2023_07_15_falcon_finetune_qlora_langchain.ipynb

{
 "cells": [
  {
   "attachments": {},
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "MIT License\n",
    "\n",
    "Copyright (c) 2023 Pavel Shibanov [https://blog.experienced.dev/](https://blog.experienced.dev/?utm_source=notebooks)\n",
    "\n",
    "Permission is hereby granted, free of charge, to any person obtaining a copy\n",
    "of this software and associated documentation files (the \"Software\"), to deal\n",
    "in the Software without restriction, including without limitation the rights\n",
    "to use, copy, modify, merge, publish, distribute, sublicense, and/or sell\n",
    "copies of the Software, and to permit persons to whom the Software is\n",
    "furnished to do so, subject to the following conditions:\n",
    "\n",
    "The above copyright notice and this permission notice shall be included in all\n",
    "copies or substantial portions of the Software.\n",
    "\n",
    "THE SOFTWARE IS PROVIDED \"AS IS\", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR\n",
    "IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,\n",
    "FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE\n",
    "AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER\n",
    "LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,\n",
    "OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE\n",
    "SOFTWARE."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/jpeg": "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",
      "text/html": [
       "\n",
       "        <iframe\n",
       "            width=\"560\"\n",
       "            height=\"315\"\n",
       "            src=\"https://www.youtube.com/embed/X4-2zw5QLns\"\n",
       "            frameborder=\"0\"\n",
       "            allowfullscreen\n",
       "            \n",
       "        ></iframe>\n",
       "        "
      ],
      "text/plain": [
       "<IPython.lib.display.YouTubeVideo at 0x7f8c3360f8b0>"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import YouTubeVideo\n",
    "\n",
    "YouTubeVideo(\"X4-2zw5QLns\", width=\"560\", height=\"315\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "!nvidia-smi"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%env PIP_QUIET=1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%pip install \\\n",
    "    python-dotenv==1.0.0 \\\n",
    "    bitsandbytes==0.40.0 \\\n",
    "    datasets==2.13.1 \\\n",
    "    scipy==1.11.1 \\\n",
    "    einops==0.6.1 \\\n",
    "    xformers==0.0.20 \\\n",
    "    langchain==0.0.234 \\\n",
    "    git+https://github.com/huggingface/transformers.git@91d7df58b6537d385e90578dac40204cb550f706 \\\n",
    "    git+https://github.com/huggingface/accelerate.git@bb47344c774b9508d8171f746632794524160410 \\\n",
    "    git+https://github.com/huggingface/peft.git@06755411549a12da27ba07e4f0ff9699bd6d9194 \\\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_name = \"tiiuae/falcon-7b-instruct\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoTokenizer\n",
    "\n",
    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
    "tokenizer.pad_token = tokenizer.eos_token"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from transformers import AutoModelForCausalLM\n",
    "\n",
    "base_model = AutoModelForCausalLM.from_pretrained(\n",
    "    model_name,\n",
    "    torch_dtype=torch.float16,\n",
    "    load_in_8bit=False,\n",
    "    device_map=\"auto\",\n",
    "    trust_remote_code=True,\n",
    ")\n",
    "model = base_model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from datasets import load_dataset\n",
    "\n",
    "data = load_dataset(\"truthful_qa\", \"generation\")\n",
    "data = data[\"validation\"].filter(lambda item: item[\"category\"] == \"Misconceptions\")\n",
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import transformers\n",
    "from langchain.llms import HuggingFacePipeline\n",
    "from langchain.chains import ConversationChain\n",
    "from langchain import PromptTemplate, LLMChain\n",
    "from IPython.display import display, Markdown\n",
    "\n",
    "\n",
    "default_template = \"\"\"\n",
    "Human: {input} \n",
    "AI:\"\"\"\n",
    "\n",
    "verbose_template = \"\"\"\n",
    "The following is a friendly conversation between a human and an AI.\n",
    "The AI is talkative and provides lots of specific details from its context. \n",
    "If the AI does not know the answer to a question, it truthfully says it does not know.\n",
    "\n",
    "Current conversation:\n",
    "Human: {input} \n",
    "AI:\"\"\"\n",
    "\n",
    "\n",
    "def get_chain(model, template, verbose=False):\n",
    "    pipeline = transformers.pipeline(\n",
    "        model=model,\n",
    "        tokenizer=tokenizer,\n",
    "        return_full_text=True,\n",
    "        task=\"text-generation\",\n",
    "        stop_sequence=\"\\nAI:\",\n",
    "        temperature=0.7,\n",
    "        max_new_tokens=512,\n",
    "        repetition_penalty=1.2,\n",
