Datasets:
Tags:
Not-For-All-Audiences
Delete kaggle-prune-loras.ipynb
Browse files- kaggle-prune-loras.ipynb +0 -1
kaggle-prune-loras.ipynb
DELETED
|
@@ -1 +0,0 @@
|
|
| 1 |
-
{"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"name":"python","version":"3.12.12","mimetype":"text/x-python","codemirror_mode":{"name":"ipython","version":3},"pygments_lexer":"ipython3","nbconvert_exporter":"python","file_extension":".py"},"kaggle":{"accelerator":"nvidiaTeslaT4","dataSources":[{"sourceType":"datasetVersion","sourceId":16197737,"datasetId":8022630,"databundleVersionId":17176669}],"dockerImageVersionId":31329,"isInternetEnabled":true,"language":"python","sourceType":"notebook","isGpuEnabled":true}},"nbformat_minor":4,"nbformat":4,"cells":[{"cell_type":"code","source":"# =============================================================================\n# CELL 1: Install dependencies\n# =============================================================================\n!pip install -q safetensors torch matplotlib numpy huggingface_hub pynacl py7zr\n\nimport os\nimport torch\nimport gc\nimport random\nimport datetime\n\nprint(f\"✅ PyTorch version: {torch.__version__}\")\nprint(f\"✅ Available GPUs: {torch.cuda.device_count()}\")\nfor i in range(torch.cuda.device_count()):\n print(f\" GPU {i}: {torch.cuda.get_device_name(i)}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-11T15:06:52.769749Z","iopub.execute_input":"2026-05-11T15:06:52.770401Z","iopub.status.idle":"2026-05-11T15:06:58.955400Z","shell.execute_reply.started":"2026-05-11T15:06:52.770368Z","shell.execute_reply":"2026-05-11T15:06:58.954609Z"}},"outputs":[{"name":"stdout","text":"\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m1.4/1.4 MB\u001b[0m \u001b[31m27.7 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m00:01\u001b[0m\n\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m71.3/71.3 kB\u001b[0m \u001b[31m6.3 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m494.3/494.3 kB\u001b[0m \u001b[31m16.1 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m100.6/100.6 kB\u001b[0m \u001b[31m8.2 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m51.5/51.5 kB\u001b[0m \u001b[31m4.4 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n\u001b[2K \u001b[90m━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━\u001b[0m \u001b[32m144.3/144.3 kB\u001b[0m \u001b[31m12.6 MB/s\u001b[0m eta \u001b[36m0:00:00\u001b[0m\n\u001b[?25h✅ PyTorch version: 2.10.0+cu128\n✅ Available GPUs: 2\n GPU 0: Tesla T4\n GPU 1: Tesla T4\n","output_type":"stream"}],"execution_count":2},{"cell_type":"code","source":"# =============================================================================\n# CELL 2: Secrets + Config from Kaggle Dataset\n# =============================================================================\nfrom kaggle_secrets import UserSecretsClient\nfrom huggingface_hub import snapshot_download, login\nimport torch\nimport safetensors.torch\nfrom safetensors import safe_open\nimport os\nimport random\nimport datetime\nimport gc\nfrom pathlib import Path\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\nfrom collections import defaultdict\n\nprint(\"✅ Libraries imported!