
+
76
+
77
+
78
+ 48%|█████████████████████████████████████████████████████████████████████████████████████████████████████▉ | 70/147 [25:25<27:59, 21.81s/it]
79
+
80
+
81
+
82
+
83
+
84
+
85
+
86
+
87
+
88
+
89
+ 54%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▍ | 80/147 [29:03<24:18, 21.77s/it]
90
+
91
+
92
+
93
+
94
+
95
+
96
+
97
+
98
+
99
+
100
+ 61%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████ | 90/147 [32:41<20:41, 21.78s/it]
101
+
102
+
103
+
104
+
105
+
106
+
107
+
108
+
109
+
110
+
111
+ 68%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉ | 100/147 [36:10<16:04, 20.53s/it]
112
+
113
+
114
+
115
+
116
+
117
+
118
+
119
+
120
+
121
+
122
+ 75%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▍ | 110/147 [39:48<13:25, 21.77s/it]
123
+
124
+
125
+
126
+
127
+
128
+
129
+
130
+
131
+
132
+
133
+ 82%|████████████████████████████████████████████��████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▉ | 120/147 [43:26<09:48, 21.78s/it]
134
+
135
+
136
+
137
+
138
+
139
+
140
+
141
+
142
+
143
+
144
+ 88%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▎ | 130/147 [47:04<06:10, 21.78s/it]
145
+
146
+
147
+
148
+
149
+
150
+
151
+
152
+
153
+
154
+
155
+ 95%|██████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████▊ | 140/147 [50:41<02:32, 21.77s/it]
156
+
157
+
158
+
159
+
160
+
161
+
162
+
163
+ 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 147/147 [53:05<00:00, 21.67s/it]
164
+ {'train_runtime': 3187.753, 'train_samples_per_second': 46.828, 'train_steps_per_second': 0.046, 'train_loss': 1.2030427374807344, 'epoch': 3.0}
165
+ If there's a warning about missing keys above, please disregard :)
wandb/run-20230328_174559-mzmf7gh8/files/requirements.txt ADDED
@@ -0,0 +1,186 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ absl-py==1.4.0
2
+ accelerate==0.18.0
3
+ aiofiles==22.1.0
4
+ aiohttp==3.8.4
5
+ aiosignal==1.3.1
6
+ aiosqlite==0.18.0
7
+ altair==4.2.2
8
+ anyio==3.6.2
9
+ appdirs==1.4.4
10
+ argon2-cffi-bindings==21.2.0
11
+ argon2-cffi==21.3.0
12
+ arrow==1.2.3
13
+ asttokens==2.2.1
14
+ async-timeout==4.0.2
15
+ attrs==22.2.0
16
+ babel==2.11.0
17
+ backcall==0.2.0
18
+ beautifulsoup4==4.11.2
19
+ bitsandbytes==0.37.2
20
+ black==23.1.0
21
+ bleach==6.0.0
22
+ certifi==2019.11.28
23
+ cffi==1.15.1
24
+ chardet==3.0.4
25
+ charset-normalizer==3.0.1
26
+ click==8.1.3
27
+ comm==0.1.2
28
+ contourpy==1.0.7
29
+ cycler==0.11.0
30
+ datasets==2.10.1
31
+ dbus-python==1.2.16
32
+ debugpy==1.6.6
33
+ decorator==5.1.1
34
+ defusedxml==0.7.1
35
+ dill==0.3.6
36
+ docker-pycreds==0.4.0
37
+ entrypoints==0.4
38
+ executing==1.2.0
39
+ fastapi==0.95.0
40
+ fastjsonschema==2.16.2
41
+ ffmpy==0.3.0
42
+ filelock==3.10.7
43
+ fire==0.5.0
44
+ fonttools==4.39.3
45
+ fqdn==1.5.1
46
+ frozenlist==1.3.3
47
+ fsspec==2023.3.0
48
+ gitdb==4.0.10
49
+ gitpython==3.1.31
50
+ gradio==3.23.0
51
+ h11==0.14.0
52
+ httpcore==0.16.3
53
+ httpx==0.23.3
54
+ huggingface-hub==0.13.3
55
+ idna==2.8
56
+ ipykernel==6.21.2
57
+ ipython-genutils==0.2.0
58
+ ipython==8.10.0
59
+ ipywidgets==8.0.4
60
+ isoduration==20.11.0
61
+ jedi==0.18.2
62
+ jinja2==3.1.2
63
+ joblib==1.2.0
64
+ json5==0.9.11
65
+ jsonpointer==2.3
66
+ jsonschema==4.17.3
67
+ jupyter-archive==3.3.4
68
+ jupyter-client==8.0.2
69
+ jupyter-core==5.2.0
70
+ jupyter-events==0.5.0
71
+ jupyter-server-fileid==0.6.0
72
+ jupyter-server-terminals==0.4.4
