Upload smaller-completion.ipynb

smaller-completion.ipynb

@@ -0,0 +1,385 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": 1,
+   "id": "66b70728",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "Loading the tokenizer from the `special_tokens_map.json` and the `added_tokens.json` will be removed in `transformers 5`,  it is kept for forward compatibility, but it is recommended to update your `tokenizer_config.json` by uploading it again. You will see the new `added_tokens_decoder` attribute that will store the relevant information.\n"
+     ]
+    }
+   ],
+   "source": [
+    "from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig\n",
+    "import torch\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained('meta-llama/Llama-2-13b-hf')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "id": "d608fb34",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "nf4_config = BitsAndBytesConfig(\n",
+    "   load_in_4bit=True,\n",
+    "   bnb_4bit_quant_type='nf4',\n",
+    "   bnb_4bit_use_double_quant=True,\n",
+    "   bnb_4bit_compute_dtype=torch.bfloat16\n",
+    ")"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 3,
+   "id": "4d63a1ec",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "90b0a965a3c1414aba49e10f0f6612f2",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)lve/main/config.json:   0%|          | 0.00/630 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "[2023-09-28 14:06:37,070] [INFO] [real_accelerator.py:158:get_accelerator] Setting ds_accelerator to cuda (auto detect)\n"
+     ]
+    },
+    {
+     "name": "stderr",
+     "output_type": "stream",
+     "text": [
+      "2023-09-28 14:06:40.012825: I tensorflow/core/platform/cpu_feature_guard.cc:182] This TensorFlow binary is optimized to use available CPU instructions in performance-critical operations.\n",
+      "To enable the following instructions: AVX2 FMA, in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
+      "2023-09-28 14:06:40.724196: W tensorflow/compiler/tf2tensorrt/utils/py_utils.cc:38] TF-TRT Warning: Could not find TensorRT\n"
+     ]
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "b8afc07efe184ac9b051a72c74a8d8c6",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading model.safetensors:   0%|          | 0.00/1.33G [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "a8e37dc88d2745139d1c9c6f65cbb548",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)neration_config.json:   0%|          | 0.00/132 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "model_600m = AutoModelForCausalLM.from_pretrained('mesolitica/llama-600m-hf-32768-fpf', quantization_config=nf4_config)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 4,
+   "id": "5e40d735",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "5391a15d92de4895bfbf6579b4d31670",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)lve/main/config.json:   0%|          | 0.00/628 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "e50f5485e13946ba90f3fc10a71edb4a",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading model.safetensors:   0%|          | 0.00/2.14G [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "28aabf8c0e484c319c4f09cb33d62ce7",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)neration_config.json:   0%|          | 0.00/132 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "model_1b = AutoModelForCausalLM.from_pretrained('mesolitica/llama-1b-hf-32768-fpf', quantization_config=nf4_config)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 5,
+   "id": "90c73899",
+   "metadata": {},
+   "outputs": [
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "e2cdc958ceaf477d8829f9f3db58a0d8",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)lve/main/config.json:   0%|          | 0.00/628 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "f29f83231f9845a28707ca61363e7870",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading model.safetensors:   0%|          | 0.00/3.83G [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    },
+    {
+     "data": {
+      "application/vnd.jupyter.widget-view+json": {
+       "model_id": "10800e991ee9406c808d60b7e5402ee5",
+       "version_major": 2,
+       "version_minor": 0
+      },
+      "text/plain": [
+       "Downloading (…)neration_config.json:   0%|          | 0.00/132 [00:00<?, ?B/s]"
+      ]
+     },
+     "metadata": {},
+     "output_type": "display_data"
+    }
+   ],
+   "source": [
+    "model_2b = AutoModelForCausalLM.from_pretrained('mesolitica/llama-2b-hf-32768-fpf', quantization_config=nf4_config)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 18,
+   "id": "af485f12",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import time\n",
+    "from tqdm import tqdm\n",
+    "\n",
+    "kwargs = {\n",
+    "    'temperature': 0.9, \n",
+    "    'max_new_tokens': 512, \n",
+    "    'top_p': 0.95,\n",
+    "    'repetition_penalty': 1.1, \n",
+    "    'do_sample': True,\n",
+    "    'num_beams': 1,\n",
+    "}"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 19,
+   "id": "8209faad",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "inputs = tokenizer(['Kerajaan didakwa terdesak kaitkan'], return_tensors='pt').to('cuda')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 20,
+   "id": "07251c54",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "generate_kwargs = dict(inputs)\n",
+    "generate_kwargs = {**generate_kwargs, **kwargs}"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "be287c4b",