\")\n\n# === Kaggle Dataset Path ===\nkaggle_repo_path = \"/kaggle/input/datasets/nekos4lyfe/image-caption-dataset/\" #@param {type:\"string\"}\n\n# Read HF_TOKEN from HF_TOKEN.txt\nhf_token = None\ntry:\n token_path = Path(kaggle_repo_path) / \"HF_TOKEN.txt\"\n if token_path.exists():\n hf_token = token_path.read_text(encoding=\"utf-8\").strip()\n print(\"✅ HF_TOKEN loaded from HF_TOKEN.txt\")\n \n if hf_token:\n login(token=hf_token)\n print(\"✅ Successfully logged into Hugging Face\")\n else:\n print(f\"⚠️ HF_TOKEN.txt not found at {token_path}\")\nexcept Exception as e:\n print(f\"⚠️ Failed to read HF_TOKEN.txt: {e}\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-11T15:07:00.032200Z","iopub.execute_input":"2026-05-11T15:07:00.032407Z","iopub.status.idle":"2026-05-11T15:07:00.119107Z","shell.execute_reply.started":"2026-05-11T15:07:00.032387Z","shell.execute_reply":"2026-05-11T15:07:00.118289Z"}},"outputs":[{"name":"stdout","text":"✅ Libraries imported!\n✅ HF_TOKEN loaded from HF_TOKEN.txt\n✅ Successfully logged into Hugging Face\n","output_type":"stream"}],"execution_count":5},{"cell_type":"code","source":"# =============================================================================\n# CELL 3: Setup Folders\n# =============================================================================\nbase_dir = \"/kaggle/working/pruned_loras\"\nos.makedirs(base_dir, exist_ok=True)\nos.makedirs(f\"{base_dir}/processed\", exist_ok=True)\nos.makedirs(f\"{base_dir}/metadata\", exist_ok=True)\n\nprint(f\"Working directory: {base_dir}\")\n\n# Read repo_id from repo_id.txt\nrepo_id = None\ntry:\n repo_id_path = Path(kaggle_repo_path) / \"repo_id.txt\"\n if repo_id_path.exists():\n repo_id = repo_id_path.read_text(encoding=\"utf-8\").strip()\n print(\"✅ repo_id loaded from repo_id.txt\")\n else:\n print(f\"⚠️ repo_id.txt not found at {repo_id_path}\")\nexcept Exception as e:\n print(f\"⚠️ Failed to read repo_id.txt: {e}\")\n#-----#","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-11T15:07:02.534600Z","iopub.execute_input":"2026-05-11T15:07:02.535306Z","iopub.status.idle":"2026-05-11T15:07:02.547315Z","shell.execute_reply.started":"2026-05-11T15:07:02.535275Z","shell.execute_reply":"2026-05-11T15:07:02.546652Z"}},"outputs":[{"name":"stdout","text":"Working directory: /kaggle/working/pruned_loras\n✅ repo_id loaded from repo_id.txt\n","output_type":"stream"}],"execution_count":6},{"cell_type":"code","source":"#repo_id = \"Winnougan/Must_Have_Klein_9b_loras\" #@param {type:\"string\"}\n#\"Realism_Engine_Klein_V2.*\",\n#\"Flux.2 Klein 9B - Thong Panties.*\" , \"F2K9B_chairtied_cha1rt1ed_V2E10.*\", \"klein_snofs_v1_2.*\" , \"f2k_consist_20260225.*\" , \"darkBeastMar2126Latest_dbkBlitzV15.*\"], \n# =============================================================================\n# CELL 4: Download + Smart Letter Naming\n# =============================================================================\nprint(f\"Downloading .safetensors from {repo_id} ...\")\n\nsnapshot_download(\n repo_id=repo_id,\n local_dir=base_dir,\n token=hf_token,\n ignore_patterns=[\"*.md\", \"*.txt\", \"*.json\", \"*.png\", \"*.jpg\"],\n allow_patterns=[\"Flux Klein - NSFW v2.*\", \"Flux.2 Klein 9B - Tiny Panties.*\"],\n)\n\noriginal_files = sorted([f for f in os.listdir(base_dir) if f.endswith(\".safetensors\")])\nprint(f\"✅ Downloaded {len(original_files)} LoRA files.\")\n\n# ====================== SMART LETTER GENERATOR ======================\ndef generate_letter_labels(n):\n labels = []\n i = 0\n while len(labels) < n:\n if i < 26:\n labels.append(chr(65 + i))\n else:\n first = chr(65 + (i - 26) // 26)\n second = chr(65 + (i - 26) % 26)\n labels.append(first + second)\n i += 1\n return labels\n\nletters = generate_letter_labels(len(original_files))\n\n# Create mapping file BEFORE renaming\nmapping_path = os.path.join(base_dir, \"original_mapping.txt\")\nwith open(mapping_path, \"w\", encoding=\"utf-8\") as f:\n f.write(\"Original Name → New Short Name\\n\")\n f.write(\"=\" * 70 + \"\\n\")\n