73
+ jupyter-server-ydoc==0.6.1
74
+ jupyter-server==2.3.0
75
+ jupyter-ydoc==0.2.2
76
+ jupyterlab-pygments==0.2.2
77
+ jupyterlab-server==2.19.0
78
+ jupyterlab-widgets==3.0.5
79
+ jupyterlab==3.6.1
80
+ kiwisolver==1.4.4
81
+ linkify-it-py==2.0.0
82
+ markdown-it-py==2.2.0
83
+ markupsafe==2.1.2
84
+ matplotlib-inline==0.1.6
85
+ matplotlib==3.7.1
86
+ mdit-py-plugins==0.3.3
87
+ mdurl==0.1.2
88
+ mistune==2.0.5
89
+ multidict==6.0.4
90
+ multiprocess==0.70.14
91
+ mypy-extensions==1.0.0
92
+ nbclassic==0.5.1
93
+ nbclient==0.7.2
94
+ nbconvert==7.2.9
95
+ nbformat==5.7.3
96
+ nest-asyncio==1.5.6
97
+ nltk==3.8.1
98
+ notebook-shim==0.2.2
99
+ notebook==6.5.2
100
+ numpy==1.24.2
101
+ openai==0.27.2
102
+ orjson==3.8.9
103
+ packaging==23.0
104
+ pandas==1.5.3
105
+ pandocfilters==1.5.0
106
+ parso==0.8.3
107
+ pathspec==0.11.1
108
+ pathtools==0.1.2
109
+ peft==0.3.0.dev0
110
+ pexpect==4.8.0
111
+ pickleshare==0.7.5
112
+ pillow==9.4.0
113
+ pip==23.0.1
114
+ platformdirs==3.0.0
115
+ prometheus-client==0.16.0
116
+ prompt-toolkit==3.0.36
117
+ protobuf==4.22.1
118
+ psutil==5.9.4
119
+ ptyprocess==0.7.0
120
+ pure-eval==0.2.2
121
+ pyarrow==11.0.0
122
+ pycparser==2.21
123
+ pydantic==1.10.7
124
+ pydub==0.25.1
125
+ pygments==2.14.0
126
+ pygobject==3.36.0
127
+ pyparsing==3.0.9
128
+ pyrsistent==0.19.3
129
+ python-apt==2.0.1
130
+ python-dateutil==2.8.2
131
+ python-json-logger==2.0.6
132
+ python-multipart==0.0.6
133
+ pytz==2022.7.1
134
+ pyyaml==6.0
135
+ pyzmq==25.0.0
136
+ regex==2023.3.23
137
+ requests-unixsocket==0.2.0
138
+ requests==2.28.2
139
+ responses==0.18.0
140
+ rfc3339-validator==0.1.4
141
+ rfc3986-validator==0.1.1
142
+ rfc3986==1.5.0
143
+ rouge-score==0.1.2
144
+ semantic-version==2.10.0
145
+ send2trash==1.8.0
146
+ sentencepiece==0.1.97
147
+ sentry-sdk==1.18.0
148
+ setproctitle==1.3.2
149
+ setuptools==67.6.1
150
+ six==1.14.0
151
+ smmap==5.0.0
152
+ sniffio==1.3.0
153
+ soupsieve==2.4
154
+ stack-data==0.6.2
155
+ starlette==0.26.1
156
+ termcolor==2.2.0
157
+ terminado==0.17.1
158
+ tinycss2==1.2.1
159
+ tokenize-rt==5.0.0
160
+ tokenizers==0.12.1
161
+ tomli==2.0.1
162
+ toolz==0.12.0
163
+ torch==1.13.1+cu116
164
+ torchaudio==0.13.1+cu116
165
+ torchvision==0.14.1+cu116
166
+ tornado==6.2
167
+ tqdm==4.65.0
168
+ traitlets==5.9.0
169
+ transformers==4.28.0.dev0
170
+ typing-extensions==4.5.0
171
+ uc-micro-py==1.0.1
172
+ uri-template==1.2.0
173
+ urllib3==1.26.15
174
+ uvicorn==0.21.1
175
+ wandb==0.14.0
176
+ wcwidth==0.2.6
177
+ webcolors==1.12
178
+ webencodings==0.5.1
179
+ websocket-client==1.5.1
180
+ websockets==10.4
181
+ wheel==0.38.4
182
+ widgetsnbextension==4.0.5
183
+ xxhash==3.2.0
184
+ y-py==0.5.5
185
+ yarl==1.8.2
186
+ ypy-websocket==0.8.2
wandb/run-20230328_174559-mzmf7gh8/files/wandb-metadata.json ADDED
@@ -0,0 +1,708 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "os": "Linux-5.4.0-113-generic-x86_64-with-glibc2.31",
3
+ "python": "3.10.10",
4
+ "heartbeatAt": "2023-03-28T17:45:59.806175",
5
+ "startedAt": "2023-03-28T17:45:59.318188",
6
+ "docker": null,
7
+ "cuda": null,
8
+ "args": [
9
+ "--base_model",
10
+ "/root/llama-7b-hf/",
11
+ "--data_path",
12
+ "./alpaca_data_cleaned.json",
13
+ "--output_dir",
14
+ "./lora-alpaca",
15
+ "--batch_size",
16
+ "1024",
17
+ "--micro_batch_size",
18
+ "128"
19
+ ],
20
+ "state": "running",
21
+ "program": "/root/alpaca-lora/finetune.py",
