Fine-tuning model

CARSE_02_FineTuning.ipynb

@@ -50,20 +50,209 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "### Subimos nuestro primer file de entrenamiento"
+    "### Separamos en Training y Validation cada file"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 16,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Función para separar la data\n",
+    "def dividir_training_validation(ruta_archivo, proporcion_training=0.8):\n",
+    "    # Leer todas las líneas del archivo\n",
+    "    with open(ruta_archivo, 'r', encoding='utf-8') as file:\n",
+    "        lineas = file.readlines()\n",
+    "\n",
+    "    # Calcular el punto de corte para el conjunto de entrenamiento\n",
+    "    corte = int(len(lineas) * proporcion_training)\n",
+    "\n",
+    "    # Dividir las líneas en conjuntos de entrenamiento y validación\n",
+    "    lineas_training = lineas[:corte]\n",
+    "    lineas_validation = lineas[corte:]\n",
+    "\n",
+    "    # Crear archivos para training y validation\n",
+    "    ruta_archivo_base = ruta_archivo.replace('.jsonl', '')\n",
+    "    archivo_training = f'{ruta_archivo_base}_train.jsonl'\n",
+    "    archivo_validation = f'{ruta_archivo_base}_val.jsonl'\n",
+    "\n",
+    "    # Escribir el conjunto de entrenamiento\n",
+    "    with open(archivo_training, 'w', encoding='utf-8') as file:\n",
+    "        file.writelines(lineas_training)\n",
+    "\n",
+    "    # Escribir el conjunto de validación\n",
+    "    with open(archivo_validation, 'w', encoding='utf-8') as file:\n",
+    "        file.writelines(lineas_validation)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 17,
    "metadata": {},
    "outputs": [],
    "source": [
-    "#upload_file_1 = client.files.create(\n",
-    "  file=open(\"Training_Data/Training_Prompts_1.jsonl\", \"rb\"),\n",
+    "# Llamadas a la función para crear la separación\n",
+    "\"\"\"\n",
+    "dividir_training_validation('Training_Data/Training_Prompts_1.jsonl')\n",
+    "dividir_training_validation('Training_Data/Training_Prompts_2.jsonl')\n",
+    "dividir_training_validation('Training_Data/Training_Prompts_3.jsonl')\n",
+    "\"\"\""
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Eliminamos los archivos \"originales\"\n",
+    "\"\"\"\n",
+    "os.remove('Training_Data/Training_Prompts_1.jsonl')\n",
+    "os.remove('Training_Data/Training_Prompts_2.jsonl')\n",
+    "os.remove('Training_Data/Training_Prompts_3.jsonl')\n",
+    "\"\"\""
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Subimos files de entrenamiento y validación"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 29,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Para Training\n",
+    "upload_train_response = client.files.create(\n",
+    "  file=open(\"Training_Data/Training_Prompts_1_train.jsonl\", \"rb\"),\n",
+    "  purpose=\"fine-tune\"\n",
+    ")\n",
+    "\n",
+    "# Para Validation\n",
+    "upload_val_response = client.files.create(\n",
+    "  file=open(\"Training_Data/Training_Prompts_1_val.jsonl\", \"rb\"),\n",
     "  purpose=\"fine-tune\"\n",
     ")"
    ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 30,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Training file id:\tfile-eQOE8MxF51oFiGSVT48x0vLw\n",
+      "Validation file id:\tfile-3LSttIrULCZUz5a4pXc3Fsk4\n"
+     ]
+    }
+   ],
+   "source": [
+    "train_file_id = upload_train_response.id\n",
+    "val_file_id = upload_val_response.id\n",
+    "\n",
+    "print(f'Training file id:\\t{train_file_id}')\n",
+    "print(f'Validation file id:\\t{val_file_id}')"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "# Trabajo de fine-tuning"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 32,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "fine_tune_response = client.fine_tuning.jobs.create(\n",
+    "  training_file=\"file-eQOE8MxF51oFiGSVT48x0vLw\", \n",
+    "  validation_file=\"file-3LSttIrULCZUz5a4pXc3Fsk4\",\n",
+    "  model=\"gpt-3.5-turbo-1106\", \n",
+    "  suffix=\"CARSE\",\n",
+    "  hyperparameters={\n",
+    "    \"n_epochs\":5\n",
+    "  }\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 33,
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Fine-tune id:\tftjob-oKdrQdLt4j9ijBvKuQWXl9C9\n"
+     ]
+    }
+   ],
+   "source": [
+    "fine_tune_id = fine_tune_response.id\n",
+    "\n",
+    "print(f'Fine-tune id:\\t{fine_tune_id}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 35,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "FineTuningJob(id='ftjob-oKdrQdLt4j9ijBvKuQWXl9C9', created_at=1702185828, error=None, fine_tuned_model=None, finished_at=None, hyperparameters=Hyperparameters(n_epochs=5, batch_size=1, learning_rate_multiplier=2), model='gpt-3.5-turbo-1106', object='fine_tuning.job', organization_id='org-IXFDgE8ZZcQzb9yKJmEuFxvC', result_files=[], status='running', trained_tokens=None, training_file='file-eQOE8MxF51oFiGSVT48x0vLw', validation_file='file-3LSttIrULCZUz5a4pXc3Fsk4')"
