Upload training.ipynb with huggingface_hub

training.ipynb

@@ -0,0 +1,605 @@
+{
+ "cells": [
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7ef3e090-1986-4080-827e-fdef2deda5ba",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline\n",
+    "import torch\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ee142e5a-92ac-400b-a048-89a3df0060f6",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "print(f\"Device set to: {device}\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ba2eea5c-108e-4305-a64e-c35800cf9bf2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Load CLI Q&A dataset\n",
+    "with open(\"cli_questions.json\", \"r\", encoding=\"utf-8\") as f:\n",
+    "    data = json.load(f)\n",
+    "\n",
+    "# Access the list of entries inside \"data\" key\n",
+    "qa_list = data[\"data\"]\n",
+    "\n",
+    "# Show a sample\n",
+    "print(f\"Total entries: {len(qa_list)}\")\n",
+    "print(\"Sample entry:\", qa_list[0])\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "81490ae9-b6f9-4004-b098-c09677c1dcd3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model_id = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
+    "model = AutoModelForCausalLM.from_pretrained(model_id)\n",
+    "model.to(device)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "5eb00a02-a5a5-4746-bc1f-685ce4865600",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "generator = pipeline(\"text-generation\", model=model, tokenizer=tokenizer, device=-1)  # -1 for CPU\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "0f2b0688-a24d-4d86-90e5-9b8237620f6c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Pick sample questions\n",
+    "sample_questions = [entry[\"question\"] for entry in qa_list[:5]]\n",
+    "\n",
+    "# Generate and print answers\n",
+    "for i, question in enumerate(sample_questions):\n",
+    "    print(f\"Q{i+1}: {question}\")\n",
+    "    output = generator(question, max_new_tokens=150, do_sample=True, temperature=0.7)\n",
+    "    print(f\"A{i+1}: {output[0]['generated_text']}\\n{'-'*60}\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "0f52ebb0-e2b9-4971-b66c-5353257b7a1c",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "prompt = f\"Q: {question}\\nA:\"\n",
+    "output = generator(prompt, max_new_tokens=100, do_sample=True, temperature=0.7)\n",
+    "print(output[0][\"generated_text\"])\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "49fcf984-bd0d-48b7-857a-e6a6e04585b8",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "\n",
+    "# Load the dataset\n",
+    "with open(\"cli_questions.json\", \"r\") as f:\n",
+    "    raw = json.load(f)\n",
+    "    data = raw[\"data\"]  # ensure this matches your JSON structure\n",
+    "\n",
+    "# Generate answers\n",
+    "results = []\n",
+    "for i, item in enumerate(data[:50]):  # run on subset first\n",
+    "    question = item[\"question\"]\n",
+    "    prompt = f\"Q: {question}\\nA:\"\n",
+    "    output = generator(prompt, max_new_tokens=150, temperature=0.7, do_sample=True)\n",
+    "    answer = output[0][\"generated_text\"].split(\"A:\")[1].strip() if \"A:\" in output[0][\"generated_text\"] else output[0][\"generated_text\"]\n",
+    "    results.append({\"question\": question, \"answer\": answer})\n",
+    "    print(f\"Q{i+1}: {question}\\nA{i+1}: {answer}\\n{'-'*60}\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "819b988d-c6a1-4b11-b09d-1f1892e18158",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "!pip install transformers datasets peft accelerate bitsandbytes trl --quiet\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6b3c1312-3499-4462-b435-9fe72f0d6f06",
+   "metadata": {
+    "scrolled": true
+   },
+   "outputs": [],
+   "source": [
+    "print(\"Top-level keys:\", data.keys() if isinstance(data, dict) else \"Not a dict\")\n",
+    "print(\"Preview:\", str(data)[:500])  # Print first 500 chars of the content\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "96748b74-a5c7-439e-8428-680cba84e06d",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "from datasets import Dataset\n",
+    "\n",
+    "# Load and extract Q&A list\n",
+    "with open(\"cli_questions.json\", \"r\") as f:\n",
+    "    raw = json.load(f)\n",
+    "    data_list = raw[\"data\"]  # ✅ correct key now\n",
+    "\n",
+    "# Convert to prompt/response format\n",
+    "for sample in data_list:\n",
+    "    sample[\"prompt\"] = sample[\"question\"]\n",
+    "    sample[\"response\"] = sample[\"answer\"]\n",
+    "\n",
+    "# Create HuggingFace Dataset\n",
+    "dataset = Dataset.from_list(data_list)\n",
+    "dataset = dataset.train_test_split(test_size=0.1)\n",
+    "\n",
+    "print(\"Loaded dataset:\", dataset)\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7a7560e5-b04f-480c-b989-0bb3d3611701",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoTokenizer, AutoModelForCausalLM\n",
+    "\n",
+    "model_name = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"  # or try \"microsoft/phi-2\"\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    device_map=\"auto\",\n",