22
+ "codePath": "finetune.py",
23
+ "git": {
24
+ "remote": "https://github.com/tloen/alpaca-lora.git",
25
+ "commit": "345c8fbb7bec0eb05e81c13693d06465287a7e3d"
26
+ },
27
+ "email": null,
28
+ "root": "/root/alpaca-lora",
29
+ "host": "c35059eec1c7",
30
+ "username": "root",
31
+ "executable": "/usr/bin/python",
32
+ "cpu_count": 64,
33
+ "cpu_count_logical": 128,
34
+ "cpu_freq": {
35
+ "current": 1804.408890624999,
36
+ "min": 1500.0,
37
+ "max": 2500.0
38
+ },
39
+ "cpu_freq_per_core": [
40
+ {
41
+ "current": 1450.74,
42
+ "min": 1500.0,
43
+ "max": 2500.0
44
+ },
45
+ {
46
+ "current": 1499.058,
47
+ "min": 1500.0,
48
+ "max": 2500.0
49
+ },
50
+ {
51
+ "current": 1498.815,
52
+ "min": 1500.0,
53
+ "max": 2500.0
54
+ },
55
+ {
56
+ "current": 1798.417,
57
+ "min": 1500.0,
58
+ "max": 2500.0
59
+ },
60
+ {
61
+ "current": 3334.731,
62
+ "min": 1500.0,
63
+ "max": 2500.0
64
+ },
65
+ {
66
+ "current": 1488.036,
67
+ "min": 1500.0,
68
+ "max": 2500.0
69
+ },
70
+ {
71
+ "current": 1486.123,
72
+ "min": 1500.0,
73
+ "max": 2500.0
74
+ },
75
+ {
76
+ "current": 1488.285,
77
+ "min": 1500.0,
78
+ "max": 2500.0
79
+ },
80
+ {
81
+ "current": 1799.92,
82
+ "min": 1500.0,
83
+ "max": 2500.0
84
+ },
85
+ {
86
+ "current": 1800.269,
87
+ "min": 1500.0,
88
+ "max": 2500.0
89
+ },
90
+ {
91
+ "current": 1799.298,
92
+ "min": 1500.0,
93
+ "max": 2500.0
94
+ },
95
+ {
96
+ "current": 1798.883,
97
+ "min": 1500.0,
98
+ "max": 2500.0
99
+ },
100
+ {
101
+ "current": 1798.093,
102
+ "min": 1500.0,
103
+ "max": 2500.0
104
+ },
105
+ {
106
+ "current": 1798.227,
107
+ "min": 1500.0,
108
+ "max": 2500.0
109
+ },
110
+ {
111
+ "current": 1499.322,
112
+ "min": 1500.0,
113
+ "max": 2500.0
114
+ },
115
+ {
116
+ "current": 1499.779,
117
+ "min": 1500.0,
118
+ "max": 2500.0
119
+ },
120
+ {
121
+ "current": 1798.645,
122
+ "min": 1500.0,
123
+ "max": 2500.0
124
+ },
125
+ {
126
+ "current": 1799.544,
127
+ "min": 1500.0,
128
+ "max": 2500.0
129
+ },
130
+ {
131
+ "current": 1798.881,
132
+ "min": 1500.0,
133
+ "max": 2500.0
134
+ },
135
+ {
136
+ "current": 1799.731,
137
+ "min": 1500.0,
138
+ "max": 2500.0
139
+ },
140
+ {
141
+ "current": 2338.39,
142
+ "min": 1500.0,
143
+ "max": 2500.0
144
+ },
145
+ {
146
+ "current": 1489.223,
147
+ "min": 1500.0,
148
+ "max": 2500.0
149
+ },
150
+ {
151
+ "current": 1494.851,
152
+ "min": 1500.0,
153
+ "max": 2500.0
154
+ },
155
+ {
156
+ "current": 1489.069,
157
+ "min": 1500.0,
158
+ "max": 2500.0
159
+ },
160
+ {
161
+ "current": 1666.105,
162
+ "min": 1500.0,
163
+ "max": 2500.0
164
+ },
165
+ {
166
+ "current": 1443.075,
167
+ "min": 1500.0,
168
+ "max": 2500.0
169
+ },
170
+ {
171
+ "current": 1499.572,
172
+ "min": 1500.0,
173
+ "max": 2500.0
174
+ },
175
+ {
176
+ "current": 1800.395,
177
+ "min": 1500.0,
178
+ "max": 2500.0
179
+ },
180
+ {
181
+ "current": 3340.524,
182
+ "min": 1500.0,
183
+ "max": 2500.0
184
+ },
185
+ {
186
+ "current": 3246.902,
187
+ "min": 1500.0,
188
+ "max": 2500.0
189
+ },
190
+ {
191
+ "current": 1673.485,
192
+ "min": 1500.0,
193
+ "max": 2500.0
194
+ },
195
+ {
196
+ "current": 3349.424,
197
+ "min": 1500.0,
198
+ "max": 2500.0
199
+ },
200
+ {
201
+ "current": 1499.191,
202
+ "min": 1500.0,
203
+ "max": 2500.0
204
+ },
205
+ {
206
+ "current": 1498.633,
207
+ "min": 1500.0,
208
+ "max": 2500.0
209
+ },
210
+ {
211
+ "current": 1498.76,
212
+ "min": 1500.0,
