{"cells":[{"cell_type":"code","source":["# ============================= CELL 1: Prepare Latents + Distill Correct 768-dim Text Encoder =============================\n","# @title 1. Full Preparation: Latents + 768-dim Distilled Text Encoder\n","\n","import os\n","import zipfile\n","import torch\n","import torch.nn as nn\n","import torch.nn.functional as F\n","import numpy as np\n","from google.colab import drive\n","from PIL import Image\n","from tqdm import tqdm\n","from diffusers import AutoencoderKL\n","from transformers import AutoTokenizer, AutoModel\n","from datasets import Dataset as HFDataset\n","from torch.utils.data import Dataset\n","from peft import LoraConfig, get_peft_model\n","from transformers import Trainer, TrainingArguments, set_seed\n","\n","set_seed(42)\n","drive.mount('/content/drive', force_remount=True)\n","\n","zip_path = '/content/drive/MyDrive/my_set.zip'  # @param {type:'string'}\n","\n","# ====================== 1. Extract Data ======================\n","print(\"📦 Extracting zip...\")\n","extract_dir = \"/content/data\"\n","os.makedirs(extract_dir, exist_ok=True)\n","\n","with zipfile.ZipFile(zip_path, 'r') as zip_ref:\n","    zip_ref.extractall(extract_dir)\n","\n","image_files = [f for f in os.listdir(extract_dir) if f.lower().endswith(('.png', '.jpg', '.jpeg', '.webp'))]\n","print(f\"✅ Found {len(image_files)} images\")\n","\n","text_files = sorted([f for f in os.listdir(extract_dir) if f.endswith('.txt') and f[0].isdigit()])\n","texts = []\n","for tf in text_files:\n","    with open(os.path.join(extract_dir, tf), \"r\", encoding=\"utf-8\") as f:\n","        content = f.read().strip()\n","        if content:\n","            texts.append(content)\n","\n","print(f\"✅ Loaded {len(texts)} captions\")\n","\n","# ====================== 2. Encode Images → Flux VAE Latents ======================\n","latent_dir = \"/content/drive/MyDrive/flux_klein_latents\"\n","if os.path.exists(latent_dir) and len(os.listdir(latent_dir)) == len(image_files):\n","    print(f\"✅ Using existing latents from {latent_dir}\")\n","else:\n","    print(\"\\n🌀 Encoding images to Flux VAE latents...\")\n","    vae = AutoencoderKL.from_pretrained(\n","        \"black-forest-labs/FLUX.1-dev\",\n","        subfolder=\"vae\",\n","        torch_dtype=torch.float32,\n","        device_map=\"auto\"\n","    )\n","    vae.eval()\n","\n","    os.makedirs(latent_dir, exist_ok=True)\n","\n","    with torch.no_grad():\n","        for img_file in tqdm(image_files, desc=\"Encoding latents\"):\n","            img_path = os.path.join(extract_dir, img_file)\n","            image = Image.open(img_path).convert(\"RGB\").resize((1024, 1024), Image.LANCZOS)\n","\n","            pixel_values = (torch.from_numpy(np.array(image)).permute(2, 0, 1).unsqueeze(0).float() / 255.0)\n","            pixel_values = pixel_values.to(vae.device, dtype=vae.dtype) * 2.0 - 1.0\n","\n","            latents = vae.encode(pixel_values).latent_dist.sample() * vae.config.scaling_factor\n","            latent_name = os.path.splitext(img_file)[0] + \".pt\"\n","            torch.save(latents.cpu(), os.path.join(latent_dir, latent_name))\n","\n","    del vae\n","    torch.cuda.empty_cache()\n","    print(f\"✅ Latents saved to {latent_dir}\")\n","\n","# ====================== 3. Compute & Project Teacher Embeddings to 768 ======================\n","print(\"\\n📝 Computing teacher embeddings and projecting to 768-dim...