Granther commited on
Commit
1ca7b19
1 Parent(s): 97eb20d

Upload sft_phi3.ipynb with huggingface_hub

Browse files
Files changed (1) hide show
  1. sft_phi3.ipynb +496 -0
sft_phi3.ipynb ADDED
@@ -0,0 +1,496 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "cells": [
3
+ {
4
+ "cell_type": "code",
5
+ "execution_count": 32,
6
+ "id": "578786b8-092a-4de8-9955-4e87da557639",
7
+ "metadata": {
8
+ "scrolled": true
9
+ },
10
+ "outputs": [
11
+ {
12
+ "name": "stdout",
13
+ "output_type": "stream",
14
+ "text": [
15
+ "Requirement already satisfied: peft in /opt/conda/lib/python3.10/site-packages (0.11.1)\n",
16
+ "Requirement already satisfied: numpy>=1.17 in /opt/conda/lib/python3.10/site-packages (from peft) (1.26.3)\n",
17
+ "Requirement already satisfied: packaging>=20.0 in /opt/conda/lib/python3.10/site-packages (from peft) (23.1)\n",
18
+ "Requirement already satisfied: psutil in /opt/conda/lib/python3.10/site-packages (from peft) (5.9.0)\n",
19
+ "Requirement already satisfied: pyyaml in /opt/conda/lib/python3.10/site-packages (from peft) (6.0.1)\n",
20
+ "Requirement already satisfied: torch>=1.13.0 in /opt/conda/lib/python3.10/site-packages (from peft) (2.2.0)\n",
21
+ "Requirement already satisfied: transformers in /opt/conda/lib/python3.10/site-packages (from peft) (4.42.3)\n",
22
+ "Requirement already satisfied: tqdm in /opt/conda/lib/python3.10/site-packages (from peft) (4.66.4)\n",
23
+ "Requirement already satisfied: accelerate>=0.21.0 in /opt/conda/lib/python3.10/site-packages (from peft) (0.32.0)\n",
24
+ "Requirement already satisfied: safetensors in /opt/conda/lib/python3.10/site-packages (from peft) (0.4.3)\n",
25
+ "Requirement already satisfied: huggingface-hub>=0.17.0 in /opt/conda/lib/python3.10/site-packages (from peft) (0.23.4)\n",
26
+ "Requirement already satisfied: filelock in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft) (3.13.1)\n",
27
+ "Requirement already satisfied: fsspec>=2023.5.0 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft) (2023.12.2)\n",
28
+ "Requirement already satisfied: requests in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft) (2.32.3)\n",
29
+ "Requirement already satisfied: typing-extensions>=3.7.4.3 in /opt/conda/lib/python3.10/site-packages (from huggingface-hub>=0.17.0->peft) (4.9.0)\n",
30
+ "Requirement already satisfied: sympy in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft) (1.12)\n",
31
+ "Requirement already satisfied: networkx in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft) (3.1)\n",
32
+ "Requirement already satisfied: jinja2 in /opt/conda/lib/python3.10/site-packages (from torch>=1.13.0->peft) (3.1.2)\n",
33
+ "Requirement already satisfied: regex!=2019.12.17 in /opt/conda/lib/python3.10/site-packages (from transformers->peft) (2024.5.15)\n",
34
+ "Requirement already satisfied: tokenizers<0.20,>=0.19 in /opt/conda/lib/python3.10/site-packages (from transformers->peft) (0.19.1)\n",
35
+ "Requirement already satisfied: MarkupSafe>=2.0 in /opt/conda/lib/python3.10/site-packages (from jinja2->torch>=1.13.0->peft) (2.1.3)\n",
36
+ "Requirement already satisfied: charset-normalizer<4,>=2 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft) (2.0.4)\n",
37
+ "Requirement already satisfied: idna<4,>=2.5 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft) (3.4)\n",
38
+ "Requirement already satisfied: urllib3<3,>=1.21.1 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft) (1.26.18)\n",
