Spaces:
Runtime error
Runtime error
create: new pay file from collar
Browse files
psy.ipynb
CHANGED
@@ -1,20 +1,83 @@
|
|
1 |
{
|
2 |
-
"
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
|
|
|
|
|
|
|
|
7 |
},
|
8 |
"language_info": {
|
9 |
-
|
10 |
-
"name" : "the programming language of the kernel",
|
11 |
-
"version": "the version of the language",
|
12 |
-
"codemirror_mode": "The name of the codemirror mode to use [optional]"
|
13 |
}
|
14 |
},
|
15 |
-
"
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
20 |
}
|
|
|
1 |
{
|
2 |
+
"nbformat": 4,
|
3 |
+
"nbformat_minor": 0,
|
4 |
+
"metadata": {
|
5 |
+
"colab": {
|
6 |
+
"provenance": []
|
7 |
+
},
|
8 |
+
"kernelspec": {
|
9 |
+
"name": "python3",
|
10 |
+
"display_name": "Python 3"
|
11 |
},
|
12 |
"language_info": {
|
13 |
+
"name": "python"
|
|
|
|
|
|
|
14 |
}
|
15 |
},
|
16 |
+
"cells": [
|
17 |
+
{
|
18 |
+
"cell_type": "code",
|
19 |
+
"execution_count": null,
|
20 |
+
"metadata": {
|
21 |
+
"id": "IAGKskIWS9C0"
|
22 |
+
},
|
23 |
+
"outputs": [],
|
24 |
+
"source": [
|
25 |
+
"from datasets import load_dataset\n",
|
26 |
+
"from transformers import AutoTokenizer, AutoModelForSequenceClassification, TrainingArguments, Trainer\n",
|
27 |
+
"import numpy as np\n",
|
28 |
+
"import evaluate\n",
|
29 |
+
"\n",
|
30 |
+
"\n",
|
31 |
+
"DATA_SEED = 9843203\n",
|
32 |
+
"QUICK_TEST = True\n",
|
33 |
+
"\n",
|
34 |
+
"# This is our baseline dataset\n",
|
35 |
+
"dataset = load_dataset(\"ClaudiaRichard/mbti_classification_v2\")\n",
|
36 |
+
"\n",
|
37 |
+
"# LLama3 8b\n",
|
38 |
+
"tokeniser = AutoTokenizer.from_pretrained(\"meta-llama/Meta-Llama-3-8B\")\n",
|
39 |
+
"\n",
|
40 |
+
"def tokenise_function(examples):\n",
|
41 |
+
" return tokeniser(examples[\"text\"], padding=\"max_length\", truncation=True)\n",
|
42 |
+
"\n",
|
43 |
+
"tokenised_dataset = dataset.map(tokenise_function, batched=True)\n",
|
44 |
+
"\n",
|
45 |
+
"\n",
|
46 |
+
"# Different sized datasets will allow for different training times\n",
|
47 |
+
"train_dataset = tokenised_datasets[\"train\"].shuffle(seed=DATA_SEED).select(range(1000)) if QUICK_TEST else tokenised_datasets[\"train\"].shuffle(seed=DATA_SEED)\n",
|
48 |
+
"test_dataset = tokenised_datasets[\"test\"].shuffle(seed=DATA_SEED).select(range(1000)) if QUICK_TEST else tokenised_datasets[\"test\"].shuffle(seed=DATA_SEED)\n",
|
49 |
+
"\n",
|
50 |
+
"\n",
|
51 |
+
"# Each of our Mtbi types has a specific label here\n",
|
52 |
+
"model = AutoModelForSequenceClassification.from_pretrained(\"meta-llama/Meta-Llama-3-8B\", num_labels=16)\n",
|
53 |
+
"\n",
|
54 |
+
"# Using default hyperparameters at the moment\n",
|
55 |
+
"training_args = TrainingArguments(output_dir=\"test_trainer\")\n",
|
56 |
+
"\n",
|
57 |
+
"# A default metric for checking accuracy\n",
|
58 |
+
"metric = evaluate.load(\"accuracy\")\n",
|
59 |
+
"\n",
|
60 |
+
"def compute_metrics(eval_pred):\n",
|
61 |
+
" logits, labels = eval_pred\n",
|
62 |
+
" predictions = np.argmax(logits, axis=-1)\n",
|
63 |
+
" return metric.compute(predictions=predictions, references=labels)\n",
|
64 |
+
"\n",
|
65 |
+
"# Extract arguments from training\n",
|
66 |
+
"training_args = TrainingArguments(output_dir=\"test_trainer\", evaluation_strategy=\"epoch\")\n",
|
67 |
+
"\n",
|
68 |
+
"# Builds a training object using previously defined data\n",
|
69 |
+
"trainer = Trainer(\n",
|
70 |
+
" model=model,\n",
|
71 |
+
" args=training_args,\n",
|
72 |
+
" train_dataset=train_dataset,\n",
|
73 |
+
" eval_dataset=test_dataset,\n",
|
74 |
+
" compute_metrics=compute_metrics,\n",
|
75 |
+
")\n",
|
76 |
+
"\n",
|
77 |
+
"# Finally, fine-tune!\n",
|
78 |
+
"if __name__ == \"__main__\":\n",
|
79 |
+
" trainer.train()"
|
80 |
+
]
|
81 |
+
}
|
82 |
+
]
|
83 |
}
|