213
+ "max": 2500.0
214
+ },
215
+ {
216
+ "current": 1499.294,
217
+ "min": 1500.0,
218
+ "max": 2500.0
219
+ },
220
+ {
221
+ "current": 1798.914,
222
+ "min": 1500.0,
223
+ "max": 2500.0
224
+ },
225
+ {
226
+ "current": 1498.084,
227
+ "min": 1500.0,
228
+ "max": 2500.0
229
+ },
230
+ {
231
+ "current": 1499.238,
232
+ "min": 1500.0,
233
+ "max": 2500.0
234
+ },
235
+ {
236
+ "current": 1799.262,
237
+ "min": 1500.0,
238
+ "max": 2500.0
239
+ },
240
+ {
241
+ "current": 3345.499,
242
+ "min": 1500.0,
243
+ "max": 2500.0
244
+ },
245
+ {
246
+ "current": 3345.563,
247
+ "min": 1500.0,
248
+ "max": 2500.0
249
+ },
250
+ {
251
+ "current": 1673.989,
252
+ "min": 1500.0,
253
+ "max": 2500.0
254
+ },
255
+ {
256
+ "current": 1674.992,
257
+ "min": 1500.0,
258
+ "max": 2500.0
259
+ },
260
+ {
261
+ "current": 3349.952,
262
+ "min": 1500.0,
263
+ "max": 2500.0
264
+ },
265
+ {
266
+ "current": 1488.692,
267
+ "min": 1500.0,
268
+ "max": 2500.0
269
+ },
270
+ {
271
+ "current": 1488.34,
272
+ "min": 1500.0,
273
+ "max": 2500.0
274
+ },
275
+ {
276
+ "current": 1486.725,
277
+ "min": 1500.0,
278
+ "max": 2500.0
279
+ },
280
+ {
281
+ "current": 1498.786,
282
+ "min": 1500.0,
283
+ "max": 2500.0
284
+ },
285
+ {
286
+ "current": 1498.932,
287
+ "min": 1500.0,
288
+ "max": 2500.0
289
+ },
290
+ {
291
+ "current": 1498.896,
292
+ "min": 1500.0,
293
+ "max": 2500.0
294
+ },
295
+ {
296
+ "current": 1498.624,
297
+ "min": 1500.0,
298
+ "max": 2500.0
299
+ },
300
+ {
301
+ "current": 1499.255,
302
+ "min": 1500.0,
303
+ "max": 2500.0
304
+ },
305
+ {
306
+ "current": 1499.346,
307
+ "min": 1500.0,
308
+ "max": 2500.0
309
+ },
310
+ {
311
+ "current": 1798.475,
312
+ "min": 1500.0,
313
+ "max": 2500.0
314
+ },
315
+ {
316
+ "current": 1499.345,
317
+ "min": 1500.0,
318
+ "max": 2500.0
319
+ },
320
+ {
321
+ "current": 1498.571,
322
+ "min": 1500.0,
323
+ "max": 2500.0
324
+ },
325
+ {
326
+ "current": 1498.301,
327
+ "min": 1500.0,
328
+ "max": 2500.0
329
+ },
330
+ {
331
+ "current": 1799.514,
332
+ "min": 1500.0,
333
+ "max": 2500.0
334
+ },
335
+ {
336
+ "current": 1799.265,
337
+ "min": 1500.0,
338
+ "max": 2500.0
339
+ },
340
+ {
341
+ "current": 1486.883,
342
+ "min": 1500.0,
343
+ "max": 2500.0
344
+ },
345
+ {
346
+ "current": 3343.41,
347
+ "min": 1500.0,
348
+ "max": 2500.0
349
+ },
350
+ {
351
+ "current": 1673.837,
352
+ "min": 1500.0,
353
+ "max": 2500.0
354
+ },
355
+ {
356
+ "current": 1673.336,
357
+ "min": 1500.0,
358
+ "max": 2500.0
359
+ },
360
+ {
361
+ "current": 1487.121,
362
+ "min": 1500.0,
363
+ "max": 2500.0
364
+ },
365
+ {
366
+ "current": 1486.946,
367
+ "min": 1500.0,
368
+ "max": 2500.0
369
+ },
370
+ {
371
+ "current": 1488.835,
372
+ "min": 1500.0,
373
+ "max": 2500.0
374
+ },
375
+ {
376
+ "current": 2690.15,
377
+ "min": 1500.0,
378
+ "max": 2500.0
379
+ },
380
+ {
381
+ "current": 3351.673,
382
+ "min": 1500.0,
383
+ "max": 2500.0
384
+ },
385
+ {
386
+ "current": 1487.586,
387
+ "min": 1500.0,
388
+ "max": 2500.0
389
+ },
390
+ {
391
+ "current": 1487.318,
392
+ "min": 1500.0,
393
+ "max": 2500.0
394
+ },
395
+ {
396
+ "current": 1488.857,
397
+ "min": 1500.0,
398
+ "max": 2500.0
399
+ },
400
+ {
401
+ "current": 3348.481,
402
+ "min": 1500.0,
403
+ "max": 2500.0
404
+ },
405
+ {
406
+ "current": 3350.093,
407
+ "min": 1500.0,
408
+ "max": 2500.0
409
+ },
410
+ {
411
+ "current": 3349.644,
412
+ "min": 1500.0,