39
+ "Requirement already satisfied: certifi>=2017.4.17 in /opt/conda/lib/python3.10/site-packages (from requests->huggingface-hub>=0.17.0->peft) (2023.11.17)\n",
40
+ "Requirement already satisfied: mpmath>=0.19 in /opt/conda/lib/python3.10/site-packages (from sympy->torch>=1.13.0->peft) (1.3.0)\n",
41
+ "\u001b[33mWARNING: Running pip as the 'root' user can result in broken permissions and conflicting behaviour with the system package manager. It is recommended to use a virtual environment instead: https://pip.pypa.io/warnings/venv\u001b[0m\u001b[33m\n",
42
+ "\u001b[0m"
43
+ ]
44
+ }
45
+ ],
46
+ "source": [
47
+ "#!pip install huggingface_hub torch transformers datasets trl \n",
48
+ "#!pip install flash-attn --no-build-isolation\n",
49
+ "!pip install --upgrade peft"
50
+ ]
51
+ },
52
+ {
53
+ "cell_type": "code",
54
+ "execution_count": 2,
55
+ "id": "4a74bec4-4bf0-47be-802a-046073da573e",
56
+ "metadata": {},
57
+ "outputs": [],
58
+ "source": [
59
+ "import sys\n",
60
+ "import logging\n",
61
+ "\n",
62
+ "import datasets\n",
63
+ "from datasets import load_dataset\n",
64
+ "from peft import LoraConfig\n",
65
+ "import torch\n",
66
+ "import transformers\n",
67
+ "from trl import SFTTrainer, SFTConfig\n",
68
+ "from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, BitsAndBytesConfig"
69
+ ]
70
+ },
71
+ {
72
+ "cell_type": "code",
73
+ "execution_count": 3,
74
+ "id": "8a9bc6f8-4a1e-42d8-897d-5225e1b5011a",
75
+ "metadata": {},
76
+ "outputs": [],
77
+ "source": [
78
+ "dataset_id = (\"wikitext\", \"wikitext-103-raw-v1\")\n",
79
+ "dataset_id = \"HuggingFaceH4/ultrachat_200k\"\n",
80
+ "\n",
81
+ "dataset = load_dataset(dataset_id)"
82
+ ]
83
+ },
84
+ {
85
+ "cell_type": "code",
86
+ "execution_count": 4,
87
+ "id": "f3b226eb-b159-4533-bd33-2746181a80b3",
88
+ "metadata": {},
89
+ "outputs": [],
90
+ "source": [
91
+ "training_config = {\n",
92
+ " \"bf16\": True,\n",
93
+ " \"do_eval\": False,\n",
94
+ " \"do_train\": True, # defualts to False, not sure where this fits\n",
95
+ " \"learning_rate\": 5.0e-06,\n",
96
+ " \"log_level\": \"info\",\n",
97
+ " \"logging_steps\": 20,\n",
98
+ " \"logging_strategy\": \"steps\",\n",
99
+ " \"lr_scheduler_type\": \"cosine\",\n",
100
+ " \"num_train_epochs\": 1,\n",
101
+ " \"max_steps\": -1,\n",
102
+ " \"output_dir\": \"./checkpoint_dir\", # model predictions and checkpoints\n",
103
+ " \"overwrite_output_dir\": True,\n",
104
+ " \"per_device_eval_batch_size\": 4,\n",
105
+ " \"per_device_train_batch_size\": 4,\n",
106
+ " \"remove_unused_columns\": True,\n",
107
+ " \"save_steps\": 100,\n",
108
+ " \"save_total_limit\": 1,\n",
109
+ " \"seed\": 0,\n",
110
+ " \"gradient_checkpointing\": True,\n",
111
+ " \"gradient_checkpointing_kwargs\":{\"use_reentrant\": False},\n",
112
+ " \"gradient_accumulation_steps\": 1, # number of steps to accumulate before beckprop\n",
113
+ " \"warmup_ratio\": 0.2,\n",
114
+ " \"packing\": False,\n",
115
+ " \"max_seq_length\": 2048,\n",
116
+ " \"dataset_text_field\": \"text\",\n",
117
+ " }\n",
118
+ "\n",
119
+ "peft_config = {\n",
120
+ " \"r\": 16, # default values VV\n",
121
+ " \"lora_alpha\": 32,\n",
122
+ " \"lora_dropout\": 0.05,\n",
123
+ " \"bias\": \"none\",\n",
124
+ " \"task_type\": \"CAUSAL_LM\",\n",
125
+ " \"target_modules\": \"all-linear\",\n",
126
+ " \"modules_to_save\": None,\n",
127
+ "}\n",
128
+ "\n",
129
+ "train_conf = SFTConfig(**training_config)\n",
130
+ "#train_conf = TrainingArguments(**training_config)\n",
