main.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5060a3f5",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d1f1cb89",
   "metadata": {},
   "outputs": [],
   "source": [
    "# lines = read_lines('.secret')\n",
    "# env = dict(parse_dotenv(lines))\n",
    "# openai.api_key = env['OPENAI_TOKEN']"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "91801c88",
   "metadata": {},
   "source": [
    "# tasks"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "de9e413f",
   "metadata": {},
   "outputs": [],
   "source": [
    "# table = pd.read_excel('tasks/user_oriented_annot.xlsx')\n",
    "# table = table[~table.done.isnull()]\n",
    "# table = table.where(pd.notnull(table), None)\n",
    "\n",
    "# view = table[['id', 'orig_instruction', 'orig_input']]\n",
    "# view = view.rename(columns={\n",
    "#     'orig_instruction': 'instruction',\n",
    "#     'orig_input': 'input'\n",
    "# })\n",
    "# items = view.to_dict(orient='records')\n",
    "# lines = format_jsonl(items)\n",
    "# write_lines('tasks/user_oriented_en.jsonl', lines)\n",
    "\n",
    "# view = table[['id', 'instruction', 'input']]\n",
    "# items = view.to_dict(orient='records')\n",
    "# lines = format_jsonl(items)\n",
    "# write_lines('tasks/user_oriented_ru.jsonl', lines)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4da2a44e",
   "metadata": {},
   "outputs": [],
   "source": [
    "# table = pd.read_excel('tasks/vicuna_question_annot.xlsx')\n",
    "\n",
    "# view = table[['id', 'category', 'instruction']]\n",
    "# items = view.to_dict(orient='records')\n",
    "# lines = format_jsonl(items)\n",
    "# write_lines('tasks/vicuna_question_ru.jsonl', lines)\n",
    "\n",
    "# view = table[['id', 'category', 'orig_instruction']]\n",
    "# view = view.rename(columns={'orig_instruction': 'instruction'})\n",
    "# items = view.to_dict(orient='records')\n",
    "# lines = format_jsonl(items)\n",
    "# write_lines('tasks/vicuna_question_en.jsonl', lines)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "6277a70a",
   "metadata": {},
   "source": [
    "# models"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "75304a05",
   "metadata": {},
   "source": [
    "## gusev_ru_alpaca_7b_lora"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "822527ef",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "model_name = 'IlyaGusev/llama_7b_ru_turbo_alpaca_lora'\n",
    "config = PeftConfig.from_pretrained(model_name)\n",
    "generation_config = GenerationConfig.from_pretrained(model_name)\n",
    "\n",
    "tokenizer = LlamaTokenizer.from_pretrained(model_name)\n",
    "tokenizer.padding_side = 'left'\n",
    "\n",
    "model = LlamaForCausalLM.from_pretrained(\n",
    "    config.base_model_name_or_path,\n",
    "    load_in_8bit=True,\n",
    "\n",
    "    # Overriding torch_dtype=None with `torch_dtype=torch.float16\n",
    "    torch_dtype=torch.float16,\n",
    "    \n",
    "    # A device map needs to be passed to run convert models into mixed-int8\n",
    "    device_map='auto'\n",
    ")\n",
    "model = PeftModel.from_pretrained(\n",
    "    model, model_name,\n",
    "    torch_dtype=torch.float16,\n",
    ").eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "558d3866",
   "metadata": {},
   "source": [
    "## gusev_ru_alpaca_16b_lora"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fd03fbdd",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'IlyaGusev/llama_13b_ru_turbo_alpaca_lora'\n",
    "config = PeftConfig.from_pretrained(model_name)\n",
    "generation_config = GenerationConfig.from_pretrained(model_name)\n",
    "\n",
    "tokenizer = LlamaTokenizer.from_pretrained(model_name)\n",
    "tokenizer.padding_side = 'left'\n",
    "\n",
    "model = LlamaForCausalLM.from_pretrained(\n",
    "    config.base_model_name_or_path,\n",
    "    load_in_8bit=True,\n",
    "    torch_dtype=torch.float16,\n",
    "    device_map='auto'\n",
    ")\n",
    "model = PeftModel.from_pretrained(\n",
    "    model, model_name,\n",
    "    torch_dtype=torch.float16,\n",
    ").eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "316b693c",
   "metadata": {},
   "source": [