+    "    load_in_4bit=True  # For LoRA on low-resource\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ae23057e-b741-4541-946d-77f9c5b8c9dc",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoTokenizer, AutoModelForCausalLM\n",
+    "\n",
+    "model_name = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    torch_dtype=\"auto\",  # or torch.float32 if you get another dtype error\n",
+    "    device_map=\"cpu\"     # force CPU since no supported GPU found\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ac99fe95-b5f3-4591-bc7c-793e195eeb86",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig\n",
+    "\n",
+    "bnb_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.float16,\n",
+    ")\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    device_map=\"auto\",\n",
+    "    quantization_config=bnb_config\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7bde0e33-3bed-4940-907f-e0c2e7af1cd3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import torch\n",
+    "from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig\n",
+    "\n",
+    "model_name = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"\n",
+    "\n",
+    "bnb_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.float16,\n",
+    ")\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    device_map=\"auto\",\n",
+    "    quantization_config=bnb_config\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "51e0d14a-18c7-410f-9821-0eb00d3d1bbc",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoTokenizer, AutoModelForCausalLM\n",
+    "\n",
+    "model_name = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    device_map=\"auto\",   # This will still use CPU if no GPU is found\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "f4e4786e-e67c-4c0f-b169-6996a2966558",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "model = AutoModelForCausalLM.from_pretrained(\n",
+    "    model_name,\n",
+    "    device_map=\"auto\",\n",
+    "    torch_dtype=torch.float32  # or float16 if your CPU supports it\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "dfd328ef-9362-426b-894e-923e70c7ace3",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "device = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n",
+    "print(f\"Device set to: {device}\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6743ec8e-8bd9-4a73-8786-fd71a6790d78",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "import torch\n",
+    "from datasets import Dataset\n",
+    "from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer, DataCollatorForLanguageModeling\n",
+    "from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4252cc0c-62fe-4871-8095-ab07959b7884",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "import torch\n",
+    "from datasets import Dataset\n",
+    "from transformers import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer, DataCollatorForLanguageModeling\n",
+    "from peft import LoraConfig, get_peft_model, prepare_model_for_kbit_training"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7153b443-8059-42d1-96fa-699d0f19f9cf",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "\n",
+    "with open(\"cli_questions.json\") as f:\n",
+    "    data = json.load(f)\n",
+    "\n",
+    "# Check the top-level structure\n",
+    "print(type(data))  # Should print <class 'dict'>\n",
+    "print(data.keys())  # See what keys are at the top\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fbfa8025-233e-47c5-9044-146f95bb24eb",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import json\n",
+    "from datasets import Dataset\n",
+    "\n",
+    "# Load the JSON and extract the list\n",
+    "with open(\"cli_questions.json\") as f:\n",
+    "    raw = json.load(f)\n",
+    "\n",
+    "qa_list = raw[\"data\"]  # access the list inside the 'data' key\n",
+    "\n",
+    "# Format for instruction tuning\n",
+    "formatted_data = [\n",
+    "    {\"text\": f\"### Question:\\n{item['question']}\\n\\n### Answer:\\n{item['answer']}\"}\n",
+    "    for item in qa_list\n",
+    "]\n",
+    "\n",
+    "# Convert to Hugging Face dataset\n",
+    "dataset = Dataset.from_list(formatted_data)\n",
+    "\n",
+    "# Preview\n",
+    "print(f\"Loaded {len(dataset)} formatted examples\")\n",
+    "print(dataset[0])\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "893c412e-0f09-44fd-b6f8-fe3557a071aa",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoTokenizer\n",
+    "\n",
+    "model_id = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"  # You can switch to Phi-2 if you prefer\n",
+    "\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_id)\n",
+    "tokenizer.pad_token = tokenizer.eos_token  # Needed for causal LM padding\n",
+    "\n",
+    "# Tokenization function\n",
+    "def tokenize(example):\n",
+    "    return tokenizer(example[\"text\"], padding=\"max_length\", truncation=True, max_length=512)\n",