\")\n","\n","teacher_model_name = \"Qwen/Qwen3-Embedding-0.6B\"\n","tokenizer = AutoTokenizer.from_pretrained(teacher_model_name)\n","teacher_model = AutoModel.from_pretrained(\n","    teacher_model_name,\n","    torch_dtype=torch.float16,\n","    device_map=\"auto\",\n","    trust_remote_code=True\n",")\n","teacher_model.eval()\n","\n","teacher_embeddings_1024 = []\n","with torch.no_grad():\n","    for text in tqdm(texts, desc=\"Teacher encoding\"):\n","        inputs = tokenizer(text, padding=True, truncation=True, max_length=512, return_tensors=\"pt\").to(teacher_model.device)\n","        outputs = teacher_model(**inputs)\n","        emb = outputs.last_hidden_state.mean(dim=1).squeeze(0).cpu()\n","        teacher_embeddings_1024.append(emb)\n","\n","teacher_embeddings_1024 = torch.stack(teacher_embeddings_1024)\n","print(f\"✅ Teacher embeddings (1024): {teacher_embeddings_1024.shape}\")\n","\n","# Project teacher to 768-dim (correct target for FLUX.2-klein)\n","teacher_proj = nn.Linear(1024, 768).to(\"cuda\")\n","with torch.no_grad():\n","    teacher_embeddings_768 = teacher_proj(teacher_embeddings_1024.to(\"cuda\")).cpu()\n","\n","print(f\"✅ Projected teacher embeddings (768): {teacher_embeddings_768.shape}\")\n","\n","# Save projected teacher embeddings\n","torch.save({\n","    \"embeddings\": teacher_embeddings_768,\n","    \"texts\": texts,\n","    \"dim\": 768\n","}, \"/content/drive/MyDrive/qwen_embeddings_768.pt\")\n","\n","del teacher_model, teacher_proj\n","torch.cuda.empty_cache()\n","\n","# ====================== 4. Distill Student to 768-dim ======================\n","print(\"\\n👨‍🎓 Distilling student model to 768-dim...\")\n","\n","student_model_name = \"Qwen/Qwen2.5-0.5B\"\n","base_student = AutoModel.from_pretrained(\n","    student_model_name, torch_dtype=torch.float32, device_map=\"auto\", trust_remote_code=True\n",")\n","\n","lora_config = LoraConfig(\n","    r=16,\n","    lora_alpha=32,\n","    target_modules=[\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\"],\n","    lora_dropout=0.05,\n","    bias=\"none\",\n","    task_type=\"FEATURE_EXTRACTION\"\n",")\n","student_model = get_peft_model(base_student, lora_config)\n","\n","projection = nn.Linear(base_student.config.hidden_size, 768).to(\"cuda\")  # Correct dimension\n","projection.train()\n","\n","hf_dataset = HFDataset.from_dict({\"text\": texts})\n","\n","class DistillationDataset(Dataset):\n","    def __init__(self, hf_dataset, tokenizer, teacher_embs, max_length=512):\n","        self.dataset = hf_dataset\n","        self.tokenizer = tokenizer\n","        self.teacher_embs = teacher_embs\n","        self.max_length = max_length\n","\n","    def __len__(self): return len(self.dataset)\n","\n","    def __getitem__(self, idx):\n","        text = self.dataset[idx][\"text\"]\n","        inputs = self.tokenizer(text, padding=\"max_length\", truncation=True, max_length=self.max_length, return_tensors=\"pt\")\n","        return {\n","            \"input_ids\": inputs[\"input_ids\"].squeeze(0),\n","            \"attention_mask\": inputs[\"attention_mask\"].squeeze(0),\n","            \"labels\": self.teacher_embs[idx],\n","        }\n","\n","distill_dataset = DistillationDataset(hf_dataset, tokenizer, teacher_embeddings_768)\n","\n","def collate_fn(batch):\n","    return {\n","        \"input_ids\": torch.stack([item[\"input_ids\"] for