    "## gusev_saiga_7b_lora, 13b"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "e42712c6",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'IlyaGusev/saiga_13b_lora'\n",
    "config = PeftConfig.from_pretrained(model_name)\n",
    "generation_config = GenerationConfig.from_pretrained(model_name)\n",
    "generation_config.max_new_tokens = 512\n",
    "\n",
    "tokenizer = LlamaTokenizer.from_pretrained(model_name)\n",
    "tokenizer.padding_side = 'left'\n",
    "\n",
    "model = LlamaForCausalLM.from_pretrained(\n",
    "    config.base_model_name_or_path,\n",
    "    load_in_8bit=True,\n",
    "    torch_dtype=torch.float16,\n",
    "    device_map='auto'\n",
    ")\n",
    "model = PeftModel.from_pretrained(\n",
    "    model, model_name,\n",
    "    torch_dtype=torch.float16,\n",
    ").eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5e6f7fcc",
   "metadata": {},
   "source": [
    "## gusev_ru_alpaca_7b_llamacpp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4b0d9985",
   "metadata": {},
   "outputs": [],
   "source": [
    "home_dir = Path.home()\n",
    "main_path = home_dir / 'src/llama.cpp/main'\n",
    "model_path = home_dir / 'models/llama_7b_ru_turbo_alpaca_lora_llamacpp/7B/ggml-model-q4_0.bin'\n",
    "generation_args = [\n",
    "    '--top_k', '40',\n",
    "    '--top_p', '0.9',\n",
    "    '--repeat_penalty', '1.1',\n",
    "    '--ctx_size', '256',\n",
    "    '--n_predict', '512',\n",
    "    '--temp', '0.95'\n",
    "]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "f9cb88bb",
   "metadata": {},
   "source": [
    "## gusev_alpaca_7b_lora"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "50098bb1",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'IlyaGusev/alpaca_7b_lora_reproduce'\n",
    "\n",
    "config = PeftConfig.from_pretrained(model_name)\n",
    "config.base_model_name_or_path = 'decapoda-research/llama-7b-hf'\n",
    "\n",
    "tokenizer = LlamaTokenizer.from_pretrained(model_name)\n",
    "tokenizer.padding_side = 'left'\n",
    "\n",
    "model = LlamaForCausalLM.from_pretrained(\n",
    "    config.base_model_name_or_path,\n",
    "    load_in_8bit=True,\n",
    "    torch_dtype=torch.float16,\n",
    "    device_map='auto'\n",
    ")\n",
    "model = PeftModel.from_pretrained(\n",
    "    model, model_name,\n",
    "    torch_dtype=torch.float16,\n",
    ").eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "8716b5c1",
   "metadata": {},
   "source": [
    "## gusev_fred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "ed5b069a",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "model_name = 'IlyaGusev/fred_t5_ru_turbo_alpaca'\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
    "\n",
    "generation_config = GenerationConfig.from_pretrained(model_name)\n",
    "generation_config.max_new_tokens = 512\n",
    "\n",
    "device = 'cuda:0'\n",
    "model = AutoModelForSeq2SeqLM.from_pretrained(model_name)\n",
    "model = model.to(device).eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ab9b8dde",
   "metadata": {},
   "source": [
    "## chainyo_alpaca_7b_lora"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "49bd4f0e",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'chainyo/alpaca-lora-7b'\n",
    "tokenizer = LlamaTokenizer.from_pretrained(model_name)\n",
    "generation_config = GenerationConfig(\n",
    "    temperature=0.2,\n",
    "    top_p=0.75,\n",
    "    top_k=40,\n",
    "    num_beams=4,\n",
    "    max_new_tokens=128,\n",
    ")\n",
    "model = LlamaForCausalLM.from_pretrained(\n",
    "    model_name,\n",
    "    load_in_8bit=True,\n",
    "    torch_dtype=torch.float16,\n",
    "    device_map='auto',\n",
    ").eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "eb55b23b",
   "metadata": {},
   "source": [
    "## wortega_instruct_rugpt_large"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "039def6e",
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = 'AlexWortega/instruct_rugptlarge'\n",
    "\n",
    "tokenizer = GPT2TokenizerFast.from_pretrained(model_name)\n",
    "special_tokens_dict = {\n",
    "    'additional_special_tokens': [\n",
    "        '<code>', '</code>',\n",
    "        '<instructionS>', '<instructionE>',\n",
    "        '<next>'\n",