413
+ "max": 2500.0
414
+ },
415
+ {
416
+ "current": 3350.818,
417
+ "min": 1500.0,
418
+ "max": 2500.0
419
+ },
420
+ {
421
+ "current": 3349.572,
422
+ "min": 1500.0,
423
+ "max": 2500.0
424
+ },
425
+ {
426
+ "current": 3345.53,
427
+ "min": 1500.0,
428
+ "max": 2500.0
429
+ },
430
+ {
431
+ "current": 1488.753,
432
+ "min": 1500.0,
433
+ "max": 2500.0
434
+ },
435
+ {
436
+ "current": 1488.811,
437
+ "min": 1500.0,
438
+ "max": 2500.0
439
+ },
440
+ {
441
+ "current": 3346.66,
442
+ "min": 1500.0,
443
+ "max": 2500.0
444
+ },
445
+ {
446
+ "current": 2576.897,
447
+ "min": 1500.0,
448
+ "max": 2500.0
449
+ },
450
+ {
451
+ "current": 2611.075,
452
+ "min": 1500.0,
453
+ "max": 2500.0
454
+ },
455
+ {
456
+ "current": 2614.701,
457
+ "min": 1500.0,
458
+ "max": 2500.0
459
+ },
460
+ {
461
+ "current": 3348.767,
462
+ "min": 1500.0,
463
+ "max": 2500.0
464
+ },
465
+ {
466
+ "current": 1487.161,
467
+ "min": 1500.0,
468
+ "max": 2500.0
469
+ },
470
+ {
471
+ "current": 1488.665,
472
+ "min": 1500.0,
473
+ "max": 2500.0
474
+ },
475
+ {
476
+ "current": 1457.393,
477
+ "min": 1500.0,
478
+ "max": 2500.0
479
+ },
480
+ {
481
+ "current": 1912.878,
482
+ "min": 1500.0,
483
+ "max": 2500.0
484
+ },
485
+ {
486
+ "current": 1491.9,
487
+ "min": 1500.0,
488
+ "max": 2500.0
489
+ },
490
+ {
491
+ "current": 1492.847,
492
+ "min": 1500.0,
493
+ "max": 2500.0
494
+ },
495
+ {
496
+ "current": 2614.927,
497
+ "min": 1500.0,
498
+ "max": 2500.0
499
+ },
500
+ {
501
+ "current": 3347.835,
502
+ "min": 1500.0,
503
+ "max": 2500.0
504
+ },
505
+ {
506
+ "current": 2497.816,
507
+ "min": 1500.0,
508
+ "max": 2500.0
509
+ },
510
+ {
511
+ "current": 2543.329,
512
+ "min": 1500.0,
513
+ "max": 2500.0
514
+ },
515
+ {
516
+ "current": 3349.205,
517
+ "min": 1500.0,
518
+ "max": 2500.0
519
+ },
520
+ {
521
+ "current": 1498.763,
522
+ "min": 1500.0,
523
+ "max": 2500.0
524
+ },
525
+ {
526
+ "current": 1499.926,
527
+ "min": 1500.0,
528
+ "max": 2500.0
529
+ },
530
+ {
531
+ "current": 1499.647,
532
+ "min": 1500.0,
533
+ "max": 2500.0
534
+ },
535
+ {
536
+ "current": 1498.029,
537
+ "min": 1500.0,
538
+ "max": 2500.0
539
+ },
540
+ {
541
+ "current": 2556.933,
542
+ "min": 1500.0,
543
+ "max": 2500.0
544
+ },
545
+ {
546
+ "current": 1494.516,
547
+ "min": 1500.0,
548
+ "max": 2500.0
549
+ },
550
+ {
551
+ "current": 1493.284,
552
+ "min": 1500.0,
553
+ "max": 2500.0
554
+ },
555
+ {
556
+ "current": 1851.977,
557
+ "min": 1500.0,
558
+ "max": 2500.0
559
+ },
560
+ {
561
+ "current": 3351.242,
562
+ "min": 1500.0,
563
+ "max": 2500.0
564
+ },
565
+ {
566
+ "current": 3350.497,
567
+ "min": 1500.0,
568
+ "max": 2500.0
569
+ },
570
+ {
571
+ "current": 2474.341,
572
+ "min": 1500.0,
573
+ "max": 2500.0
574
+ },
575
+ {
576
+ "current": 2491.446,
577
+ "min": 1500.0,
578
+ "max": 2500.0
579
+ },
580
+ {
581
+ "current": 3349.844,
582
+ "min": 1500.0,
583
+ "max": 2500.0
584
+ },
585
+ {
586
+ "current": 1488.604,
587
+ "min": 1500.0,
588
+ "max": 2500.0
589
+ },
590
+ {
591
+ "current": 1488.116,
592
+ "min": 1500.0,
593
+ "max": 2500.0
594
+ },
595
+ {
596
+ "current": 1487.271,
597
+ "min": 1500.0,
598
+ "max": 2500.0
599
+ },
600
+ {
601
+ "current": 1497.468,
602
+ "min": 1500.0,
603
+ "max": 2500.0
604
+ },
605
+ {
606
+ "current": 1498.769,
607
+ "min": 1500.0,
608
+ "max": 2500.0
609
+ },
610
+ {
611
+ "current": 1498.175,
612
+ "min": 1500.0,