131
+ "peft_conf = LoraConfig(**peft_config)"
132
+ ]
133
+ },
134
+ {
135
+ "cell_type": "code",
136
+ "execution_count": null,
137
+ "id": "20c9d834-50fe-4495-b003-7d80495c8439",
138
+ "metadata": {
139
+ "scrolled": true
140
+ },
141
+ "outputs": [
142
+ {
143
+ "data": {
144
+ "application/vnd.jupyter.widget-view+json": {
145
+ "model_id": "08aed232727444ab814beb2c188090eb",
146
+ "version_major": 2,
147
+ "version_minor": 0
148
+ },
149
+ "text/plain": [
150
+ "Loading checkpoint shards: 0%| | 0/2 [00:00<?, ?it/s]"
151
+ ]
152
+ },
153
+ "metadata": {},
154
+ "output_type": "display_data"
155
+ },
156
+ {
157
+ "name": "stderr",
158
+ "output_type": "stream",
159
+ "text": [
160
+ "Exception ignored in: <bound method IPythonKernel._clean_thread_parent_frames of <ipykernel.ipkernel.IPythonKernel object at 0x76c288d725c0>>\n",
161
+ "Traceback (most recent call last):\n",
162
+ " File \"/opt/conda/lib/python3.10/site-packages/ipykernel/ipkernel.py\", line 775, in _clean_thread_parent_frames\n",
163
+ " def _clean_thread_parent_frames(\n",
164
+ "KeyboardInterrupt: \n",
165
+ "Exception ignored in: <bound method IPythonKernel._clean_thread_parent_frames of <ipykernel.ipkernel.IPythonKernel object at 0x76c288d725c0>>\n",
166
+ "Traceback (most recent call last):\n",
167
+ " File \"/opt/conda/lib/python3.10/site-packages/ipykernel/ipkernel.py\", line 775, in _clean_thread_parent_frames\n",
168
+ " def _clean_thread_parent_frames(\n",
169
+ "KeyboardInterrupt: \n",
170
+ "Exception ignored in: <bound method IPythonKernel._clean_thread_parent_frames of <ipykernel.ipkernel.IPythonKernel object at 0x76c288d725c0>>\n",
171
+ "Traceback (most recent call last):\n",
172
+ " File \"/opt/conda/lib/python3.10/site-packages/ipykernel/ipkernel.py\", line 775, in _clean_thread_parent_frames\n",
173
+ " def _clean_thread_parent_frames(\n",
174
+ "KeyboardInterrupt: \n",
175
+ "Exception ignored in: <bound method IPythonKernel._clean_thread_parent_frames of <ipykernel.ipkernel.IPythonKernel object at 0x76c288d725c0>>\n",
176
+ "Traceback (most recent call last):\n",
177
+ " File \"/opt/conda/lib/python3.10/site-packages/ipykernel/ipkernel.py\", line 775, in _clean_thread_parent_frames\n",
178
+ " def _clean_thread_parent_frames(\n",
179
+ "KeyboardInterrupt: \n"
180
+ ]
181
+ }
182
+ ],
183
+ "source": [
184
+ "checkpoint_path = \"microsoft/Phi-3-mini-128k-instruct\"\n",
185
+ "model_kwargs = dict(\n",
186
+ " use_cache=False,\n",
187
+ " trust_remote_code=True,\n",
188
+ " attn_implementation=\"flash_attention_2\", # loading the model with flash-attenstion support\n",
189
+ " torch_dtype=torch.bfloat16,\n",
190
+ " device_map=\"auto\"\n",
191
+ ")\n",
192
+ "\n",
193
+ "model = AutoModelForCausalLM.from_pretrained(checkpoint_path, **model_kwargs)\n",
194
+ "tokenizer = AutoTokenizer.from_pretrained(checkpoint_path, truncation=True)"
195
+ ]
196
+ },
197
+ {
198
+ "cell_type": "code",
199
+ "execution_count": 6,
200
+ "id": "d684252c-2151-4601-8ebb-398bd3a63f00",
201
+ "metadata": {},
202
+ "outputs": [],
203
+ "source": [
204
+ "tokenizer.model_max_length = 2048\n",
205
+ "#tokenizer.pad_token = tokenizer.unk_token # use unk rather than eos token to prevent endless generation\n",
206
+ "#tokenizer.pad_token_id = tokenizer.convert_tokens_to_ids(tokenizer.pad_token)\n",
207
+ "tokenizer.padding_side = 'right'"
208
+ ]
209
+ },
210
+ {
211
+ "cell_type": "code",
212
+ "execution_count": 7,