    "    ]\n",
    "}\n",
    "tokenizer.add_special_tokens(special_tokens_dict)\n",
    "\n",
    "generation_config = GenerationConfig(\n",
    "    min_length=20,\n",
    "    max_new_tokens=512,\n",
    "    top_k=50,\n",
    "    top_p=0.7,\n",
    "    do_sample=True,  \n",
    "    early_stopping=True,\n",
    "    no_repeat_ngram_size=2,\n",
    "    eos_token_id=tokenizer.eos_token_id,\n",
    "    pad_token_id=tokenizer.eos_token_id,\n",
    "    use_cache=True,\n",
    "    repetition_penalty=1.5,  \n",
    "    length_penalty=1.2,  \n",
    "    num_beams=4,\n",
    ")\n",
    "\n",
    "device = 'cuda:0'\n",
    "model = GPT2LMHeadModel.from_pretrained(model_name).to(device).eval()\n",
    "model.resize_token_embeddings(len(tokenizer))"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "32955b83",
   "metadata": {},
   "source": [
    "## oasst_pythia"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "003919c2",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "model_name = 'OpenAssistant/oasst-sft-4-pythia-12b-epoch-3.5'\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
    "\n",
    "generation_config = GenerationConfig.from_pretrained(model_name)\n",
    "generation_config.max_new_tokens = 512\n",
    "generation_config.repetition_penalty = 1.1\n",
    "generation_config.no_repeat_ngram_size = 8\n",
    "\n",
    "device = 'cuda:0'\n",
    "model = AutoModelForCausalLM.from_pretrained(model_name)\n",
    "model = model.to(device).eval()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "06652557",
   "metadata": {},
   "source": [
    "# evals"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "4402cc05",
   "metadata": {},
   "outputs": [],
   "source": [
    "task_path = Path('tasks/vicuna_question_ru.jsonl')\n",
    "eval_path = Path('evals/vicuna_question_ru/oasst_pythia.jsonl')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "f44a735e",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "task_records = load_task(task_path)\n",
    "eval_records = [\n",
    "    EvalRecord(\n",
    "        id=_.id,\n",
    "        prompt=oasst_prompt(_),\n",
    "        output=None\n",
    "    )\n",
    "    for _ in task_records\n",
    "]\n",
    "random.sample(eval_records, 10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "a89e7ed3",
   "metadata": {},
   "outputs": [],
   "source": [
    "dump_eval(eval_path, eval_records)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "6cbeeaac",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "eval_records = list(load_eval(eval_path))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "2c56a1f4",
   "metadata": {},
   "outputs": [],
   "source": [
    "total = len(eval_records)\n",
    "have_output = sum(\n",
    "    _.output is not None\n",
    "    for _ in eval_records\n",
    ")\n",
    "print(have_output, total)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "aba9f9f2",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "model_batch_complete(eval_records, model, tokenizer, generation_config, batch_size=8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "083fad12",
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "for record in eval_records:\n",
    "    if record.output is None:\n",
    "        continue\n",
    "\n",
    "    prompt = re.sub(r'<\\|.+?\\|>', '', record.prompt)\n",
    "    output = record.output.removeprefix(prompt)\n",
    "#     # output = output[:output.find('<end>')]\n",
    "#     print(output)\n",
    "#     print('\\n\\n\\n')\n",
    "\n",
    "    record.output = output"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "974f6eb8",
   "metadata": {},
   "outputs": [],
   "source": [
    "# %run -n main.py\n",
    "# await batch_llamacpp_complete(\n",
    "#     eval_records, main_path, model_path, generation_args,\n",
    "#     threads=8, pool_size=3\n",
    "# )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "20412d0a",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "dump_eval(eval_path, eval_records)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2727d70c",
   "metadata": {},
   "source": [
    "# annot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "7c06fe56",
   "metadata": {},