613
+ "max": 2500.0
614
+ },
615
+ {
616
+ "current": 1498.385,
617
+ "min": 1500.0,
618
+ "max": 2500.0
619
+ },
620
+ {
621
+ "current": 1492.725,
622
+ "min": 1500.0,
623
+ "max": 2500.0
624
+ },
625
+ {
626
+ "current": 1494.054,
627
+ "min": 1500.0,
628
+ "max": 2500.0
629
+ },
630
+ {
631
+ "current": 2521.72,
632
+ "min": 1500.0,
633
+ "max": 2500.0
634
+ },
635
+ {
636
+ "current": 1498.841,
637
+ "min": 1500.0,
638
+ "max": 2500.0
639
+ },
640
+ {
641
+ "current": 1492.901,
642
+ "min": 1500.0,
643
+ "max": 2500.0
644
+ },
645
+ {
646
+ "current": 1493.652,
647
+ "min": 1500.0,
648
+ "max": 2500.0
649
+ },
650
+ {
651
+ "current": 1858.326,
652
+ "min": 1500.0,
653
+ "max": 2500.0
654
+ },
655
+ {
656
+ "current": 1798.766,
657
+ "min": 1500.0,
658
+ "max": 2500.0
659
+ },
660
+ {
661
+ "current": 1488.242,
662
+ "min": 1500.0,
663
+ "max": 2500.0
664
+ },
665
+ {
666
+ "current": 3353.958,
667
+ "min": 1500.0,
668
+ "max": 2500.0
669
+ },
670
+ {
671
+ "current": 1674.086,
672
+ "min": 1500.0,
673
+ "max": 2500.0
674
+ },
675
+ {
676
+ "current": 1674.16,
677
+ "min": 1500.0,
678
+ "max": 2500.0
679
+ }
680
+ ],
681
+ "disk": {
682
+ "total": 100.0,
683
+ "used": 28.334514617919922
684
+ },
685
+ "gpu": "NVIDIA A100 80GB PCIe",
686
+ "gpu_count": 4,
687
+ "gpu_devices": [
688
+ {
689
+ "name": "NVIDIA A100 80GB PCIe",
690
+ "memory_total": 85899345920
691
+ },
692
+ {
693
+ "name": "NVIDIA A100 80GB PCIe",
694
+ "memory_total": 85899345920
695
+ },
696
+ {
697
+ "name": "NVIDIA A100 80GB PCIe",
698
+ "memory_total": 85899345920
699
+ },
700
+ {
701
+ "name": "NVIDIA A100 80GB PCIe",
702
+ "memory_total": 85899345920
703
+ }
704
+ ],
705
+ "memory": {
706
+ "total": 1007.7458572387695
707
+ }
708
+ }
wandb/run-20230328_174559-mzmf7gh8/files/wandb-summary.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"train/loss": 0.8147, "train/learning_rate": 4.468085106382978e-05, "train/epoch": 3.0, "train/global_step": 147, "_timestamp": 1680028745.4920614, "_runtime": 3186.164462327957, "_step": 14, "train/train_runtime": 3187.753, "train/train_samples_per_second": 46.828, "train/train_steps_per_second": 0.046, "train/total_flos": 1.5159002374476923e+18, "train/train_loss": 1.2030427374807344, "_wandb": {"runtime": 3185}}
wandb/run-20230328_174559-mzmf7gh8/logs/debug-internal.log ADDED
The diff for this file is too large to render. See raw diff
 
wandb/run-20230328_174559-mzmf7gh8/logs/debug.log ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Configure stats pid to 4415
2
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Loading settings from /root/.config/wandb/settings
3
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Loading settings from /root/alpaca-lora/wandb/settings
4
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Loading settings from environment variables: {'project': 'alpaca'}
5
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Applying setup settings: {'_disable_service': False}
6
+ 2023-03-28 17:45:59,321 INFO MainThread:4415 [wandb_setup.py:_flush():76] Inferring run settings from compute environment: {'program_relpath': 'finetune.py', 'program': '/root/alpaca-lora/finetune.py'}
7
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:_log_setup():506] Logging user logs to /root/alpaca-lora/wandb/run-20230328_174559-mzmf7gh8/logs/debug.log
8