213
+ "id": "75869100-99f7-49c7-a9d3-7a3950dd7d72",
214
+ "metadata": {
215
+ "scrolled": true
216
+ },
217
+ "outputs": [],
218
+ "source": [
219
+ "def preproc(examples, tokenizer):\n",
220
+ " messages = examples['messages']\n",
221
+ " examples['text'] = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=False) #, return_dict=True)\n",
222
+ " return examples\n",
223
+ "\n",
224
+ "train_dataset = dataset['train_sft']\n",
225
+ "test_dataset = dataset['test_sft']\n",
226
+ "\n",
227
+ "train_dataset = train_dataset.map(preproc,\n",
228
+ " fn_kwargs={'tokenizer':tokenizer},\n",
229
+ " num_proc=24,\n",
230
+ " #batched=True,\n",
231
+ " remove_columns=list(train_dataset.features)).select(range(1000))\n",
232
+ "\n",
233
+ "test_dataset = test_dataset.map(preproc,\n",
234
+ " fn_kwargs={'tokenizer':tokenizer},\n",
235
+ " num_proc=24,\n",
236
+ " #batched=True,\n",
237
+ " remove_columns=list(test_dataset.features))#[10000:]"
238
+ ]
239
+ },
240
+ {
241
+ "cell_type": "code",
242
+ "execution_count": 8,
243
+ "id": "56cd1b31-6f7e-4c7d-8524-b12cf94b9c9f",
244
+ "metadata": {},
245
+ "outputs": [
246
+ {
247
+ "data": {
248
+ "application/vnd.jupyter.widget-view+json": {
249
+ "model_id": "5d79f04152484f9494e389b264fc7176",
250
+ "version_major": 2,
251
+ "version_minor": 0
252
+ },
253
+ "text/plain": [
254
+ "Map: 0%| | 0/1000 [00:00<?, ? examples/s]"
255
+ ]
256
+ },
257
+ "metadata": {},
258
+ "output_type": "display_data"
259
+ },
260
+ {
261
+ "name": "stderr",
262
+ "output_type": "stream",
263
+ "text": [
264
+ "Using auto half precision backend\n"
265
+ ]
266
+ }
267
+ ],
268
+ "source": [
269
+ "trainer = SFTTrainer(\n",
270
+ " model=model,\n",
271
+ " args=train_conf,\n",
272
+ " peft_config=peft_conf,\n",
273
+ " train_dataset=train_dataset,\n",
274
+ " #eval_dataset=test_dataset,\n",
275
+ " # max_seq_length=tokenizer.model_max_length,\n",
276
+ " # dataset_text_field=\"text\",\n",
277
+ " tokenizer=tokenizer,\n",
278
+ " # packing=True\n",
279
+ ")"
280
+ ]
281
+ },
282
+ {
283
+ "cell_type": "code",
284
+ "execution_count": 16,
285
+ "id": "d8e6b669-1717-429a-9c43-3c02adb8a3d1",
286
+ "metadata": {},
287
+ "outputs": [
288
+ {
289
+ "name": "stderr",
290
+ "output_type": "stream",
291
+ "text": [
292
+ "***** Running training *****\n",
293
+ " Num examples = 1,000\n",
294
+ " Num Epochs = 1\n",
295
+ " Instantaneous batch size per device = 4\n",
296
+ " Total train batch size (w. parallel, distributed & accumulation) = 4\n",
297
+ " Gradient Accumulation steps = 1\n",
298
+ " Total optimization steps = 250\n",
299
+ " Number of trainable parameters = 25,165,824\n"
300
+ ]
301
+ },
302
+ {
303
+ "data": {
304
+ "text/html": [
305
+ "\n",
306
+ " <div>\n",
307
+ " \n",
308
+ " <progress value='4' max='250' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
309
+ " [ 4/250 00:04 < 09:17, 0.44 it/s, Epoch 0.01/1]\n",
310
+ " </div>\n",
311
+ " <table border=\"1\" class=\"dataframe\">\n",
312
+ " <thead>\n",
313
+ " <tr style=\"text-align: left;\">\n",
314
+ " <th>Step</th>\n",
315
+ " <th>Training Loss</th>\n",
316
+ " </tr>\n",
317
+ " </thead>\n",
318
+ " <tbody>\n",
319
+ " </tbody>\n",
320
+ "</table><p>"
321
+ ],
322
+ "text/plain": [
323
+ "<IPython.core.display.HTML object>"
324
+ ]
325
+ },
326
+ "metadata": {},
327
+ "output_type": "display_data"
328
+ },
329
+ {
330
+ "ename": "KeyboardInterrupt",
331
+ "evalue": "",
332
+ "output_type": "error",
333
+ "traceback": [
334
