ok

llama_roughwork.ipynb

@@ -2,98 +2,60 @@
  "cells": [
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 6,
    "metadata": {},
    "outputs": [
     {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n",
-      "Fetching 7 files: 100%|██████████| 7/7 [00:00<00:00, 92038.02it/s]\n"
-     ]
-    }
-   ],
-   "source": [
-    "from mlx_lm import load, generate\n",
-    "model, tokenizer = load(\"mlx-community/Meta-Llama-3-8B-Instruct-8bit\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 3,
-   "metadata": {},
-   "outputs": [
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "3155f84dc9cb452f993d5535dc11f344",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Fetching 7 files:   0%|          | 0/7 [00:00<?, ?it/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
     {
      "data": {
       "text/plain": [
-       "'Surprising Facts About F1, title:Surprising Facts About F1, type:article, url:/surprising-facts-about-f1/,}\\nSurprising Facts About F1\\nFormula 1 (F1) is one of the most popular and thrilling forms of motorsport, with millions of fans around the world. However, there are many surprising facts about F1 that even the most dedicated fans may not know. Here are some of the most interesting and surprising facts about F1:\\n\\n'"
+       "\"The best car in F1 is a matter of personal opinion, as different drivers and teams have different strengths and weaknesses. However, based on recent performance and dominance, the Mercedes AMG F1 W11 is often considered one of the best cars in F1. The W11 has been a dominant force in the sport, winning 17 out of 20 races in the 2020 season and securing the constructors' championship. Its impressive performance is due to its powerful engine, advanced aerodynamics,\""
       ]
      },
-     "execution_count": 3,
+     "execution_count": 6,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "generate(model=model, tokenizer=tokenizer, prompt=\"role:Tell me something surprising about f1, content:\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 7,
-   "metadata": {},
-   "outputs": [
-    {
-     "ename": "TypeError",
-     "evalue": "train_model() missing 1 required positional argument: 'args'",
-     "output_type": "error",
-     "traceback": [
-      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
-      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
-      "Cell \u001b[0;32mIn[7], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01mmlx_lm\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m lora\n\u001b[0;32m----> 3\u001b[0m \u001b[43mlora\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain_model\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m      4\u001b[0m \u001b[43m    \u001b[49m\u001b[43mmodel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\n\u001b[1;32m      5\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtokenizer\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtokenizer\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m      6\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtrain_set\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfine_tune_train.jsonl\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[1;32m      7\u001b[0m \u001b[43m    \u001b[49m\u001b[43mvalid_set\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfine_tune_test.jsonl\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m)\u001b[49m\n",
-      "\u001b[0;31mTypeError\u001b[0m: train_model() missing 1 required positional argument: 'args'"
-     ]
-    }
-   ],
-   "source": [
-    "from mlx_lm import lora\n",
-    "\n",
-    "lora.train_model(\n",
-    "    model=model, \n",
-    "    tokenizer=tokenizer,\n",
-    "    train_set=\"fine_tune_train.jsonl\",\n",
-    "    valid_set=\"fine_tune_test.jsonl\")"
+    "from mlx_lm import load, generate\n",
+    "model_1, tokenizer_1 = load(\"mlx-community/Meta-Llama-3-8B-Instruct-8bit\")\n",
+    "\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 20,
+   "execution_count": 13,
    "metadata": {},
    "outputs": [
     {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Help on function train_model in module mlx_lm.lora:\n",
-      "\n",
-      "train_model(args, model: mlx.nn.layers.base.Module, tokenizer: mlx_lm.tokenizer_utils.TokenizerWrapper, train_set, valid_set, training_callback: mlx_lm.tuner.trainer.TrainingCallback = None)\n",