    "    )\n",
    "    return LLMChain(\n",
    "        llm=HuggingFacePipeline(pipeline=pipeline),\n",
    "        prompt=PromptTemplate.from_template(template),\n",
    "        verbose=verbose,\n",
    "    )\n",
    "\n",
    "\n",
    "base_chain_verbose = get_chain(base_model, verbose_template)\n",
    "fine_tuned_verbose = get_chain(model, verbose_template)\n",
    "\n",
    "\n",
    "def compare_results(base_model, fine_tuned_model, item, template=None, verbose=False):\n",
    "    if template is None:\n",
    "        template = default_template\n",
    "    base_chain = get_chain(base_model, template, verbose)\n",
    "    fine_tuned_chain = get_chain(fine_tuned_model, template, verbose)\n",
    "    base_res = base_chain.run(item[\"question\"])\n",
    "    fine_tuned_res = fine_tuned_chain.run(item[\"question\"])\n",
    "    display(\n",
    "        Markdown(\n",
    "            f\"\"\"\n",
    "### question: \n",
    "{item['question']}\n",
    "#### base_model:\n",
    "{base_res}\n",
    "#### fine_tuned_model:\n",
    "{fine_tuned_res}\n",
    "#### best answer:\n",
    "{item['best_answer']}\n",
    "#### source:\n",
    "{item['source']}\n",
    "\"\"\"\n",
    "        )\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "compare_results(base_model, model, data[46])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "compare_results(base_model, model, random.choice(data))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "compare_results(base_model, model, random.choice(data), verbose_template, verbose=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "fine_tuning_template = \"\"\"\n",
    "Human: {question}\n",
    "AI: {best_answer}\n",
    "\"\"\"\n",
    "fine_tune_prompt = PromptTemplate.from_template(fine_tuning_template)\n",
    "\n",
    "\n",
    "def tokenize(item):\n",
    "    return tokenizer(\n",
    "        fine_tune_prompt.format(\n",
    "            question=item[\"question\"], best_answer=item[\"best_answer\"]\n",
    "        ),\n",
    "        padding=True,\n",
    "        truncation=True,\n",
    "    )\n",
    "\n",
    "\n",
    "train_dataset = data.map(tokenize)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from transformers import BitsAndBytesConfig\n",
    "\n",
    "quantization_config = BitsAndBytesConfig(\n",
    "    load_in_4bit=True,\n",
    "    bnb_4bit_use_double_quant=True,\n",
    "    bnb_4bit_quant_type=\"nf4\",\n",
    "    bnb_4bit_compute_dtype=torch.bfloat16,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from transformers import AutoModelForCausalLM\n",
    "\n",
    "model = AutoModelForCausalLM.from_pretrained(\n",
    "    model_name,\n",
    "    device_map=\"auto\",\n",
    "    quantization_config=quantization_config,\n",
    "    trust_remote_code=True,\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def print_num_params(model):\n",
    "    params = [\n",
    "        (param.numel(), param.numel() if param.requires_grad else 0)\n",
    "        for _, param in model.named_parameters()\n",
    "    ]\n",
    "    all, train = map(sum, zip(*params))\n",
    "    print(f\"{train=} / {all=} {train/all:f}\")\n",
    "\n",
    "\n",
    "print_num_params(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# model.gradient_checkpointing_enable()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "from peft import prepare_model_for_kbit_training, LoraConfig, get_peft_model\n",
    "\n",
    "model = prepare_model_for_kbit_training(model)\n",
    "config = LoraConfig(\n",
    "    r=16,\n",
    "    lora_alpha=32,\n",
    "    target_modules=[\"query_key_value\"],\n",
    "    lora_dropout=0.05,\n",
    "    bias=\"none\",\n",
    "    task_type=\"CAUSAL_LM\",\n",
    ")\n",
    "model = get_peft_model(model, config)\n",
    "print_num_params(model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import transformers\n",
    "\n",
    "output_dir = \"fine_tuned\"\n",
    "\n",
    "training_args = transformers.TrainingArguments(\n",
    "    per_device_train_batch_size=32,\n",
    "    gradient_accumulation_steps=4,\n",
    "    num_train_epochs=1,\n",
    "    learning_rate=2e-4,\n",
    "    fp16=True,\n",
    "    save_total_limit=3,\n",
    "    logging_steps=1,\n",
    "    output_dir=output_dir,\n",
    "    max_steps=100,\n",
    "    optim=\"paged_adamw_8bit\",\n",
    "    lr_scheduler_type=\"cosine\",\n",
    "    warmup_ratio=0.05,\n",
    ")\n",
    "trainer = transformers.Trainer(\n",
    "    model=model,\n",
    "    train_dataset=train_dataset,\n",
    "    args=training_args,\n",
    "    data_collator=transformers.DataCollatorForLanguageModeling(tokenizer, mlm=False),\n",
    ")\n",
    "model.config.use_cache = False\n",
    "trainer.train()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "compare_results(base_model, model, data[6])"
   ]
  }
 ],
 "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": "3.10.10"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}