item in batch]),\n","        \"attention_mask\": torch.stack([item[\"attention_mask\"] for item in batch]),\n","        \"labels\": torch.stack([item[\"labels\"] for item in batch])\n","    }\n","\n","class DistillTrainer(Trainer):\n","    def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None):\n","        labels = inputs.pop(\"labels\").to(\"cuda\")   # 768-dim\n","\n","        outputs = model(input_ids=inputs[\"input_ids\"], attention_mask=inputs[\"attention_mask\"])\n","        hidden = outputs.last_hidden_state.mean(dim=1)\n","        student_emb = projection(hidden)   # 768-dim\n","\n","        student_norm = F.normalize(student_emb, p=2, dim=1)\n","        teacher_norm = F.normalize(labels, p=2, dim=1)\n","\n","        mse_loss = F.mse_loss(student_norm, teacher_norm)\n","        cos_loss = (1 - F.cosine_similarity(student_norm, teacher_norm, dim=1)).mean()\n","        loss = 0.25 * mse_loss + 0.75 * cos_loss\n","\n","        return (loss, outputs) if return_outputs else loss\n","\n","training_args = TrainingArguments(\n","    output_dir=\"./distilled_qwen_768\",\n","    per_device_train_batch_size=4,      # safe for Colab\n","    num_train_epochs=30,\n","    learning_rate=2e-4,\n","    fp16=True,\n","    logging_steps=50,\n","    save_strategy=\"no\",\n","    report_to=\"none\",\n","    remove_unused_columns=False,\n",")\n","\n","trainer = DistillTrainer(\n","    model=student_model,\n","    args=training_args,\n","    train_dataset=distill_dataset,\n","    data_collator=collate_fn,\n",")\n","\n","print(\"🚀 Starting distillation with correct 768-dim targets...\")\n","trainer.train()\n","\n","# ====================== Save Distilled Model ======================\n","distilled_save_dir = \"/content/drive/MyDrive/distilled_qwen_768_for_flux\"\n","os.makedirs(distilled_save_dir, exist_ok=True)\n","\n","student_model.save_pretrained(distilled_save_dir)\n","tokenizer.save_pretrained(distilled_save_dir)\n","torch.save(projection.state_dict(), f\"{distilled_save_dir}/projection.pth\")\n","\n","print(f\"\\n✅ SUCCESS! 768-dim distilled encoder saved to: {distilled_save_dir}\")\n","print(\"   Latents are in: /content/drive/MyDrive/flux_klein_latents\")\n","print(\"   You can now safely run Cell 2 to save everything and disconnect.\")\n","\n","torch.cuda.empty_cache()"],"metadata":{"id":"oluEkFV0KMXo"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["# ============================= CELL 2: Save All Assets to Drive =============================\n","# @title 2. Save Latents + New Distilled Encoder\n","\n","import os\n","import torch\n","\n","print(\"💾 Saving all assets to Google Drive...\")\n","\n","# Ensure directories exist\n","os.makedirs(\"/content/drive/MyDrive/flux_klein_latents\", exist_ok=True)\n","os.makedirs(\"/content/drive/MyDrive/distilled_qwen_768_for_flux\", exist_ok=True)\n","\n","# Move latents if not already there\n","# (assuming they are already saved in Cell 1)\n","\n","print(\"✅ Latents are in /content/drive/MyDrive/flux_klein_latents\")\n","print(\"✅ New 768-dim distilled model is in /content/drive/MyDrive/distilled_qwen_768_for_flux\")\n","\n","print(\"\\n🎉 All data is safely saved on Google Drive.\")\n","print(\"   You can now **disconnect and delete the runtime** if you want.\")\n","print(\"   Everything needed for training is on Drive.\")\n","print(\"   When you come back, start from Cell 3.