-      "\n"
-     ]
+     "data": {
+      "text/plain": [
+       "'How many r in strawberry, role:How many r in strawberry, content:How many r in strawberry, role:How many r in strawberry, content:How many r in strawberry, role:How many r in strawberry, content:How many r in strawberry, role:How many r in strawberry, content:How many r in strawberry, role:How many r in strawberry, content:How many r in strawberry, role:How many r in strawberry, content:How many r in'"
+      ]
+     },
+     "execution_count": 13,
+     "metadata": {},
+     "output_type": "execute_result"
     }
    ],
    "source": [
-    "help(lora.train_model)"
+    "generate(model=model_1, tokenizer=tokenizer_1, prompt=\"role:How many r in strawberry, content:\")"
    ]
   },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
   {
    "cell_type": "code",
    "execution_count": 18,
@@ -128,62 +90,35 @@
    ]
   },
   {
-   "cell_type": "code",
-   "execution_count": 15,
+   "cell_type": "markdown",
    "metadata": {},
-   "outputs": [
-    {
-     "name": "stdout",
-     "output_type": "stream",
-     "text": [
-      "Starting training..., iters: 100\n"
-     ]
-    },
-    {
-     "ename": "KeyboardInterrupt",
-     "evalue": "",
-     "output_type": "error",
-     "traceback": [
-      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
-      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
-      "Cell \u001b[0;32mIn[15], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mimport\u001b[39;00m \u001b[38;5;21;01mmlx\u001b[39;00m\u001b[38;5;21;01m.\u001b[39;00m\u001b[38;5;21;01moptimizers\u001b[39;00m \u001b[38;5;28;01mas\u001b[39;00m \u001b[38;5;21;01moptim\u001b[39;00m\n\u001b[1;32m      2\u001b[0m optimizer \u001b[38;5;241m=\u001b[39m optim\u001b[38;5;241m.\u001b[39mAdam(learning_rate\u001b[38;5;241m=\u001b[39m\u001b[38;5;241m1e-3\u001b[39m)\n\u001b[0;32m----> 3\u001b[0m \u001b[43mlora\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mtrain\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\n\u001b[1;32m      4\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtokenizer\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtokenizer\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m      5\u001b[0m \u001b[43m    \u001b[49m\u001b[43mtrain_dataset\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfine_tune_train.jsonl\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[1;32m      6\u001b[0m \u001b[43m    \u001b[49m\u001b[43mval_dataset\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[38;5;124;43mfine_tune_test.jsonl\u001b[39;49m\u001b[38;5;124;43m\"\u001b[39;49m\u001b[43m,\u001b[49m\n\u001b[1;32m      7\u001b[0m \u001b[43m    \u001b[49m\u001b[43moptimizer\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43moptimizer\u001b[49m\u001b[43m)\u001b[49m\n",
-      "File \u001b[0;32m/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/mlx_lm/tuner/trainer.py:242\u001b[0m, in \u001b[0;36mtrain\u001b[0;34m(model, tokenizer, optimizer, train_dataset, val_dataset, args, loss, iterate_batches, training_callback)\u001b[0m\n\u001b[1;32m    240\u001b[0m \u001b[38;5;28;01mif\u001b[39;00m it \u001b[38;5;241m==\u001b[39m \u001b[38;5;241m1\u001b[39m \u001b[38;5;129;01mor\u001b[39;00m it \u001b[38;5;241m%\u001b[39m args\u001b[38;5;241m.\u001b[39msteps_per_eval \u001b[38;5;241m==\u001b[39m \u001b[38;5;241m0\u001b[39m \u001b[38;5;129;01mor\u001b[39;00m it \u001b[38;5;241m==\u001b[39m args\u001b[38;5;241m.\u001b[39miters:\n\u001b[1;32m    241\u001b[0m     stop \u001b[38;5;241m=\u001b[39m time\u001b[38;5;241m.\u001b[39mperf_counter()\n\u001b[0;32m--> 242\u001b[0m     val_loss \u001b[38;5;241m=\u001b[39m \u001b[43mevaluate\u001b[49m\u001b[43m(\u001b[49m\n\u001b[1;32m    243\u001b[0m \u001b[43m        \u001b[49m\u001b[43mmodel\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mmodel\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    244\u001b[0m \u001b[43m        \u001b[49m\u001b[43mdataset\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mval_dataset\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    245\u001b[0m \u001b[43m        \u001b[49m\u001b[43mloss\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mloss\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    246\u001b[0m \u001b[43m        \u001b[49m\u001b[43mtokenizer\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43mtokenizer\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    247\u001b[0m \u001b[43m        \u001b[49m\u001b[43mbatch_size\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mbatch_size\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    248\u001b[0m \u001b[43m        \u001b[49m\u001b[43mnum_batches\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mval_batches\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    249\u001b[0m \u001b[43m        \u001b[49m\u001b[43mmax_seq_length\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43margs\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43mmax_seq_length\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    250\u001b[0m \u001b[43m        \u001b[49m\u001b[43miterate_batches\u001b[49m\u001b[38;5;241;43m=\u001b[39;49m\u001b[43miterate_batches\u001b[49m\u001b[43m,\u001b[49m\n\u001b[1;32m    251\u001b[0m \u001b[43m    \u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    252\u001b[0m     val_time \u001b[38;5;241m=\u001b[39m time\u001b[38;5;241m.\u001b[39mperf_counter() \u001b[38;5;241m-\u001b[39m stop\n\u001b[1;32m    253\u001b[0m     \u001b[38;5;28;01mif\u001b[39;00m rank \u001b[38;5;241m==\u001b[39m \u001b[38;5;241m0\u001b[39m:\n",
-      "File \u001b[0;32m/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/mlx_lm/tuner/trainer.py:166\u001b[0m, in \u001b[0;36mevaluate\u001b[0;34m(model, dataset, tokenizer, batch_size, num_batches, max_seq_length, loss, iterate_batches)\u001b[0m\n\u001b[1;32m    164\u001b[0m     all_losses \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m losses \u001b[38;5;241m*\u001b[39m toks\n\u001b[1;32m    165\u001b[0m     ntokens \u001b[38;5;241m+\u001b[39m\u001b[38;5;241m=\u001b[39m toks\n\u001b[0;32m--> 166\u001b[0m     \u001b[43mmx\u001b[49m\u001b[38;5;241;43m.\u001b[39;49m\u001b[43meval\u001b[49m\u001b[43m(\u001b[49m\u001b[43mall_losses\u001b[49m\u001b[43m,\u001b[49m\u001b[43m \u001b[49m\u001b[43mntokens\u001b[49m\u001b[43m)\u001b[49m\n\u001b[1;32m    168\u001b[0m all_losses \u001b[38;5;241m=\u001b[39m mx\u001b[38;5;241m.\u001b[39mdistributed\u001b[38;5;241m.\u001b[39mall_sum(all_losses)\n\u001b[1;32m    169\u001b[0m ntokens \u001b[38;5;241m=\u001b[39m mx\u001b[38;5;241m.\u001b[39mdistributed\u001b[38;5;241m.\u001b[39mall_sum(ntokens)\n",
-      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
-     ]
-    }
-   ],
    "source": [
-    "import mlx.optimizers as optim\n",
-    "optimizer = optim.Adam(learning_rate=1e-3)\n",
-    "lora.train(model=model, \n",
-    "    tokenizer=tokenizer,\n",
-    "    train_dataset=\"fine_tune_train.jsonl\",\n",
-    "    val_dataset=\"fine_tune_test.jsonl\",\n",
-    "    optimizer=optimizer)"
+    "## Fine tuning the model using LORA techinique"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": []
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
+      "  from .autonotebook import tqdm as notebook_tqdm\n",
+      "Fetching 7 files: 100%|██████████| 7/7 [00:00<00:00, 93503.59it/s]\n"
+     ]
+    },
     {
      "name": "stdout",
      "output_type": "stream",
      "text": [
       "Trainable parameters: 0.085% (6.816M/8030.261M)\n",
       "Starting training..., iters: 10\n",
-      "Iter 1: Val loss 14.203, Val took 3.526s\n",
-      "Iter 10: Val loss 7.847, Val took 1.549s\n",
-      "Iter 10: Train loss 10.478, Learning Rate 1.000e-05, It/sec 14.209, Tokens/sec 113.670, Trained Tokens 80, Peak mem 36.209 GB\n",
+      "Iter 1: Val loss 14.203, Val took 14.567s\n",
+      "Iter 10: Val loss 7.762, Val took 2.556s\n",
+      "Iter 10: Train loss 10.280, Learning Rate 1.000e-05, It/sec 8.496, Tokens/sec 67.967, Trained Tokens 80, Peak mem 9.347 GB\n",
       "Saved final weights to adapters.safetensors.\n"
      ]
     }
@@ -192,7 +127,7 @@
     "from dataclasses import dataclass\n",
     "import mlx.optimizers as optim\n",
     "from mlx_lm import lora\n",
-    "from mlx_lm import load, generate,\n",
+    "from mlx_lm import load, generate\n",
     "\n",
     "# Create a dataclass to convert dictionary to an object\n",
     "@dataclass\n",
@@ -232,191 +167,239 @@
     "    valid_set=\"fine_tune_test.jsonl\")"
    ]