+ "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
335
+ "\u001b[0;31mKeyboardInterrupt\u001b[0m Traceback (most recent call last)",
336
+ "Cell \u001b[0;32mIn[16], line 1\u001b[0m\n\u001b[0;32m----> 1\u001b[0m train_result \u001b[38;5;241m=\u001b[39m \u001b[43mtrainer\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2\u001b[0m metrics \u001b[38;5;241m=\u001b[39m train_result\u001b[38;5;241m.\u001b[39mmetrics\n\u001b[1;32m 3\u001b[0m trainer\u001b[38;5;241m.\u001b[39msave_state()\n",
337
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/trl/trainer/sft_trainer.py:440\u001b[0m, in \u001b[0;36mSFTTrainer.train\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m 437\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mneftune_noise_alpha \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trainer_supports_neftune:\n\u001b[1;32m 438\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trl_activate_neftune(\u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mmodel)\n\u001b[0;32m--> 440\u001b[0m output \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43msuper\u001b[39;49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 442\u001b[0m \u001b[38;5;66;03m# After training we make sure to retrieve back the original forward pass method\u001b[39;00m\n\u001b[1;32m 443\u001b[0m \u001b[38;5;66;03m# for the embedding layer by removing the forward post hook.\u001b[39;00m\n\u001b[1;32m 444\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mneftune_noise_alpha \u001b[38;5;129;01mis\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28;01mNone\u001b[39;00m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_trainer_supports_neftune:\n",
338
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/transformers/trainer.py:1932\u001b[0m, in \u001b[0;36mTrainer.train\u001b[0;34m(self, resume_from_checkpoint, trial, ignore_keys_for_eval, **kwargs)\u001b[0m\n\u001b[1;32m 1930\u001b[0m hf_hub_utils\u001b[38;5;241m.\u001b[39menable_progress_bars()\n\u001b[1;32m 1931\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 1932\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m \u001b[43minner_training_loop\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 1933\u001b[0m \u001b[43m \u001b[49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 1934\u001b[0m \u001b[43m \u001b[49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mresume_from_checkpoint\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 1935\u001b[0m \u001b[43m \u001b[49m\u001b[43mtrial\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtrial\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 1936\u001b[0m \u001b[43m \u001b[49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mignore_keys_for_eval\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 1937\u001b[0m \u001b[43m \u001b[49m\u001b[43m)\u001b[49m\n",
339
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/transformers/trainer.py:2268\u001b[0m, in \u001b[0;36mTrainer._inner_training_loop\u001b[0;34m(self, batch_size, args, resume_from_checkpoint, trial, ignore_keys_for_eval)\u001b[0m\n\u001b[1;32m 2265\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol \u001b[38;5;241m=\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcallback_handler\u001b[38;5;241m.\u001b[39mon_step_begin(args, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate, \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mcontrol)\n\u001b[1;32m 2267\u001b[0m \u001b[38;5;28;01mwith\u001b[39;00m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39maccelerator\u001b[38;5;241m.