+   "metadata": {},
+   "source": [
+    "## 600M"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 23,
+   "id": "4711a5e7",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "o = model_600m.generate(**generate_kwargs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 24,
+   "id": "f891dbbb",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Kerajaan didakwa terdesak kaitkan dengan keputusan Majlis Peguam Negara (AGC) mengarahkan jumlah mangsa banjir di lima lokasi banjir tersebut. Terdapat bekas Menteri Undang-undang Persekutuan, Datuk Seri Ismail Sabri Yaakob semalam memuat naik semula sehingga musim tengkujuh dari beberapa kumpulan yang menyebabkan terdapat ribut di kawasan luar bandar ini. “Saya ingin berkongsi tarikh pendakwanya kerana mencecah 69 tahun lalu berbanding 1,239 orang daripada 50 kaki manakala sebanyak RM23 juta,” katanya dalam satu hantaran di Facebook rasmi bersama warga emas dan 48 tahun hampir sebulan lepas. Beliau memberitahu, perinciannya mempunyai kadar vaksinasi antara Januari depan dan Jun ini serta pengurup terkenal dengan usaha memastikan tiada penjarakkan sosial itu dituruti kegiatan banjir (FELDA) walaupun negeri tidak menerima dua dos vaksin. Bagaimanapun, pihaknya akan melakukan kegiatan haram tertunggu-tunggu selain kira-kira 3,792 atau 85.19 malam ini. – MalaysiaGazette\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(tokenizer.decode(o[0], skip_special_tokens = True).split('[/INST]')[-1].strip())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5ce2c89a",
+   "metadata": {},
+   "source": [
+    "## 1B"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 25,
+   "id": "25999783",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "o = model_1b.generate(**generate_kwargs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 26,
+   "id": "ffed0ef2",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Kerajaan didakwa terdesak kaitkan seks luar jika samun PASIR MAS – Seorang pemimpin DAP mengecam tindakan kerajaan yang dilihat mengawal perlakuan tidak senonoh terhadap Pas dan parti Islam itu kerana melakukan serangan di Kuala Terengganu. Nasihat Pas hari ini, Datuk Seri Abdul Hadi Awang berkata, perbuatan sedemikian tidak boleh disamun sekiranya keadaan jenayah itu terus berlaku dalam masyarakat. “Apa sahaja isu jenayah seperti ini akan dimudaratkan lagi dengan anggapan sempit dan kotor,” katanya dipetik daripada portal Sinar Harian. Ia susulan tuduhan berkaitan jenayah indeks kerjasama antarabangsa dengan Barisan Nasional (BN) selain Pakatan Rakyat membabitkan kerjasama dengan Perikatan Nasional (PN). Sinar Harian difahamkan polis turut menahan dua wanita yang disyaki terlibat dengan sindiket terserempak di Kampung Bukit Malawati, Chini, dekat sini. Menurutnya, mereka merupakan rakyat Malaysia pertama yang ditahan polis selepas didakwa cuba menjerit namun gagal kerana kerasnya memadam api dengan kenderaan lain. Terdahulu, peguam kanan persekutuan Shamsul Bolhassan dan Saravanan Muthusamy mewakili Menteri Pembangunan Usahawan Mohd Redzuan Yusof memberitahu Malaysiakini bahawa Polis Diraja Malaysia (PDRM) telah membuka siasatan ke atas pencerobohan kumpulan militan Abu Sayyaf kepada beberapa pihak. Malaysiakini\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(tokenizer.decode(o[0], skip_special_tokens = True).split('[/INST]')[-1].strip())"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "a8ed43ec",
+   "metadata": {},
+   "source": [
+    "## 2B"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 27,
+   "id": "6cd30bbb",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "o = model_2b.generate(**generate_kwargs)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 28,
+   "id": "2c8b086b",
+   "metadata": {},
+   "outputs": [
+    {
+     "name": "stdout",
+     "output_type": "stream",
+     "text": [
+      "Kerajaan didakwa terdesak kaitkan isu tiket loteri KUALA LUMPUR, Dec 13 — The Attorney General’s Chambers (AGC) akan membuka semula pertuduhan baharu bagi pihak yang melibatkan kerajaan-kerajaan negeri dalam kejadian samun dua unit loteri awam. Selain itu, kesalahan mengedar bahan lucah dan seksyen 96 Akta Kanun Pelesenan Perniagaan 1956 (Akta 128) serta Seksyen 17(2) Akta Suruhanjaya Komunikasi dan Multimedia (AKMM). Menteri di Jabatan Perdana Menteri (Tugas-tugas Khas), Datuk Mohd. Shafie Apdal berkata beliau sendiri mendapati pertuduhan yang dikenakan terhadap kerajaan adalah lebih afdal. “Setelah selesai perbicaraan saya meminta maklum balas daripada pihak pendakwa raya sama ada mahu meneruskan atau tidak seseorang yang telah berdepan dengan tuduhan sebegitu melakukannya,” katanya kepada media di lobi Parlimen hari ini. Beliau menjawab soalan Ahmad Maslan (BN-Pontian) dan Timbalan Menteri Di Jabatan Perdana Menteri (Hal Ehwal Ekonomi), Datuk Dr. Abdul Latiff Ahmad. Pengguna Twitter @maslmydan94 dibenarkan membuat sebarang ulasan atau pandangan, namun beliau mengaku sebagai warganegara Malaysia, penyokong parti politik, dan juga orang awam yang berada di luar Kuala Lumpur. Menteri di Jabatan Perdana Menteri (Hal Ehwal Ekonomi), Datuk Seri Azalina Othman Said hari ini dilapor akan mengetuai\n"
+     ]
+    }
+   ],
+   "source": [
+    "print(tokenizer.decode(o[0], skip_special_tokens = True).split('[/INST]')[-1].strip())"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python 3 (ipykernel)",
+   "language": "python",
+   "name": "python3"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.12"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}