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:_log_setup():507] Logging internal logs to /root/alpaca-lora/wandb/run-20230328_174559-mzmf7gh8/logs/debug-internal.log
9
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:init():546] calling init triggers
10
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:init():552] wandb.init called with sweep_config: {}
11
+ config: {}
12
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:init():602] starting backend
13
+ 2023-03-28 17:45:59,322 INFO MainThread:4415 [wandb_init.py:init():606] setting up manager
14
+ 2023-03-28 17:45:59,324 INFO MainThread:4415 [backend.py:_multiprocessing_setup():106] multiprocessing start_methods=fork,spawn,forkserver, using: spawn
15
+ 2023-03-28 17:45:59,327 INFO MainThread:4415 [wandb_init.py:init():613] backend started and connected
16
+ 2023-03-28 17:45:59,329 INFO MainThread:4415 [wandb_init.py:init():701] updated telemetry
17
+ 2023-03-28 17:45:59,335 INFO MainThread:4415 [wandb_init.py:init():741] communicating run to backend with 60.0 second timeout
18
+ 2023-03-28 17:45:59,540 INFO MainThread:4415 [wandb_run.py:_on_init():2133] communicating current version
19
+ 2023-03-28 17:45:59,703 INFO MainThread:4415 [wandb_run.py:_on_init():2142] got version response
20
+ 2023-03-28 17:45:59,703 INFO MainThread:4415 [wandb_init.py:init():789] starting run threads in backend
21
+ 2023-03-28 17:45:59,827 INFO MainThread:4415 [wandb_run.py:_console_start():2114] atexit reg
22
+ 2023-03-28 17:45:59,828 INFO MainThread:4415 [wandb_run.py:_redirect():1969] redirect: SettingsConsole.WRAP_RAW
23
+ 2023-03-28 17:45:59,828 INFO MainThread:4415 [wandb_run.py:_redirect():2034] Wrapping output streams.
24
+ 2023-03-28 17:45:59,828 INFO MainThread:4415 [wandb_run.py:_redirect():2059] Redirects installed.
25
+ 2023-03-28 17:45:59,829 INFO MainThread:4415 [wandb_init.py:init():831] run started, returning control to user process
26
+ 2023-03-28 17:45:59,832 INFO MainThread:4415 [wandb_run.py:_config_callback():1251] config_cb None None {'vocab_size': 32000, 'hidden_size': 4096, 'intermediate_size': 11008, 'num_hidden_layers': 32, 'num_attention_heads': 32, 'hidden_act': 'silu', 'initializer_range': 0.02, 'rms_norm_eps': 1e-06, 'use_cache': False, 'return_dict': True, 'output_hidden_states': False, 'output_attentions': False, 'torchscript': False, 'torch_dtype': 'float16', 'use_bfloat16': False, 'tf_legacy_loss': False, 'pruned_heads': {}, 'tie_word_embeddings': False, 'is_encoder_decoder': False, 'is_decoder': False, 'cross_attention_hidden_size': None, 'add_cross_attention': False, 'tie_encoder_decoder': False, 'max_length': 20, 'min_length': 0, 'do_sample': False, 'early_stopping': False, 'num_beams': 1, 'num_beam_groups': 1, 'diversity_penalty': 0.0, 'temperature': 1.0, 'top_k': 50, 'top_p': 1.0, 'typical_p': 1.0, 'repetition_penalty': 1.0, 'length_penalty': 1.0, 'no_repeat_ngram_size': 0, 'encoder_no_repeat_ngram_size': 0, 'bad_words_ids': None, 'num_return_sequences': 1, 'chunk_size_feed_forward': 0, 'output_scores': False, 'return_dict_in_generate': False, 'forced_bos_token_id': None, 'forced_eos_token_id': None, 'remove_invalid_values': False, 'exponential_decay_length_penalty': None, 'suppress_tokens': None, 'begin_suppress_tokens': None, 'architectures': ['LlamaForCausalLM'], 'finetuning_task': None, 'id2label': {0: 'LABEL_0', 1: 'LABEL_1'}, 'label2id': {'LABEL_0': 0, 'LABEL_1': 1}, 'tokenizer_class': None, 