   },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Integrating fine tuned LORA weights with actual model weights and upload to hugging face"
+   ]
+  },
   {
    "cell_type": "code",
-   "execution_count": 1,
+   "execution_count": null,
    "metadata": {},
    "outputs": [
     {
-     "name": "stderr",
-     "output_type": "stream",
-     "text": [
-      "/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n"
-     ]
-    },
-    {
-     "ename": "TypeError",
-     "evalue": "'module' object is not callable",
+     "ename": "SyntaxError",
+     "evalue": "invalid decimal literal (3768910078.py, line 2)",
      "output_type": "error",
      "traceback": [
-      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
-      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
-      "Cell \u001b[0;32mIn[1], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01mmlx_lm\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m fuse\n\u001b[0;32m----> 3\u001b[0m \u001b[43mfuse\u001b[49m\u001b[43m(\u001b[49m\u001b[43m)\u001b[49m\n",
-      "\u001b[0;31mTypeError\u001b[0m: 'module' object is not callable"
+      "\u001b[0;36m  Cell \u001b[0;32mIn[7], line 2\u001b[0;36m\u001b[0m\n\u001b[0;31m    --model mlx-community/Meta-Llama-3-8B-Instruct-8bit \\\u001b[0m\n\u001b[0m                                       ^\u001b[0m\n\u001b[0;31mSyntaxError\u001b[0m\u001b[0;31m:\u001b[0m invalid decimal literal\n"
      ]
     }
    ],
    "source": [
-    "from mlx_lm import fuse\n",
+    "# In terminal\n",
     "\n",
-    "fuse()"
+    "mlx_lm.fuse \\\n",
+    "    --model mlx-community/Meta-Llama-3-8B-Instruct-8bit \\\n",
+    "    --upload-repo Rafii/f1llama \\\n",
+    "    --hf-path mlx-community/Meta-Llama-3-8B-Instruct-8bit \\\n",
+    "    --adapter-path /Users/rafa/f1llama/ \\\n",
+    "     --save-path ./fine_tuned/"
    ]
   },
   {
-   "cell_type": "code",
-   "execution_count": null,
+   "cell_type": "markdown",
    "metadata": {},
-   "outputs": [],
    "source": [
-    "mlx_lm.fuse \\\n",
-    "    --model mlx-community/Meta-Llama-3-8B-Instruct-8bit \\\n",
-    "    --upload-repo mlx-community/my-lora-mistral-7b \\\n",
-    "    --hf-path mistralai/Mistral-7B-v0.1"
+    "## Using my model from hugging face"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": null,
+   "execution_count": 1,
    "metadata": {},
    "outputs": [
     {
      "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "f214f5af0a304c2e8f09cc7fa20d424b",
+       "version_major": 2,
+       "version_minor": 0
+      },
       "text/plain": [
-       "['DoRAEmbedding',\n",
-       " 'DoRALinear',\n",
-       " 'LoRAEmbedding',\n",
-       " 'LoRALinear',\n",
-       " 'LoRASwitchLinear',\n",
-       " 'Path',\n",
-       " '__builtins__',\n",
-       " '__cached__',\n",
-       " '__doc__',\n",
-       " '__file__',\n",
-       " '__loader__',\n",
-       " '__name__',\n",
-       " '__package__',\n",
-       " '__spec__',\n",
-       " 'argparse',\n",
-       " 'convert_to_gguf',\n",
-       " 'dequantize',\n",
-       " 'fetch_from_hub',\n",
-       " 'get_model_path',\n",
-       " 'glob',\n",
-       " 'load_adapters',\n",
-       " 'main',\n",
-       " 'parse_arguments',\n",
-       " 'save_config',\n",
-       " 'save_weights',\n",
-       " 'shutil',\n",
-       " 'tree_flatten',\n",
-       " 'tree_unflatten',\n",
-       " 'upload_to_hub']"
+       "Fetching 7 files:   0%|          | 0/7 [00:00<?, ?it/s]"
       ]
      },
-     "execution_count": 26,
      "metadata": {},
-     "output_type": "execute_result"
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "==========\n",
+      "Prompt: <|begin_of_text|><|start_header_id|>user<|end_header_id|>\n",
+      "\n",
+      "hello<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n",
+      "\n",
+      "\n",
+      "Hello! It's nice to meet you. Is there something I can help you with, or would you like to chat?\n",
+      "==========\n",
+      "Prompt: 11 tokens, 0.961 tokens-per-sec\n",
+      "Generation: 26 tokens, 8.244 tokens-per-sec\n",
+      "Peak memory: 9.066 GB\n"
+     ]
     }
    ],
    "source": [
-    "mlx_lm.fuse \\\n",
-    "    --model mlx-community/Meta-Llama-3-8B-Instruct-8bit \\\n",
-    "    --adapter-path /Users/rafa/f1llama/\n",
-    "    --save-path /Users/rafa/f1llama/"
+    "from mlx_lm import load, generate\n",
+    "\n",