\")"],"metadata":{"colab":{"base_uri":"https://localhost:8080/"},"id":"9IGpdiL9BBr6","executionInfo":{"status":"ok","timestamp":1774999215889,"user_tz":-120,"elapsed":11,"user":{"displayName":"fukU Google","userId":"02763165356193834046"}},"outputId":"28cffe5f-fa50-4572-bd91-c36ed16bd507"},"execution_count":4,"outputs":[{"output_type":"stream","name":"stdout","text":["💾 Saving all assets to Google Drive...\n","✅ Latents are in /content/drive/MyDrive/flux_klein_latents\n","✅ New 768-dim distilled model is in /content/drive/MyDrive/distilled_qwen_768_for_flux\n","\n","🎉 All data is safely saved on Google Drive.\n","   You can now **disconnect and delete the runtime** if you want.\n","   Everything needed for training is on Drive.\n","   When you come back, start from Cell 3.\n"]}]},{"cell_type":"markdown","source":["You can disconnect the colab past thispoint. All data from cells 1  and 2 are saved to drive."],"metadata":{"id":"cfshTDIFM5ND"}},{"cell_type":"code","source":["# ============================= CELL 3: Install Dependencies & Setup =============================\n","# @title 3. Install Dependencies + Setup Parameters\n","\n","!pip install -q diffusers transformers peft accelerate datasets tqdm\n","\n","import os\n","import torch\n","from google.colab import drive, userdata\n","from transformers import AutoTokenizer, AutoModel\n","from peft import PeftModel\n","import gc\n","\n","drive.mount('/content/drive', force_remount=True)\n","\n","# ====================== Parameters ======================\n","DISTILLED_DIR = \"/content/drive/MyDrive/distilled_qwen_768_for_flux\"\n","LATENT_DIR = \"/content/drive/MyDrive/flux_klein_latents\"\n","FINAL_LORA_DIR = \"/content/drive/MyDrive/flux_klein_lora_final\"\n","\n","BATCH_SIZE = 1          # Keep low for safety\n","NUM_EPOCHS = 8\n","LEARNING_RATE = 1e-4\n","LORA_RANK = 32\n","LORA_ALPHA = 32\n","\n","print(\"✅ Dependencies installed and parameters set.\")\n","print(f\"   Distilled encoder: {DISTILLED_DIR}\")\n","print(f\"   Batch size: {BATCH_SIZE} | Epochs: {NUM_EPOCHS}\")\n","\n","# Optional: quick check\n","print(\"\\n🔍 Quick VRAM check:\")\n","!nvidia-smi --query-gpu=memory.used,memory.total --format=csv"],"metadata":{"id":"ZZaadi1VBK6Z"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["# ============================= CELL 4: LoRA Training on FLUX.2-klein-base-4B (Clean & Reliable) =============================\n","# @title 4. LoRA Training — FLUX.2-klein-base-4B + 768-dim Distilled Qwen\n","\n","import os\n","import torch\n","import torch.nn as nn\n","import torch.nn.functional as F\n","import gc\n","from torch.utils.data import Dataset\n","from tqdm import tqdm\n","from google.colab import drive\n","from transformers import Trainer, TrainingArguments, set_seed\n","from peft import LoraConfig, get_peft_model\n","from diffusers import FluxTransformer2DModel\n","\n","set_seed(42)\n","drive.mount('/content/drive', force_remount=True)\n","\n","print(\"=== CELL 4 START - Clean Restart ===\")\n","print(f\"Current VRAM used: {torch.cuda.memory_allocated()/1024**3:.2f} GB\")\n","\n","# ====================== 1. Load Distilled 768-dim Encoder ======================\n","print(\"\\n[1/5] Loading distilled 768-dim Qwen encoder...