'prefix': None, 'bos_token_id': 1, 'pad_token_id': 0, 'eos_token_id': 2, 'sep_token_id': None, 'decoder_start_token_id': None, 'task_specific_params': None, 'problem_type': None, '_name_or_path': '/root/llama-7b-hf/', 'transformers_version': '4.28.0.dev0', 'model_type': 'llama', 'output_dir': './lora-alpaca', 'overwrite_output_dir': False, 'do_train': False, 'do_eval': True, 'do_predict': False, 'evaluation_strategy': 'steps', 'prediction_loss_only': False, 'per_device_train_batch_size': 128, 'per_device_eval_batch_size': 8, 'per_gpu_train_batch_size': 'None', 'per_gpu_eval_batch_size': 'None', 'gradient_accumulation_steps': 2, 'eval_accumulation_steps': 'None', 'eval_delay': 0, 'learning_rate': 0.0003, 'weight_decay': 0.0, 'adam_beta1': 0.9, 'adam_beta2': 0.999, 'adam_epsilon': 1e-08, 'max_grad_norm': 1.0, 'num_train_epochs': 3, 'max_steps': -1, 'lr_scheduler_type': 'linear', 'warmup_ratio': 0.0, 'warmup_steps': 100, 'log_level': 'passive', 'log_level_replica': 'warning', 'log_on_each_node': True, 'logging_dir': './lora-alpaca/runs/Mar28_17-45-56_c35059eec1c7', 'logging_strategy': 'steps', 'logging_first_step': False, 'logging_steps': 10, 'logging_nan_inf_filter': True, 'save_strategy': 'steps', 'save_steps': 200, 'save_total_limit': 3, 'save_on_each_node': False, 'no_cuda': False, 'use_mps_device': False, 'seed': 42, 'data_seed': 'None', 'jit_mode_eval': False, 'use_ipex': False, 'bf16': False, 'fp16': True, 'fp16_opt_level': 'O1', 'half_precision_backend': 'cuda_amp', 'bf16_full_eval': False, 'fp16_full_eval': False, 'tf32': 'None', 'local_rank': 0, 'xpu_backend': 'None', 'tpu_num_cores': 'None', 'tpu_metrics_debug': False, 'debug': '[]', 'dataloader_drop_last': False, 'eval_steps': 200, 'dataloader_num_workers': 0, 'past_index': -1, 'run_name': '', 'disable_tqdm': False, 'remove_unused_columns': True, 'label_names': 'None', 'load_best_model_at_end': True, 'metric_for_best_model': 'loss', 'greater_is_better': False, 'ignore_data_skip': False, 'sharded_ddp': '[]', 'fsdp': '[]', 'fsdp_min_num_params': 0, 'fsdp_config': "{'fsdp_min_num_params': 0, 'xla': False, 'xla_fsdp_grad_ckpt': False}", 'fsdp_transformer_layer_cls_to_wrap': 'None', 'deepspeed': 'None', 'label_smoothing_factor': 0.0, 'optim': 'adamw_torch', 'optim_args': 'None', 'adafactor': False, 'group_by_length': False, 'length_column_name': 'length', 'report_to': "['wandb']", 'ddp_find_unused_parameters': False, 'ddp_bucket_cap_mb': 'None', 'dataloader_pin_memory': True, 'skip_memory_metrics': True, 'use_legacy_prediction_loop': False, 'push_to_hub': False, 'resume_from_checkpoint': 'None', 'hub_model_id': 'None', 'hub_strategy': 'every_save', 'hub_token': '<HUB_TOKEN>', 'hub_private_repo': False, 'gradient_checkpointing': False, 'include_inputs_for_metrics': False, 'fp16_backend': 'auto', 'push_to_hub_model_id': 'None', 'push_to_hub_organization': 'None', 'push_to_hub_token': '<PUSH_TO_HUB_TOKEN>', 'mp_parameters': '', 'auto_find_batch_size': False, 'full_determinism': False, 'torchdynamo': 'None', 'ray_scope': 'last', 'ddp_timeout': 1800, 'torch_compile': False, 'torch_compile_backend': 'None', 'torch_compile_mode': 'None', 'train_batch_size': 128, 'eval_batch_size': 8}
27
+ 2023-03-28 18:39:10,751 WARNING MsgRouterThr:4415 [router.py:message_loop():77] message_loop has been closed
wandb/run-20230328_174559-mzmf7gh8/run-mzmf7gh8.wandb ADDED
Binary file (598 kB). View file