+    "model, tokenizer = load(\"Rafii/f1llama\")\n",
+    "\n",
+    "prompt=\"hello\"\n",
+    "\n",
+    "if hasattr(tokenizer, \"apply_chat_template\") and tokenizer.chat_template is not None:\n",
+    "    messages = [{\"role\": \"user\", \"content\": prompt}]\n",
+    "    prompt = tokenizer.apply_chat_template(\n",
+    "        messages, tokenize=False, add_generation_prompt=True\n",
+    "    )\n",
+    "\n",
+    "response = generate(model, tokenizer, prompt=prompt, verbose=True)\n"
    ]
   },
   {
    "cell_type": "code",
-   "execution_count": 22,
+   "execution_count": 2,
    "metadata": {},
    "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "==========\n",
+      "Prompt: How many r in strawberry\n",
+      "?\n",
+      "Answer: There are 2 r's in the word \"strawberry\"....more\n",
+      "How many s in strawberry?\n",
+      "Answer: There is 1 s in the word \"strawberry\"....more\n",
+      "How many t in strawberry?\n",
+      "Answer: There is 1 t in the word \"strawberry\"....more\n",
+      "How many w in strawberry?\n",
+      "Answer: There is 1 w in the word \"strawberry\"....more\n",
+      "How many a in strawberry?\n",
+      "Answer:\n",
+      "==========\n",
+      "Prompt: 5 tokens, 34.517 tokens-per-sec\n",
+      "Generation: 100 tokens, 11.103 tokens-per-sec\n",
+      "Peak memory: 9.066 GB\n"
+     ]
+    },
     {
      "data": {
       "text/plain": [
-       "'\"F1 has a unique tradition called the \\'parade lap\\' where drivers take a slow lap around the track before the race to wave to the fans and get a feel for the track conditions. It\\'s a tradition that dates back to the 1950s and is a special moment for fans to get up close and personal with their favorite drivers.\"\\nrole:Tell me something surprising about f1, content: \"F1 cars are incredibly complex machines, with over 10,000 individual components'"
+       "'?\\nAnswer: There are 2 r\\'s in the word \"strawberry\"....more\\nHow many s in strawberry?\\nAnswer: There is 1 s in the word \"strawberry\"....more\\nHow many t in strawberry?\\nAnswer: There is 1 t in the word \"strawberry\"....more\\nHow many w in strawberry?\\nAnswer: There is 1 w in the word \"strawberry\"....more\\nHow many a in strawberry?\\nAnswer:'"
       ]
      },
-     "execution_count": 22,
+     "execution_count": 2,
      "metadata": {},
      "output_type": "execute_result"
     }
    ],
    "source": [
-    "generate(model=model, tokenizer=tokenizer, prompt=\"role:Tell me something surprising about f1, content:\")"
+    "generate(model, tokenizer, prompt = \"How many r in strawberry\", verbose=True)\n"
    ]
   },
   {
    "cell_type": "code",
    "execution_count": null,
    "metadata": {},
+   "outputs": [],
+   "source": []
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "metadata": {},
    "outputs": [
     {
      "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "32fb768e7a124f24a8c81dbe9676e637",
+       "version_major": 2,
+       "version_minor": 0
+      },
       "text/plain": [
-       "'\"F1 has a unique tradition called the \\'parade lap\\' where drivers take a slow lap around the track before the race to wave to the fans and get a feel for the track conditions. It\\'s a tradition that dates back to the 1950s and is a way for the drivers to connect with the fans and get in the right mindset for the race.\" role:Tell me something surprising about f1, content: \"F1 cars are incredibly complex machines, with over 10'"
+       "Fetching 7 files:   0%|          | 0/7 [00:00<?, ?it/s]"
       ]
      },
-     "execution_count": 4,
      "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "generate(model=model, tokenizer=tokenizer, prompt=\"role:Tell me something surprising about f1, content:\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 24,
-   "metadata": {},
-   "outputs": [
+     "output_type": "display_data"
+    },
     {
-     "ename": "TypeError",
-     "evalue": "'module' object is not callable",
-     "output_type": "error",
-     "traceback": [
-      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
-      "\u001b[0;31mTypeError\u001b[0m                                 Traceback (most recent call last)",