\u001b[39maccumulate(model):\n\u001b[0;32m-> 2268\u001b[0m tr_loss_step \u001b[38;5;241m=\u001b[39m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtraining_step\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43minputs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 2270\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m (\n\u001b[1;32m 2271\u001b[0m args\u001b[38;5;241m.\u001b[39mlogging_nan_inf_filter\n\u001b[1;32m 2272\u001b[0m \u001b[38;5;129;01mand\u001b[39;00m \u001b[38;5;129;01mnot\u001b[39;00m is_torch_xla_available()\n\u001b[1;32m 2273\u001b[0m \u001b[38;5;129;01mand\u001b[39;00m (torch\u001b[38;5;241m.\u001b[39misnan(tr_loss_step) \u001b[38;5;129;01mor\u001b[39;00m torch\u001b[38;5;241m.\u001b[39misinf(tr_loss_step))\n\u001b[1;32m 2274\u001b[0m ):\n\u001b[1;32m 2275\u001b[0m \u001b[38;5;66;03m# if loss is nan or inf simply add the average of previous logged losses\u001b[39;00m\n\u001b[1;32m 2276\u001b[0m tr_loss \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m tr_loss \u001b[38;5;241m/\u001b[39m (\u001b[38;5;241m1\u001b[39m \u001b[38;5;241m+\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mstate\u001b[38;5;241m.\u001b[39mglobal_step \u001b[38;5;241m-\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39m_globalstep_last_logged)\n",
340
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/transformers/trainer.py:3324\u001b[0m, in \u001b[0;36mTrainer.training_step\u001b[0;34m(***failed resolving arguments***)\u001b[0m\n\u001b[1;32m 3322\u001b[0m scaled_loss\u001b[38;5;241m.\u001b[39mbackward()\n\u001b[1;32m 3323\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 3324\u001b[0m \u001b[38;5;28;43mself\u001b[39;49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43maccelerator\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbackward\u001b[49m\u001b[43m(\u001b[49m\u001b[43mloss\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m 3326\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m loss\u001b[38;5;241m.\u001b[39mdetach() \u001b[38;5;241m/\u001b[39m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39margs\u001b[38;5;241m.\u001b[39mgradient_accumulation_steps\n",
341
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/accelerate/accelerator.py:2151\u001b[0m, in \u001b[0;36mAccelerator.backward\u001b[0;34m(self, loss, **kwargs)\u001b[0m\n\u001b[1;32m 2149\u001b[0m \u001b[38;5;28mself\u001b[39m\u001b[38;5;241m.\u001b[39mlomo_backward(loss, learning_rate)\n\u001b[1;32m 2150\u001b[0m \u001b[38;5;28;01melse\u001b[39;00m:\n\u001b[0;32m-> 2151\u001b[0m \u001b[43mloss\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbackward\u001b[49m\u001b[43m(\u001b[49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[38;5;241;43m*\u001b[39;49m\u001b[43mkwargs\u001b[49m\u001b[43m)\u001b[49m\n",
342
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/_tensor.py:522\u001b[0m, in \u001b[0;36mTensor.backward\u001b[0;34m(self, gradient, retain_graph, create_graph, inputs)\u001b[0m\n\u001b[1;32m 512\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m has_torch_function_unary(\u001b[38;5;28mself\u001b[39m):\n\u001b[1;32m 513\u001b[0m \u001b[38;5;28;01mreturn\u001b[39;00m handle_torch_function(\n\u001b[1;32m 514\u001b[0m Tensor\u001b[38;5;241m.\u001b[39mbackward,\n\u001b[1;32m 515\u001b[0m (\u001b[38;5;28mself\u001b[39m,),\n\u001b[0;32m (...)