+      ]
+     },
+     "execution_count": 35,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "# Retrieve the state of a fine-tune\n",
+    "client.fine_tuning.jobs.retrieve(fine_tune_id)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 36,
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "text/plain": [
+       "SyncCursorPage[FineTuningJobEvent](data=[FineTuningJobEvent(id='ftevent-HJmomlGcLNxB8tYxEn2fZYge', created_at=1702186772, level='info', message='Step 471/890: training loss=1.55, validation loss=2.50', object='fine_tuning.job.event', data={'step': 471, 'train_loss': 1.5508146286010742, 'valid_loss': 2.498373487721319, 'train_mean_token_accuracy': 0.75, 'valid_mean_token_accuracy': 0.43478260869565216}, type='metrics'), FineTuningJobEvent(id='ftevent-CYGX0JMLoG4Ofjs69dz8MPk9', created_at=1702186756, level='info', message='Step 461/890: training loss=1.16, validation loss=1.79', object='fine_tuning.job.event', data={'step': 461, 'train_loss': 1.156284213066101, 'valid_loss': 1.792006492614746, 'train_mean_token_accuracy': 0.7692307829856873, 'valid_mean_token_accuracy': 0.4}, type='metrics'), FineTuningJobEvent(id='ftevent-mPG7xYX7opqQr9nlHUnn0gOM', created_at=1702186738, level='info', message='Step 451/890: training loss=1.81, validation loss=3.31', object='fine_tuning.job.event', data={'step': 451, 'train_loss': 1.806351900100708, 'valid_loss': 3.3079380989074707, 'train_mean_token_accuracy': 0.5545454621315002, 'valid_mean_token_accuracy': 0.16666666666666666}, type='metrics'), FineTuningJobEvent(id='ftevent-SZnQTTUL8QV1vD3wmIkQ9NQE', created_at=1702186720, level='info', message='Step 441/890: training loss=2.13, validation loss=1.30', object='fine_tuning.job.event', data={'step': 441, 'train_loss': 2.1261494159698486, 'valid_loss': 1.3008832931518555, 'train_mean_token_accuracy': 0.5555555820465088, 'valid_mean_token_accuracy': 0.35}, type='metrics'), FineTuningJobEvent(id='ftevent-Z70r5u6Tv1cpKIjNoz7uxnaB', created_at=1702186702, level='info', message='Step 431/890: training loss=2.31, validation loss=1.63', object='fine_tuning.job.event', data={'step': 431, 'train_loss': 2.3078231811523438, 'valid_loss': 1.6347110454852765, 'train_mean_token_accuracy': 0.5, 'valid_mean_token_accuracy': 0.23076923076923078}, type='metrics'), FineTuningJobEvent(id='ftevent-3FOJQqO26ejvt9jGJv5YtTLS', created_at=1702186686, level='info', message='Step 421/890: training loss=1.73, validation loss=2.02', object='fine_tuning.job.event', data={'step': 421, 'train_loss': 1.7264701128005981, 'valid_loss': 2.0240015983581543, 'train_mean_token_accuracy': 0.6315789222717285, 'valid_mean_token_accuracy': 0.42857142857142855}, type='metrics'), FineTuningJobEvent(id='ftevent-ErMp6gvVT4GBdGf3yFGcAScn', created_at=1702186668, level='info', message='Step 411/890: training loss=1.16, validation loss=1.91', object='fine_tuning.job.event', data={'step': 411, 'train_loss': 1.1630134582519531, 'valid_loss': 1.9137598673502605, 'train_mean_token_accuracy': 0.7058823704719543, 'valid_mean_token_accuracy': 0.5277777777777778}, type='metrics'), FineTuningJobEvent(id='ftevent-OvYkLbxmyf0IDr7RNdSZv5Qr', created_at=1702186652, level='info', message='Step 401/890: training loss=0.76, validation loss=2.10', object='fine_tuning.job.event', data={'step': 401, 'train_loss': 0.7578617334365845, 'valid_loss': 2.0966723442077635, 'train_mean_token_accuracy': 0.7272727489471436, 'valid_mean_token_accuracy': 0.25}, type='metrics'), FineTuningJobEvent(id='ftevent-BJ0JMEcNFGTfCzyXK1R5Uets', created_at=1702186634, level='info', message='Step 391/890: training loss=1.46, validation loss=2.74', object='fine_tuning.job.event', data={'step': 391, 'train_loss': 1.4621750116348267, 'valid_loss': 2.7422271728515626, 'train_mean_token_accuracy': 0.699999988079071, 'valid_mean_token_accuracy': 0.4}, type='metrics'), FineTuningJobEvent(id='ftevent-HuADV2SZtgOS5cBnyN12JnMO', created_at=1702186618, level='info', message='Step 381/890: training loss=1.10, validation loss=2.81', object='fine_tuning.job.event', data={'step': 381, 'train_loss': 1.1008758544921875, 'valid_loss': 2.8111038208007812, 'train_mean_token_accuracy': 0.875, 'valid_mean_token_accuracy': 0.1935483870967742}, type='metrics')], object='list', has_more=True)"
+      ]
+     },
+     "execution_count": 36,
+     "metadata": {},
+     "output_type": "execute_result"
+    }
+   ],
+   "source": [
+    "# List up to 10 events from a fine-tuning job\n",
+    "client.fine_tuning.jobs.list_events(fine_tuning_job_id=fine_tune_id, limit=10)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": []
   }
  ],
  "metadata": {