+    "\n",
+    "tokenized_dataset = dataset.map(tokenize, batched=True)\n",
+    "tokenized_dataset = tokenized_dataset.remove_columns([\"text\"])\n",
+    "\n",
+    "tokenized_dataset.set_format(type=\"torch\")\n",
+    "print(tokenized_dataset[0])\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fb49d005-c57c-422f-8bc5-b4037a6bb40f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_dataset = tokenized_dataset\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "a3fb419b-703f-43c8-9be0-a71815b3da82",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Use entire dataset as training set\n",
+    "train_dataset = tokenized_dataset\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "09c26c73-e7e8-4610-97d6-6c4a10004785",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "tokenized_dataset.save_to_disk(\"tokenized_dataset\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "e66f130b-b80b-42fd-9f79-60f245f2c114",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from datasets import load_from_disk\n",
+    "\n",
+    "# Load the saved dataset\n",
+    "tokenized_dataset = load_from_disk(\"tokenized_dataset\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2dbe3f16-4d82-40c8-be84-b1f85910620f",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "train_dataset = tokenized_dataset  # Use full set for training since it's only 172 examples\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "7f05e8d5-fcdf-4a11-9c51-7e8ecd255848",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import DataCollatorForLanguageModeling\n",
+    "\n",
+    "data_collator = DataCollatorForLanguageModeling(\n",
+    "    tokenizer=tokenizer,\n",
+    "    mlm=False\n",
+    ")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "ec68cba4-8413-4c7d-91de-1fe798dc39fc",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from transformers import AutoModelForCausalLM, AutoTokenizer, TrainingArguments, Trainer, DataCollatorForLanguageModeling\n",
+    "from peft import get_peft_model, LoraConfig, prepare_model_for_kbit_training\n",
+    "from datasets import load_from_disk\n",
+    "import torch\n",
+    "\n",
+    "# Load model and tokenizer (TinyLlama)\n",
+    "model_name = \"TinyLlama/TinyLlama-1.1B-Chat-v1.0\"\n",
+    "tokenizer = AutoTokenizer.from_pretrained(model_name)\n",
+    "tokenizer.pad_token = tokenizer.eos_token  # Important for Trainer padding\n",
+    "\n",
+    "model = AutoModelForCausalLM.from_pretrained(model_name)\n",
+    "\n",
+    "# Setup LoRA config\n",
+    "lora_config = LoraConfig(\n",
+    "    r=8,\n",
+    "    lora_alpha=16,\n",
+    "    target_modules=[\"q_proj\", \"v_proj\"],\n",
+    "    lora_dropout=0.1,\n",
+    "    bias=\"none\",\n",
+    "    task_type=\"CAUSAL_LM\"\n",
+    ")\n",
+    "\n",
+    "# Inject LoRA adapters\n",
+    "model = get_peft_model(model, lora_config)\n",
+    "\n",
+    "# Load the tokenized dataset\n",
+    "dataset = load_from_disk(\"tokenized_dataset\")\n",
+    "\n",
+    "# Setup data collator\n",
+    "data_collator = DataCollatorForLanguageModeling(\n",
+    "    tokenizer=tokenizer,\n",
+    "    mlm=False\n",
+    ")\n",
+    "\n",
+    "# Training args\n",
+    "training_args = TrainingArguments(\n",
+    "    output_dir=\"./lora-tinyllama-output\",\n",
+    "    per_device_train_batch_size=2,         # Small batch size for CPU\n",
+    "    gradient_accumulation_steps=4,\n",
+    "    num_train_epochs=1,                    # Reduce for quicker runs\n",
+    "    logging_steps=10,\n",
+    "    save_strategy=\"epoch\",\n",
+    "    learning_rate=2e-4,\n",
+    "    fp16=False,                            # Don't use fp16 on CPU\n",
+    "    report_to=\"none\"\n",
+    ")\n",
+    "\n",
+    "# Define Trainer\n",
+    "trainer = Trainer(\n",
+    "    model=model,\n",
+    "    args=training_args,\n",
+    "    train_dataset=dataset,\n",
+    "    tokenizer=tokenizer,\n",
+    "    data_collator=data_collator\n",
+    ")\n",
+    "\n",
+    "# Start training\n",
+    "trainer.train()\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2eaf9fa5-540c-4bd2-b6e1-9ea60c820004",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "pip install -r requirements.txt\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "fad00764-e047-4fd0-b703-c9bbd343ce46",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "login(token=\"REMOVED_TOKEN_...\")\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "075e175f-d164-420a-92fb-75150637d351",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from huggingface_hub import login\n",
+    "import os\n",
+    "\n",
+    "# Safer login using environment variable (no token exposed in notebook)\n",
+    "login(token=os.getenv(\"HF_TOKEN\"))\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "def2deab-147c-4445-8e62-96c397d72f12",
+   "metadata": {},
+   "outputs": [],
+   "source": []
+  }
+ ],
+ "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": "3.12.7"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}