Step 2.5 Fine Tune model

notebooks/demo-nm/fine-tune-model.ipynb

@@ -0,0 +1,104 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Fine Tune Model"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Neural Magic hosts many models, including BERT models sparsified on an English text corpus with a masked language modeling training scheme. These sparse models can then finetune to your dataset like you normally would with a dense model using SparseML and recipes."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# This import will install a sparsification compatible version of transformers for SparseML.\n",
+    "from sparseml.transformers import question_answering, export_onnx"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Load Dataset"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#TODO: Load dataset."
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Fine Tune Model"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "question_answering(\n",
+    "    output_dir=\"sparse_quantized_bert-question_answering_squad\",\n",
+    "    model_name_or_path=\"zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/12layer_pruned80_quant-none-vnni\",  #noqa: E501\n",
+    "    recipe=\"zoo:nlp/masked_language_modeling/bert-base/pytorch/huggingface/wikipedia_bookcorpus/12layer_pruned80_quant-none-vnni?recipe_type=transfer-question_answering\",  #noqa: E501\n",
+    "    distill_teacher=\"path/to/dense-teacher-trained\",\n",
+    "    dataset_name=\"custom-dataset\",\n",
+    "    do_train=True,\n",
+    "    do_eval=True,\n",
+    "    evaluation_strategy=\"epoch\",\n",
+    "    fp16=True,\n",
+    "    seed=21636,\n",
+    "    per_device_train_batch_size=16,\n",
+    "    per_device_eval_batch_size=24,\n",
+    "    preprocessing_num_workers=6,\n",
+    "    save_strategy=\"epoch\",\n",
+    "    save_total_limit=1\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "Store model in ONNX format for deepsparse engine"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "export_onnx(\n",
+    "    model_path=\"sparse_quantized_bert-text_classification_squad\",\n",
+    "    task='question-answering',\n",
+    "    sequence_length=128\n",
+    ")"
+   ]
+  }
+ ],
+ "metadata": {
+  "language_info": {
+   "name": "python"
+  },
+  "orig_nbformat": 4
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

overlays/nm-fine-tune/Pipfile

@@ -0,0 +1,15 @@
+[[source]]
+url = "https://pypi.org/simple"
+verify_ssl = true
+name = "pypi"
+
+[packages]
+sparsezoo = "*"
+sparseml = "*"
+torch = "<1.9"
+boto3 = "*"
+
+[dev-packages]
+
+[requires]
+python_version = "3.8"