\u001b[0m\n\u001b[1;32m 520\u001b[0m inputs\u001b[38;5;241m=\u001b[39minputs,\n\u001b[1;32m 521\u001b[0m )\n\u001b[0;32m--> 522\u001b[0m \u001b[43mtorch\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mautograd\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbackward\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m 523\u001b[0m \u001b[43m \u001b[49m\u001b[38;5;28;43mself\u001b[39;49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mgradient\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mretain_graph\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mcreate_graph\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43minputs\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43minputs\u001b[49m\n\u001b[1;32m 524\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
343
+ "File \u001b[0;32m/opt/conda/lib/python3.10/site-packages/torch/autograd/__init__.py:266\u001b[0m, in \u001b[0;36mbackward\u001b[0;34m(tensors, grad_tensors, retain_graph, create_graph, grad_variables, inputs)\u001b[0m\n\u001b[1;32m 261\u001b[0m retain_graph \u001b[38;5;241m=\u001b[39m create_graph\n\u001b[1;32m 263\u001b[0m \u001b[38;5;66;03m# The reason we repeat the same comment below is that\u001b[39;00m\n\u001b[1;32m 264\u001b[0m \u001b[38;5;66;03m# some Python versions print out the first line of a multi-line function\u001b[39;00m\n\u001b[1;32m 265\u001b[0m \u001b[38;5;66;03m# calls in the traceback and some print out the last line\u001b[39;00m\n\u001b[0;32m--> 266\u001b[0m \u001b[43mVariable\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43m_execution_engine\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mrun_backward\u001b[49m\u001b[43m(\u001b[49m\u001b[43m \u001b[49m\u001b[38;5;66;43;03m# Calls into the C++ engine to run the backward pass\u001b[39;49;00m\n\u001b[1;32m 267\u001b[0m \u001b[43m \u001b[49m\u001b[43mtensors\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 268\u001b[0m \u001b[43m \u001b[49m\u001b[43mgrad_tensors_\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 269\u001b[0m \u001b[43m \u001b[49m\u001b[43mretain_graph\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 270\u001b[0m \u001b[43m \u001b[49m\u001b[43mcreate_graph\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 271\u001b[0m \u001b[43m \u001b[49m\u001b[43minputs\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m 272\u001b[0m \u001b[43m \u001b[49m\u001b[43mallow_unreachable\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[1;32m 273\u001b[0m \u001b[43m \u001b[49m\u001b[43maccumulate_grad\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;28;43;01mTrue\u001b[39;49;00m\u001b[43m,\u001b[49m\n\u001b[1;32m 274\u001b[0m \u001b[43m\u001b[49m\u001b[43m)\u001b[49m\n",
344
+ "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
345
+ ]
346
+ }
347
+ ],
348
+ "source": [
349
+ "train_result = trainer.train()\n",
350
+ "metrics = train_result.metrics\n",
351
+ "trainer.save_state()"
352
+ ]
353
+ },
354
+ {
355
+ "cell_type": "code",
356
+ "execution_count": 12,
357
+ "id": "4d4207fc-1578-4591-a480-467fd2a5855b",
358
+ "metadata": {},
359
+ "outputs": [
360
+ {
361
+ "data": {
362
+ "text/plain": [
363
+ "{'train_runtime': 506.2204,\n",
364
+ " 'train_samples_per_second': 1.975,\n",
365
+ " 'train_steps_per_second': 0.494,\n",
366
+ " 'total_flos': 4.041582948790272e+16,\n",
367
+ " 'train_loss': 1.1037534561157227,\n",
368
+ " 'epoch': 1.0}"
369
+ ]
370
+ },
371
+ "execution_count": 12,
372
+ "metadata": {},
373