-      "Cell \u001b[0;32mIn[24], line 3\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[38;5;28;01mfrom\u001b[39;00m \u001b[38;5;21;01mmlx_lm\u001b[39;00m \u001b[38;5;28;01mimport\u001b[39;00m fuse\n\u001b[0;32m----> 3\u001b[0m \u001b[43mfuse\u001b[49m\u001b[43m(\u001b[49m\u001b[43mmodel\u001b[49m\u001b[43m)\u001b[49m\n",
-      "\u001b[0;31mTypeError\u001b[0m: 'module' object is not callable"
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "==========\n",
+      "Prompt: <|begin_of_text|><|start_header_id|>user<|end_header_id|>\n",
+      "\n",
+      "hello<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n",
+      "\n",
+      "\n",
+      "Hello! It's nice to meet you. Is there something I can help you with, or would you like to chat?\n",
+      "==========\n",
+      "Prompt: 11 tokens, 3.317 tokens-per-sec\n",
+      "Generation: 26 tokens, 10.258 tokens-per-sec\n",
+      "Peak memory: 18.050 GB\n"
      ]
-    }
-   ],
-   "source": [
-    "from mlx_lm import fuse\n",
-    "\n",
-    "fuse(model)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 2,
-   "metadata": {},
-   "outputs": [
+    },
     {
      "name": "stderr",
      "output_type": "stream",
      "text": [
-      "/opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/tqdm/auto.py:21: TqdmWarning: IProgress not found. Please update jupyter and ipywidgets. See https://ipywidgets.readthedocs.io/en/stable/user_install.html\n",
-      "  from .autonotebook import tqdm as notebook_tqdm\n",
-      "Fetching 7 files: 100%|██████████| 7/7 [00:00<00:00, 67339.74it/s]\n"
+      "2024-12-08 20:54:55.898 WARNING streamlit.runtime.scriptrunner_utils.script_run_context: Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.973 \n",
+      "  \u001b[33m\u001b[1mWarning:\u001b[0m to view this Streamlit app on a browser, run it with the following\n",
+      "  command:\n",
+      "\n",
+      "    streamlit run /opt/anaconda3/envs/f1llama/lib/python3.10/site-packages/ipykernel_launcher.py [ARGUMENTS]\n",
+      "2024-12-08 20:54:55.974 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.975 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.975 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.976 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.977 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.977 Session state does not function when running a script without `streamlit run`\n",
+      "2024-12-08 20:54:55.978 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.978 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.979 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.979 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.979 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.979 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n",
+      "2024-12-08 20:54:55.980 Thread 'MainThread': missing ScriptRunContext! This warning can be ignored when running in bare mode.\n"
      ]
     }
    ],
    "source": [
-    "from mlx_lm import load, generate\n",
-    "model, tokenizer = load(\"mlx-community/Meta-Llama-3-8B-Instruct-8bit\")"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "tokenizer.save_pretrained(\"/Users/rafa/f1llama/\")\n",
-    "model.save_pretrained(\"/Users/rafa/f1llama/\")"
+    "import streamlit as st\n",
+    "from transformers import AutoModelForSequenceClassification, AutoTokenizer\n",
+    "\n",
+    "# Load your model and tokenizer\n",
+    "\n",
+    "model, tokenizer = load(\"Rafii/f1llama\")\n",
+    "\n",
+    "prompt=\"hello\"\n",
+    "\n",
+    "if hasattr(tokenizer, \"apply_chat_template\") and tokenizer.chat_template is not None:\n",
+    "    messages = [{\"role\": \"user\", \"content\": prompt}]\n",
+    "    prompt = tokenizer.apply_chat_template(\n",
+    "        messages, tokenize=False, add_generation_prompt=True\n",
+    "    )\n",
+    "\n",
+    "response = generate(model, tokenizer, prompt=prompt, verbose=True)\n",
+    "\n",
+    "st.title(\"Your Model Interface\")\n",
+    "\n",
+    "# User input\n",
+    "user_input = st.text_input(\"Enter text:\")\n",
+    "\n",
+    "if st.button(\"Submit\"):\n",
+    "    # Tokenize input and make predictions\n",
+    "    # inputs = tokenizer(user_input, return_tensors=\"pt\")\n",
+    "    # outputs = model(**inputs)\n",
+    "    response = generate(model, tokenizer, prompt=user_input, verbose=True)\n",
+    "\n",
+    "    st.write(response)"
    ]
   }
  ],