\")\n","\n","tokenizer = AutoTokenizer.from_pretrained(DISTILLED_DIR)\n","\n","base_qwen = AutoModel.from_pretrained(\n","    \"Qwen/Qwen2.5-0.5B\",\n","    torch_dtype=torch.float32,\n","    device_map=\"auto\",\n","    trust_remote_code=True,\n","    low_cpu_mem_usage=True\n",")\n","\n","student_model = PeftModel.from_pretrained(base_qwen, DISTILLED_DIR)\n","student_model.eval()\n","\n","projection = nn.Linear(base_qwen.config.hidden_size, 768).to(\"cuda\")\n","projection.load_state_dict(torch.load(f\"{DISTILLED_DIR}/projection.pth\", map_location=\"cuda\"))\n","projection.eval()\n","\n","print(\"[DEBUG] Distilled encoder loaded (768-dim)\")\n","\n","# ====================== 2. Load Data ======================\n","print(\"\\n[2/5] Loading texts and latents...\")\n","\n","data = torch.load(\"/content/drive/MyDrive/qwen_embeddings.pt\", weights_only=False)\n","texts = data[\"texts\"]\n","print(f\"[DEBUG] Loaded {len(texts)} texts\")\n","\n","latent_files = sorted([f for f in os.listdir(LATENT_DIR) if f.endswith(\".pt\")])\n","latents = []\n","for lf in tqdm(latent_files, desc=\"Loading latents\"):\n","    latent = torch.load(os.path.join(LATENT_DIR, lf), weights_only=False)\n","    if latent.dim() == 4 and latent.shape[0] == 1:\n","        latent = latent.squeeze(0)\n","    latents.append(latent)\n","\n","latents = torch.stack(latents)\n","print(f\"[DEBUG] Latents shape: {latents.shape}\")\n","\n","# ====================== 3. Dataset ======================\n","class FluxLoRADataset(Dataset):\n","    def __init__(self, latents, texts):\n","        self.latents = latents\n","        self.texts = texts\n","\n","    def __len__(self): return len(self.latents)\n","\n","    def __getitem__(self, idx):\n","        return {\"latent\": self.latents[idx], \"text\": self.texts[idx]}\n","\n","dataset = FluxLoRADataset(latents, texts)\n","\n","def collate_fn(batch):\n","    return {\n","        \"latent\": torch.stack([item[\"latent\"] for item in batch]),\n","        \"texts\": [item[\"text\"] for item in batch]\n","    }\n","\n","# ====================== 4. Load Transformer + LoRA ======================\n","print(\"\\n[3/5] Loading FLUX.2-klein-base-4B transformer...\")\n","\n","torch.cuda.empty_cache()\n","gc.collect()\n","\n","transformer = FluxTransformer2DModel.from_pretrained(\n","    \"black-forest-labs/FLUX.2-klein-base-4B\",\n","    subfolder=\"transformer\",\n","    torch_dtype=torch.bfloat16,\n","    low_cpu_mem_usage=False,\n",").to(\"cuda\")\n","\n","print(f\"[DEBUG] Transformer loaded. VRAM: {torch.cuda.memory_allocated()/1024**3:.2f} GB\")\n","\n","lora_config = LoraConfig(\n","    r=LORA_RANK,\n","    lora_alpha=LORA_ALPHA,\n","    target_modules=[\n","        \"attn.to_q\", \"attn.to_k\", \"attn.to_v\", \"attn.to_out.0\",\n","        \"attn.to_qkv_mlp_proj\", \"attn.add_q_proj\", \"attn.add_k_proj\", \"attn.add_v_proj\", \"attn.to_add_out\",\n","        \"ff.linear_in\", \"ff.linear_out\", \"ff_context.linear_in\", \"ff_context.linear_out\"\n","    ],\n","    lora_dropout=0.05,\n","    bias=\"none\"\n",")\n","\n","transformer = get_peft_model(transformer, lora_config)\n","transformer.train()\n","\n","print(\"[DEBUG] LoRA applied successfully\")\n","\n","# ====================== 5. Trainer with Simple Repeat Trick ======================\n","class FluxLoRATrainer(Trainer):\n","    def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None):\n","        latents = inputs[\"latent\"].to(dtype=torch.bfloat16, device=model.device)\n","        raw_texts = inputs[\"texts\"]\n","\n","        # Get 768-dim embedding\n","        text_inputs = tokenizer(raw_texts, padding=True, truncation=True, max_length=512, return_tensors=\"pt\").to(\"cuda\")\n","\n","        with torch.no_grad():\n","            outputs = student_model(**text_inputs)\n","            hidden = outputs.last_hidden_state.mean(dim=1)\n","            text_emb_768 = projection(hidden).to(dtype=torch.bfloat16)   # (B, 768)\n","\n","        batch_size = latents.shape[0]\n","        timesteps = torch.rand(batch_size, device=latents.device)\n","        noise = torch.randn_like(latents)\n","\n","        noisy_latents = (1 - timesteps.view(-1, 1, 1, 1)) * latents + timesteps.view(-1, 1, 1, 1) * noise\n","\n","        # Simple repeat trick for encoder_hidden_states (most stable for single-token)\n","        encoder_hidden_states = text_emb_768.unsqueeze(1).repeat(1, 1, 10)   # (B, 1, 7680)\n","\n","        # 2D ids (no batch dimension)\n","        txt_ids = torch.zeros((1, 3), device=latents.device, dtype=torch.bfloat16)\n","        img_ids = torch.zeros((latents.shape[2] * latents.shape[3], 3), device=latents.device, dtype=torch.bfloat16)\n","\n","        model_output = model(\n","            hidden_states=noisy_latents,\n","            timestep=timesteps * 1000,\n","            encoder_hidden_states=encoder_hidden_states,\n","            pooled_projections=text_emb_768,\n","            txt_ids=txt_ids,\n","            img_ids=img_ids,\n","            return_dict=False\n","        )[0]\n","\n","        target = noise - latents\n","        loss = F.mse_loss(model_output, target)\n","\n","        print(f\"[DEBUG] Loss: {loss.item():.6f} | pooled: {text_emb_768.shape} | encoder: {encoder_hidden_states.shape}\")\n","        return (loss, model_output) if return_outputs else loss\n","\n","# ====================== Training ======================\n","training_args = TrainingArguments(\n","    output_dir=\"/content/flux_klein_lora\",\n","    per_device_train_batch_size=BATCH_SIZE,\n","    num_train_epochs=NUM_EPOCHS,\n","    learning_rate=LEARNING_RATE,\n","    lr_scheduler_type=\"cosine\",\n","    warmup_steps=50,\n","    bf16=True,\n","    logging_steps=10,\n","    save_strategy=\"epoch\",\n","    save_total_limit=2,\n","    report_to=\"none\",\n","    remove_unused_columns=False,\n",")\n","\n","trainer = FluxLoRATrainer(\n","    model=transformer,\n","    args=training_args,\n","    train_dataset=dataset,\n","    data_collator=collate_fn,\n",")\n","\n","print(\"\\n🚀 Starting LoRA training with simple repeat trick...\")\n","trainer.train()\n","\n","# ====================== Save ======================\n","final_lora_dir = \"/content/drive/MyDrive/flux_klein_lora_final\"\n","os.makedirs(final_lora_dir, exist_ok=True)\n","transformer.save_pretrained(final_lora_dir)\n","\n","print(f\"\\n✅ Training completed!\")\n","print(f\"   LoRA saved to: {final_lora_dir}\")\n","print(\"   You can now use this LoRA with your 768-dim distilled Qwen for inference.\")\n","\n","torch.cuda.empty_cache()\n","gc.collect()"],"metadata":{"id":"fYbQfjC9RBWY"},"execution_count":null,"outputs":[]}],"metadata":{"accelerator":"GPU","colab":{"gpuType":"T4","provenance":[],"mount_file_id":"1rORehICZ99xZsrwMfy2w8Jxg6M55d81L","authorship_tag":"ABX9TyPxRg2vW/ImNjqp/X8EGqgj"},"kernelspec":{"display_name":"Python 3","name":"python3"},"language_info":{"name":"python"}},"nbformat":4,"nbformat_minor":0}