+ "output_type": "execute_result"
374
+ }
375
+ ],
376
+ "source": [
377
+ "metrics"
378
+ ]
379
+ },
380
+ {
381
+ "cell_type": "code",
382
+ "execution_count": null,
383
+ "id": "f92339ec-0448-40d2-9458-6242e35b9bdc",
384
+ "metadata": {},
385
+ "outputs": [],
386
+ "source": [
387
+ "from peft import PeftConfig, PeftModel \n",
388
+ "\n",
389
+ "checkpoint_path = \"microsoft/Phi-3-mini-128k-instruct\"\n",
390
+ "adapter_path = \"./checkpoint_dir/checkpoint-250\"\n",
391
+ "\n",
392
+ "model_kwargs = dict(\n",
393
+ " use_cache=False,\n",
394
+ " trust_remote_code=True,\n",
395
+ " attn_implementation=\"flash_attention_2\", # loading the model with flash-attenstion support\n",
396
+ " torch_dtype=torch.bfloat16,\n",
397
+ " device_map=\"auto\"\n",
398
+ ")\n",
399
+ "\n",
400
+ "model = AutoModelForCausalLM.from_pretrained(checkpoint_path, **model_kwargs)"
401
+ ]
402
+ },
403
+ {
404
+ "cell_type": "code",
405
+ "execution_count": 8,
406
+ "id": "f0cf458d-8b4f-4ff9-bd60-bbe510416cea",
407
+ "metadata": {},
408
+ "outputs": [
409
+ {
410
+ "name": "stderr",
411
+ "output_type": "stream",
412
+ "text": [
413
+ "Special tokens have been added in the vocabulary, make sure the associated word embeddings are fine-tuned or trained.\n"
414
+ ]
415
+ }
416
+ ],
417
+ "source": [
418
+ "model = PeftModel.from_pretrained(model, adapter_path)\n",
419
+ "tokenizer = AutoTokenizer.from_pretrained(checkpoint_path)"
420
+ ]
421
+ },
422
+ {
423
+ "cell_type": "code",
424
+ "execution_count": 22,
425
+ "id": "b5ada882-b7d2-46c5-ba5b-54fab2556832",
426
+ "metadata": {},
427
+ "outputs": [],
428
+ "source": [
429
+ "input_text = [\n",
430
+ " {'role': 'user', 'content': \"Tell me about cats\"},\n",
431
+ "]\n",
432
+ "\n",
433
+ "input = \"Tell me about cats\"\n",
434
+ "\n",
435
+ "input = tokenizer(input, return_tensors='pt')\n",
436
+ "\n",
437
+ "output = model.generate(\n",
438
+ " input['input_ids'].to('cuda'),\n",
439
+ " max_length=50, # Maximum length of the generated text\n",
440
+ " num_return_sequences=1, # Number of sequences to generate\n",
441
+ ")"
442
+ ]
443
+ },
444
+ {
445
+ "cell_type": "code",
446
+ "execution_count": 23,
447
+ "id": "139e9973-003a-484f-95f8-42428dd436f5",
448
+ "metadata": {},
449
+ "outputs": [
450
+ {
451
+ "name": "stdout",
452
+ "output_type": "stream",
453
+ "text": [
454
+ "Tell me about cats.\n",
455
+ "\n",
456
+ "Chatbot: Cats are fascinating creatures! They are known for their agility, independence, and unique behaviors. They have a keen sense of hearing and can see well in low light\n"
457
+ ]
458
+ }
459
+ ],
460
+ "source": [
461
+ "generated_text = tokenizer.decode(output[0], skip_special_tokens=True)\n",
462
+ "\n",
463
+ "print(generated_text)"
464
+ ]
465
+ },
466
+ {
467
+ "cell_type": "code",
468
+ "execution_count": null,
469
+ "id": "6dc4ddb3-3cbf-4d6e-9f57-45acb8acbe25",
470
+ "metadata": {},
471
+ "outputs": [],
472
+ "source": []
473
+ }
474
+ ],
475
+ "metadata": {
476
+ "kernelspec": {
477
+ "display_name": "Python 3 (ipykernel)",
478
+ "language": "python",
479
+ "name": "python3"
480
+ },
481
+ "language_info": {
482
+ "codemirror_mode": {
483
+ "name": "ipython",
484
+ "version": 3
485
+ },
486
+ "file_extension": ".py",
487
+ "mimetype": "text/x-python",
488
+ "name": "python",
489
+ "nbconvert_exporter": "python",
490
+ "pygments_lexer": "ipython3",
491
+ "version": "3.10.13"
492
+ }
493
+ },
494
+ "nbformat": 4,
495
+ "nbformat_minor": 5
496
+ }