   "outputs": [],
   "source": [
    "%run -n main.py\n",
    "task = 'vicuna_question_ru'\n",
    "a_name = 'openai_davinci_003'\n",
    "b_name = 'gusev_saiga_13b_lora'\n",
    "\n",
    "lines = read_lines(f'evals/{task}/{a_name}.jsonl')\n",
    "a_items = list(parse_jsonl(lines))\n",
    "\n",
    "lines = read_lines(f'evals/{task}/{b_name}.jsonl')\n",
    "b_items = list(parse_jsonl(lines))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "7906f586",
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "id_a_items = {_['id']: _ for _ in a_items}\n",
    "id_b_items = {_['id']: _ for _ in b_items}\n",
    "items = []\n",
    "for id in sorted(id_a_items.keys() & id_b_items.keys()):\n",
    "    a_item = id_a_items[id]\n",
    "    b_item = id_b_items[id]\n",
    "\n",
    "    items.append({\n",
    "        'id': id,\n",
    "        'prompt': b_item['prompt'],\n",
    "        'a': a_item['output'],\n",
    "        'b': b_item['output'],\n",
    "        'label': None\n",
    "    })\n",
    "\n",
    "table = pd.DataFrame(items)\n",
    "table.to_excel(f'sbs/{task}/{a_name}_{b_name}.xlsx', index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2f6761ed",
   "metadata": {},
   "source": [
    "# report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "d1ddf8cf",
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "%config InlineBackend.figure_format='retina'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "fa492332",
   "metadata": {},
   "outputs": [],
   "source": [
    "+ gusev_fred\n",
    "+ oasst_pythia\n",
    "\n",
    "+ wortega_instruct_rugpt_large\n",
    "wortega_instruct_rugpt_large_v2\n",
    "\n",
    "+ gusev_ru_alpaca_7b_lora\n",
    "gusev_ru_alpaca_7b_lora_v2\n",
    "\n",
    ". gusev_ru_alpaca_13b_lora\n",
    "gusev_ru_alpaca_13b_lora_v2\n",
    "\n",
    "+ gusev_saiga_7b_lora\n",
    "gusev_saiga_7b_lora_v2\n",
    "\n",
    ". gusev_saiga_13b_lora\n",
    "gusev_saiga_30b_lora\n",
    "\n",
    "+ davinci-002\n",
    ". turbo"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "id": "94a58e8e",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x17a32f160>"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from pathlib import Path\n",
    "\n",
    "names = [\n",
    "#     'openai_turbo',\n",
    "    'gusev_saiga_7b_lora',\n",
    "    'gusev_ru_alpaca_7b_lora',\n",
    "    'openai_davinci_002',\n",
    "    'wortega_instruct_rugpt_large',\n",
    "    'gusev_fred',\n",
    "    'oasst_pythia',\n",
    "]\n",
    "\n",
    "\n",
    "data = {}\n",
    "for name in names:\n",
    "    path = f'sbs/vicuna_question_ru/openai_davinci_003_{name}.xlsx'\n",
    "    table = pd.read_excel(path, dtype='str')\n",
    "    \n",
    "    label_counts = table.label.value_counts()\n",
    "    for label, count in label_counts.items():\n",
    "        data[name, label] = count\n",
    "        \n",
    "\n",
    "table = pd.Series(data)\n",
    "table = table.unstack()\n",
    "table = table.fillna(0)\n",
    "\n",
    "table = table.reindex(\n",
    "    index=reversed(names),\n",
    "    columns=[\n",
    "        '2', '1', '0', '-1', '-2', '?',\n",
    "    ]\n",
    "    \n",
    ")\n",
    "table = table.rename(\n",
    "    columns = {\n",
    "        '?': 'пустой ответ',\n",
    "        '-2': 'хуже',\n",
    "        '-1': 'похуже',\n",
    "        '0': 'примерно одинаково',\n",
    "        '1': 'получше',\n",
    "        '2': 'лучше'\n",
    "    }\n",
    ")\n",
    "\n",
    "\n",
    "table.plot(\n",
    "    title='SbS с openai_davinci_003 на ru_vicuna_question',\n",
    "    kind='barh',\n",
    "    stacked=True,\n",
    "    width=0.9,\n",
    "    alpha=0.7,\n",
    "    xlabel='# заданий'\n",
    ").legend(\n",
    "    loc='upper left',\n",
    "    bbox_to_anchor=(1.0, 1.0)\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "72dd24e2",
   "metadata": {},
   "outputs": [],
   "source": [
    "!open sbs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "be7ca193",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "rulm-sbs",
   "language": "python",
   "name": "rulm-sbs"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.10.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}