for old_name, letter in zip(original_files, letters):\n f.write(f\"{old_name} → {letter}.safetensors\\n\")\n\nprint(f\"✅ Mapping saved to: {mapping_path}\")\n\n# Rename files\nlora_files = []\nfor old_name, letter in zip(original_files, letters):\n new_name = f\"{letter}.safetensors\"\n old_path = os.path.join(base_dir, old_name)\n new_path = os.path.join(base_dir, new_name)\n \n os.rename(old_path, new_path)\n lora_files.append(new_name)\n print(f\" {old_name} → {new_name}\")\n\nprint(f\"✅ All files renamed using A-Z + AA, AB... system.\")","metadata":{"trusted":true,"execution":{"iopub.status.busy":"2026-05-11T15:07:04.928606Z","iopub.execute_input":"2026-05-11T15:07:04.929290Z","iopub.status.idle":"2026-05-11T15:07:08.226168Z","shell.execute_reply.started":"2026-05-11T15:07:04.929259Z","shell.execute_reply":"2026-05-11T15:07:08.225223Z"}},"outputs":[{"name":"stdout","text":"Downloading .safetensors from Winnougan/Must_Have_Klein_9b_loras ...\n","output_type":"stream"},{"output_type":"display_data","data":{"text/plain":"Downloading (incomplete total...): 0.00B [00:00, ?B/s]","application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"042b1b80871e45c89a3fc86fffcd4dbb"}},"metadata":{}},{"output_type":"display_data","data":{"text/plain":"Fetching 2 files: 0%| | 0/2 [00:00<?, ?it/s]","application/vnd.jupyter.widget-view+json":{"version_major":2,"version_minor":0,"model_id":"df5d2c5016b74ff0ace04ce537d7d4da"}},"metadata":{}},{"name":"stdout","text":"✅ Downloaded 2 LoRA files.\n✅ Mapping saved to: /kaggle/working/pruned_loras/original_mapping.txt\n Flux Klein - NSFW v2.safetensors → A.safetensors\n Flux.2 Klein 9B - Tiny Panties.safetensors → B.safetensors\n✅ All files renamed using A-Z + AA, AB... system.\n","output_type":"stream"}],"execution_count":7},{"cell_type":"code","source":"# =============================================================================\n# CELL 5: Improved Pruning with Proper Key Removal + Original Name Tracking\n# =============================================================================\nfrom collections import defaultdict\nimport time\nimport datetime\nimport os\nimport random\nimport gc\nimport torch\nfrom concurrent.futures import ThreadPoolExecutor, as_completed\nfrom pathlib import Path\nfrom safetensors import safe_open\n\ndesired_accuracy = 90 #@param {type:\"slider\", min:80, max:100, step:1}\n\nprint(f\"🎯 Target Accuracy: {desired_accuracy}% | Using 2 threads\")\n\n# ====================== LOAD ORIGINAL NAME MAPPING ======================\noriginal_to_letter = {}\nmapping_path = os.path.join(base_dir, \"original_mapping.txt\")\nif os.path.exists(mapping_path):\n with open(mapping_path, \"r\", encoding=\"utf-8\") as f:\n for line in f:\n if \"→\" in line:\n old, new = line.strip().split(\"→\")\n original_to_letter[new.strip()] = old.strip()\n print(f\"✅ Loaded {len(original_to_letter)} original filename mappings\")\nelse:\n print(\"⚠️ original_mapping.txt not found - will use short names\")\n\n# ====================== WORKER FUNCTION ======================\ndef worker(thread_id: int, gpu_id: int, file_list: list):\n print(f\"[Thread {thread_id} | GPU {gpu_id}] 🚀 Started with {len(file_list)} files\")\n torch.cuda.set_device(gpu_id)\n \n for idx, filename in enumerate(file_list):\n letter = filename.replace(\".safetensors\", \"\")\n input_path = os.path.join(base_dir, filename)\n size_before = os.path.getsize(input_path) / (1024 * 1024)\n \n print(f\"[Thread {thread_id} | GPU {gpu_id}] 🔄 Processing [{letter}] {filename} ({size_before:.2f} MB)\")\n\n torch.cuda.empty_cache()\n gc.collect()\n\n # Get original filename for metadata\n original_filename = original_to_letter.get(filename, filename)\n\n random_suffix = random.randint(10000, 99999)\n output_name = f\"{letter}_{random_suffix}.safetensors\"\n output_path = f\"{base_dir}/processed/{output_name}\"\n\n # Load file\n state_dict = safetensors.torch.load_file(input_path, device=\"cpu\")\n \n # Load metadata\n original_metadata = {}\n try:\n with safe_open(input_path, framework=\"pt\", device=\"cpu\") as f:\n original_metadata = f.metadata() or {}\n except:\n pass\n\n # Parse LoRA layers (supports both lora_A/B and lora_down/up)\n lora_pairs = defaultdict(dict)\n original_ranks = {}\n\n for key, tensor in state_dict.items():\n if \".lora_A.weight\" in key or \".lora_down.weight\" in key:\n base = key.replace(\".lora_A.weight\", \"\").replace(\".lora_down.weight\", \"\")\n lora_pairs[base][\"A\"] = tensor\n original_ranks[base] = tensor.shape[0]\n elif \".lora_B.weight\" in key or \".lora_up.weight\" in key:\n base = key.replace(\".lora_B.weight\", \"\").replace(\".lora_up.weight\", \"\")\n lora_pairs[base][\"B\"] = tensor\n\n # Pruning\n new_state_dict = dict(state_dict) # copy\n changes_made = False\n pruned_count = 0\n total_layers = len(lora_pairs)\n layer_log = []\n\n for base_key, matrices in lora_pairs.items():\n if \"A\" not in matrices or \"B\" not in matrices:\n continue\n\n A = matrices[\"A\"].to(gpu_id, non_blocking=True)\n B = matrices[\"B\"].to(gpu_id, non_blocking=True)\n orig_rank = original_ranks[base_key]\n\n delta_W = torch.matmul(B, A).float()\n original_norm = torch.norm(delta_W, p='fro').item()\n\n U, S, Vh = torch.linalg.svd(delta_W, full_matrices=False)\n\n # Try different ranks\n candidates = [max(1, int(orig_rank * f)) for f in [0.95, 0.9, 0.85, 0.8, 0.75, 0.7, 0.65, 0.6]]\n test_ranks = sorted(set(r for r in candidates if r < orig_rank))\n\n best_rank = orig_rank\n best_acc = 100.0\n\n for r in test_ranks:\n if r >= len(S): continue\n S_trunc = torch.zeros_like(S)\n S_trunc[:r] = S[:r]\n approx = U @ torch.diag(S_trunc) @ Vh\n error = torch.norm(delta_W - approx, p='fro').item()\n rel_error = (error / original_norm * 100) if original_norm > 1e-8 else 0.0\n acc = 100 - rel_error\n \n if acc >= desired_accuracy:\n best_rank = r\n best_acc = acc\n break\n\n if best_rank < orig_rank:\n changes_made = True\n pruned_count += 1\n r = best_rank\n\n sqrt_S = torch.sqrt(S[:r])\n new_up = (U[:, :r] * sqrt_S.unsqueeze(0)).to(A.dtype) # lora_up\n new_down = (sqrt_S.unsqueeze(1) * Vh[:r, :]).to(A.dtype) # lora_down\n\n # === REMOVE OLD KEYS ===\n for suffix in [\".lora_A.weight\", \".lora_B.weight\", \".lora_down.weight\", \".lora_up.weight\"]:\n new_state_dict.pop(base_key + suffix, None)\n\n # === ADD NEW PRUNED KEYS (standard format) ===\n new_state_dict[base_key + \".lora_down.weight\"] = new_down.contiguous().cpu()\n new_state_dict[base_key + \".lora_up.weight\"] = new_up.contiguous().cpu()\n\n print(f\" → {base_key} | {orig_rank} → {best_rank} | Acc {best_acc:.2f}%\")\n layer_log.append(f\"{base_key} | {orig_rank:3} → {best_rank:3} | Pruned | Acc {best_acc:.2f}%\")\n else:\n layer_log.append(f\"{base_key} | {orig_rank:3} → {orig_rank:3} | No change\")\n\n # Cleanup GPU memory\n del A, B, delta_W, U, S, Vh\n torch.cuda.empty_cache()\n\n # Delete original file\n if os.path.exists(input_path):\n os.remove(input_path)\n\n if not changes_made:\n print(f\"[Thread {thread_id} | GPU {gpu_id}] → No pruning needed for [{letter}]\")\n continue\n\n # === SAVE PRUNED FILE ===\n new_metadata = dict(original_metadata)\n new_metadata.update({\n \"original_file\": original_filename,\n \"pruned_with\": f\"SVD Rank Reduction at {desired_accuracy}% accuracy\",\n \"pruning_date\": datetime.datetime.now().isoformat(),\n \"random_id\": str(random_suffix),\n \"layers_pruned\": f\"{pruned_count}/{total_layers}\"\n })\n\n # Ensure all tensors are contiguous\n for k, v in new_state_dict.items():\n if isinstance(v, torch.Tensor):\n new_state_dict[k] = v.contiguous()\n\n safetensors.torch.save_file(new_state_dict, output_path, metadata=new_metadata)\n\n size_after = os.path.getsize(output_path) / (1024 * 1024)\n\n # Save detailed log\n meta_dir = f\"{base_dir}/metadata/{letter}\"\n os.makedirs(meta_dir, exist_ok=True)\n with open(f\"{meta_dir}/{letter}_metadata.txt\", \"w\", encoding=\"utf-8\") as f:\n f.write(f\"Original filename: {original_filename}\\n\")\n f.write(f\"Pruned filename : {output_name}\\n\")\n f.write(f\"Size Before : {size_before:.2f} MB\\n\")\n f.write(f\"Size After : {size_after:.2f} MB\\n\")\n f.write(f\"Layers Pruned : {pruned_count}/{total_layers}\\n\\n\")\n f.write(\"=== LAYER DETAILS ===\\n\")\n for line in layer_log:\n f.write(line + \"\\n\")\n\n print(f\"[Thread {thread_id} | GPU {gpu_id}] ✅ Saved [{letter}] | {size_before:.1f}MB → {size_after:.1f}MB\")\n\n # Final cleanup\n del state_dict, new_state_dict\n torch.cuda.empty_cache()\n gc.collect()\n\n print(f\"[Thread {thread_id} | GPU {gpu_id}] 🏁 Worker completed\")\n return True\n\n\n# ====================== LAUNCH THREADS ======================\nprint(f\"\\nDistributing {len(lora_files)} files across threads...\")\n\nnum_threads = 2\nchunks = [[] for _ in range(num_threads)]\nfor i, f in enumerate(lora_files):\n chunks[i % num_threads].append(f)\n\nwith ThreadPoolExecutor(max_workers=num_threads) as executor:\n futures = []\n for t in range(num_threads):\n gpu_id = t % 2\n future = executor.submit(worker, t, gpu_id, chunks[t])\n futures.append(future)\n print(f\" → Thread {t} started on GPU {gpu_id} ({len(chunks[t])} files)\")\n if t < num_threads - 1:\n time.sleep(15) # stagger start\n\n for future in as_completed(futures):\n future.result()\n\nprint(\"\\n🎉 All LoRA pruning completed successfully!\")","metadata":{"_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","trusted":true,"execution":{"iopub.status.busy":"2026-05-11T15:07:08.227521Z","iopub.execute_input":"2026-05-11T15:07:08.227869Z","iopub.status.idle":"2026-05-11T15:33:01.107624Z","shell.execute_reply.started":"2026-05-11T15:07:08.227843Z","shell.execute_reply":"2026-05-11T15:33:01.106924Z"}},"outputs":[{"name":"stdout","text":"🎯 Target Accuracy: 90% | Using 2 threads\n✅ Loaded 3 original filename mappings\n\nDistributing 2 files across threads...\n[Thread 0 | GPU 0] 🚀 Started with 1 files\n → Thread 0 started on GPU 0 (1 files)\n[Thread 0 | GPU 0] 🔄 Processing [A] A.safetensors (158.03 MB)\n → diffusion_model.double_blocks.0.img_attn.proj | 32 → 24 | Acc 91.33%\n[Thread 1 | GPU 1] 🚀 Started with 1 files\n → Thread 1 started on GPU 1 (1 files)\n[Thread 1 | GPU 1] 🔄 Processing [B] B.safetensors (158.03 MB)\n → diffusion_model.double_blocks.0.img_attn.qkv | 32 → 25 | Acc 90.94%\n → diffusion_model.double_blocks.0.img_attn.proj | 32 → 24 | Acc 91.26%\n → diffusion_model.double_blocks.0.img_mlp.0 | 32 → 22 | Acc 91.10%\n → diffusion_model.double_blocks.0.img_attn.qkv | 32 → 22 | Acc 90.31%\n → diffusion_model.double_blocks.0.img_mlp.2 | 32 → 19 | Acc 92.42%\n → diffusion_model.double_blocks.0.img_mlp.0 | 32 → 19 | Acc 91.21%\n → diffusion_model.double_blocks.0.txt_attn.proj | 32 → 19 | Acc 98.71%\n → diffusion_model.double_blocks.0.img_mlp.2 | 32 → 19 | Acc 91.51%\n → diffusion_model.double_blocks.0.txt_attn.proj | 32 → 19 | Acc 95.95%\n → diffusion_model.double_blocks.0.txt_attn.qkv | 32 → 19 | Acc 96.81%\n → diffusion_model.double_blocks.0.txt_attn.qkv | 32 → 19 | Acc 97.04%\n → diffusion_model.double_blocks.0.txt_mlp.0 | 32 → 19 | Acc 97.69%\n → diffusion_model.double_blocks.0.txt_mlp.0 | 32 → 19 | Acc 96.53%\n → diffusion_model.double_blocks.0.txt_mlp.2 | 32 → 19 | Acc 98.48%\n → diffusion_model.double_blocks.0.txt_mlp.2 | 32 → 19 | Acc 98.28%\n → diffusion_model.double_blocks.1.img_attn.proj | 32 → 20 | Acc 90.07%\n → diffusion_model.double_blocks.1.img_attn.proj | 32 → 19 | Acc 90.53%\n → diffusion_model.double_blocks.1.img_attn.qkv | 32 → 19 | Acc 91.94%\n → diffusion_model.double_blocks.1.img_attn.qkv | 32 → 19 | Acc 95.55%\n → diffusion_model.double_blocks.1.img_mlp.0 | 32 → 22 | Acc 91.35%\n → diffusion_model.double_blocks.1.img_mlp.0 | 32 → 19 | Acc 90.72%\n → diffusion_model.double_blocks.1.img_mlp.2 | 32 → 19 | Acc 90.48%\n → diffusion_model.double_blocks.1.img_mlp.2 | 32 → 19 | Acc 93.38%\n → diffusion_model.double_blocks.1.txt_attn.proj | 32 → 19 | Acc 95.63%\n → diffusion_model.double_blocks.1.txt_attn.proj | 32 → 19 | Acc 98.28%\n → diffusion_model.double_blocks.1.txt_attn.qkv | 32 → 19 | Acc 97.65%\n → diffusion_model.double_blocks.1.txt_attn.qkv | 32 → 19 | Acc 98.89%\n → diffusion_model.double_blocks.1.txt_mlp.0 | 32 → 19 | Acc 97.26%\n → diffusion_model.double_blocks.1.txt_mlp.0 | 32 → 19 | Acc 97.54%\n → diffusion_model.double_blocks.1.txt_mlp.2 | 32 → 19 | Acc 95.66%\n → diffusion_model.double_blocks.1.txt_mlp.2 | 32 → 19 | Acc 98.34%\n → diffusion_model.double_blocks.2.img_attn.proj | 32 → 22 | Acc 90.51%\n → diffusion_model.double_blocks.2.img_attn.proj | 32 → 22 | Acc 90.26%\n → diffusion_model.double_blocks.2.img_attn.qkv | 32 → 19 | Acc 91.73%\n → diffusion_model.double_blocks.2.img_attn.qkv | 32 → 19 | Acc 90.35%\n → diffusion_model.double_blocks.2.img_mlp.0 | 32 → 27 | Acc 90.55%\n → diffusion_model.double_blocks.2.img_mlp.0 | 32 → 24 | Acc 90.16%\n → diffusion_model.double_blocks.2.img_mlp.2 | 32 → 24 | Acc 90.32%\n → diffusion_model.double_blocks.2.img_mlp.2 | 32 → 24 | Acc 91.02%\n → diffusion_model.double_blocks.2.txt_attn.proj | 32 → 19 | Acc 94.88%\n → diffusion_model.double_blocks.2.txt_attn.proj | 32 → 19 | Acc 96.74%\n → diffusion_model.double_blocks.2.txt_attn.qkv | 32 → 19 | Acc 96.40%\n → diffusion_model.double_blocks.2.txt_attn.qkv | 32 → 19 | Acc 98.28%\n → diffusion_model.double_blocks.2.txt_mlp.0 | 32 → 22 | Acc 90.80%\n → diffusion_model.double_blocks.2.txt_mlp.0 | 32 → 19 | Acc 95.65%\n → diffusion_model.double_blocks.2.txt_mlp.2 | 32 → 19 | Acc 91.27%\n → diffusion_model.double_blocks.3.img_attn.proj | 32 → 27 | Acc 90.81%\n → diffusion_model.double_blocks.2.txt_mlp.2 | 32 → 19 | Acc 96.08%\n → diffusion_model.double_blocks.3.img_attn.qkv | 32 → 27 | Acc 91.68%\n → diffusion_model.double_blocks.3.img_attn.proj | 32 → 24 | Acc 90.14%\n → diffusion_model.double_blocks.3.img_attn.qkv | 32 → 24 | Acc 90.76%\n → diffusion_model.double_blocks.3.img_mlp.0 | 32 → 30 | Acc 92.76%\n → diffusion_model.double_blocks.3.img_mlp.2 | 32 → 27 | Acc 91.13%\n → diffusion_model.double_blocks.3.img_mlp.0 | 32 → 24 | Acc 90.38%\n → diffusion_model.double_blocks.3.txt_attn.proj | 32 → 22 | Acc 90.35%\n → diffusion_model.double_blocks.3.img_mlp.2 | 32 → 25 | Acc 90.85%\n → diffusion_model.double_blocks.3.txt_attn.qkv | 32 → 19 | Acc 92.30%\n → diffusion_model.double_blocks.3.txt_attn.proj | 32 → 19 | Acc 95.13%\n → diffusion_model.double_blocks.3.txt_mlp.0 | 32 → 19 | Acc 91.21%\n → diffusion_model.double_blocks.3.txt_attn.qkv | 32 → 19 | Acc 96.85%\n → diffusion_model.double_blocks.3.txt_mlp.2 | 32 → 22 | Acc 90.86%\n → diffusion_model.double_blocks.3.txt_mlp.0 | 32 → 19 | Acc 94.86%\n → diffusion_model.double_blocks.4.img_attn.proj | 32 → 25 | Acc 90.58%\n → diffusion_model.double_blocks.3.txt_mlp.2 | 32 → 19 | Acc 94.56%\n → diffusion_model.double_blocks.4.img_attn.qkv | 32 → 25 | Acc 90.72%\n → diffusion_model.double_blocks.4.img_attn.proj | 32 → 24 | Acc 90.82%\n → diffusion_model.double_blocks.4.img_mlp.0 | 32 → 28 | Acc 90.61%\n → diffusion_model.double_blocks.4.img_attn.qkv | 32 → 24 | Acc 90.03%\n → diffusion_model.double_blocks.4.img_mlp.2 | 32 → 22 | Acc 90.97%\n → diffusion_model.double_blocks.4.img_mlp.0 | 32 → 27 | Acc 91.21%\n → diffusion_model.double_blocks.4.txt_attn.proj | 32 → 20 | Acc 90.31%\n → diffusion_model.double_blocks.4.img_mlp.2 | 32 → 22 | Acc 90.13%\n → diffusion_model.double_blocks.4.txt_attn.qkv | 32 → 19 | Acc 92.77%\n → diffusion_model.double_blocks.4.txt_attn.proj | 32 → 19 | Acc 96.41%\n → diffusion_model.double_blocks.4.txt_mlp.0 | 32 → 19 | Acc 91.33%\n → diffusion_model.double_blocks.4.txt_attn.qkv | 32 → 19 | Acc 95.00%\n → diffusion_model.double_blocks.4.txt_mlp.2 | 32 → 20 | Acc 90.06%\n → diffusion_model.double_blocks.4.txt_mlp.0 | 32 → 19 | Acc 92.98%\n → diffusion_model.double_blocks.5.img_attn.proj | 32 → 22 | Acc 91.19%\n → diffusion_model.double_blocks.4.txt_mlp.2 | 32 → 19 | Acc 95.22%\n → diffusion_model.double_blocks.5.img_attn.qkv | 32 → 25 | Acc 90.76%\n → diffusion_model.double_blocks.5.img_attn.proj | 32 → 22 | Acc 90.69%\n → diffusion_model.double_blocks.5.img_mlp.0 | 32 → 27 | Acc 90.46%\n → diffusion_model.double_blocks.5.img_attn.qkv | 32 → 24 | Acc 90.06%\n → diffusion_model.double_blocks.5.img_mlp.2 | 32 → 19 | Acc 92.87%\n → diffusion_model.double_blocks.5.txt_attn.proj | 32 → 19 | Acc 95.40%\n → diffusion_model.double_blocks.5.img_mlp.0 | 32 → 24 | Acc 91.16%\n → diffusion_model.double_blocks.5.txt_attn.qkv | 32 → 19 | Acc 96.60%\n → diffusion_model.double_blocks.5.img_mlp.2 | 32 → 20 | Acc 90.67%\n → diffusion_model.double_blocks.5.txt_mlp.0 | 32 → 19 | Acc 95.51%\n → diffusion_model.double_blocks.5.txt_attn.proj | 32 → 19 | Acc 95.61%\n → diffusion_model.double_blocks.5.txt_mlp.2 | 32 → 19 | Acc 90.82%\n → diffusion_model.double_blocks.5.txt_attn.qkv | 32 → 19 | Acc 96.67%\n → diffusion_model.double_blocks.6.img_attn.proj | 32 → 19 | Acc 90.32%\n → diffusion_model.double_blocks.5.txt_mlp.0 | 32 → 19 | Acc 94.48%\n → diffusion_model.double_blocks.6.img_attn.qkv | 32 → 24 | Acc 91.35%\n → diffusion_model.double_blocks.5.txt_mlp.2 | 32 → 19 | Acc 92.61%\n → diffusion_model.double_blocks.6.img_attn.proj | 32 → 19 | Acc 93.62%\n → diffusion_model.double_blocks.6.img_mlp.0 | 32 → 27 | Acc 91.53%\n → diffusion_model.double_blocks.6.img_mlp.2 | 32 → 19 | Acc 93.64%\n → diffusion_model.double_blocks.6.img_attn.qkv | 32 → 22 | Acc 90.57%\n → diffusion_model.double_blocks.6.txt_attn.proj | 32 → 22 | Acc 90.51%\n → diffusion_model.double_blocks.6.img_mlp.0 | 32 → 22 | Acc 90.50%\n → diffusion_model.double_blocks.6.txt_attn.qkv | 32 → 19 | Acc 93.88%\n → diffusion_model.double_blocks.6.img_mlp.2 | 32 → 19 | Acc 94.18%\n → diffusion_model.double_blocks.6.txt_mlp.0 | 32 → 19 | Acc 91.15%\n → diffusion_model.double_blocks.6.txt_attn.proj | 32 → 19 | Acc 96.54%\n → diffusion_model.double_blocks.6.txt_mlp.2 | 32 → 24 | Acc 90.70%\n → diffusion_model.double_blocks.6.txt_attn.qkv | 32 → 19 | Acc 97.62%\n → diffusion_model.double_blocks.7.img_attn.proj | 32 → 24 | Acc 90.59%\n → diffusion_model.double_blocks.6.txt_mlp.0 | 32 → 19 | Acc 91.31%\n → diffusion_model.double_blocks.7.img_attn.qkv | 32 → 25 | Acc 90.45%\n → diffusion_model.double_blocks.6.txt_mlp.2 | 32 → 22 | Acc 90.33%\n → diffusion_model.double_blocks.7.img_mlp.0 | 32 → 19 | Acc 91.63%\n → diffusion_model.double_blocks.7.img_attn.proj | 32 → 27 | Acc 90.96%\n → diffusion_model.double_blocks.7.img_mlp.2 | 32 → 24 | Acc 90.49%\n → diffusion_model.double_blocks.7.img_attn.qkv | 32 → 24 | Acc 91.45%\n → diffusion_model.double_blocks.7.txt_attn.proj | 32 → 19 | Acc 91.50%\n → diffusion_model.double_blocks.7.txt_attn.qkv | 32 → 19 | Acc 92.45%\n → diffusion_model.double_blocks.7.img_mlp.0 | 32 → 24 | Acc 91.86%\n → diffusion_model.double_blocks.7.txt_mlp.0 | 32 → 19 | Acc 90.43%\n → diffusion_model.double_blocks.7.img_mlp.2 | 32 → 30 | Acc 92.25%\n → diffusion_model.double_blocks.7.txt_mlp.2 | 32 → 19 | Acc 93.34%\n → diffusion_model.double_blocks.7.txt_attn.proj | 32 → 19 | Acc 93.94%\n → diffusion_model.single_blocks.0.linear1 | 32 → 22 | Acc 90.57%\n → diffusion_model.double_blocks.7.txt_attn.qkv | 32 → 19 | Acc 95.04%\n → diffusion_model.single_blocks.0.linear2 | 32 → 19 | Acc 91.79%\n → diffusion_model.double_blocks.7.txt_mlp.0 | 32 → 19 | Acc 90.78%\n → diffusion_model.double_blocks.7.txt_mlp.2 | 32 → 19 | Acc 92.14%\n → diffusion_model.single_blocks.1.linear1 | 32 → 22 | Acc 90.67%\n → diffusion_model.single_blocks.1.linear2 | 32 → 19 | Acc 90.04%\n → diffusion_model.single_blocks.0.linear1 | 32 → 25 | Acc 90.15%\n → diffusion_model.single_blocks.10.linear1 | 32 → 30 | Acc 92.09%\n → diffusion_model.single_blocks.0.linear2 | 32 → 24 | Acc 91.08%\n → diffusion_model.single_blocks.10.linear2 | 32 → 30 | Acc 92.46%\n → diffusion_model.single_blocks.1.linear1 | 32 → 27 | Acc 91.98%\n → diffusion_model.single_blocks.11.linear1 | 32 → 30 | Acc 90.85%\n → diffusion_model.single_blocks.1.linear2 | 32 → 28 | Acc 90.39%\n → diffusion_model.single_blocks.11.linear2 | 32 → 30 | Acc 92.34%\n → diffusion_model.single_blocks.12.linear1 | 32 → 30 | Acc 90.59%\n → diffusion_model.single_blocks.10.linear2 | 32 → 30 | Acc 92.14%\n → diffusion_model.single_blocks.12.linear2 | 32 → 30 | Acc 92.01%\n → diffusion_model.single_blocks.13.linear2 | 32 → 30 | Acc 90.94%\n → diffusion_model.single_blocks.14.linear2 | 32 → 30 | Acc 91.17%\n → diffusion_model.single_blocks.15.linear2 | 32 → 30 | Acc 91.01%\n → diffusion_model.single_blocks.2.linear1 | 32 → 24 | Acc 90.51%\n → diffusion_model.single_blocks.2.linear2 | 32 → 22 | Acc 90.05%\n → diffusion_model.single_blocks.2.linear1 | 32 → 27 | Acc 91.13%\n → diffusion_model.single_blocks.2.linear2 | 32 → 28 | Acc 90.66%\n → diffusion_model.single_blocks.23.linear2 | 32 → 30 | Acc 91.78%\n → diffusion_model.single_blocks.3.linear1 | 32 → 27 | Acc 90.14%\n → diffusion_model.single_blocks.3.linear2 | 32 → 24 | Acc 91.03%\n → diffusion_model.single_blocks.4.linear1 | 32 → 27 | Acc 90.46%\n → diffusion_model.single_blocks.4.linear2 | 32 → 27 | Acc 91.13%\n → diffusion_model.single_blocks.5.linear1 | 32 → 30 | Acc 91.96%\n → diffusion_model.single_blocks.5.linear2 | 32 → 27 | Acc 90.87%\n → diffusion_model.single_blocks.3.linear1 | 32 → 30 | Acc 92.66%\n → diffusion_model.single_blocks.3.linear2 | 32 → 27 | Acc 91.42%\n → diffusion_model.single_blocks.6.linear1 | 32 → 30 | Acc 92.05%\n → diffusion_model.single_blocks.6.linear2 | 32 → 28 | Acc 91.13%\n → diffusion_model.single_blocks.4.linear1 | 32 → 30 | Acc 90.21%\n → diffusion_model.single_blocks.7.linear1 | 32 → 30 | Acc 91.25%\n → diffusion_model.single_blocks.4.linear2 | 32 → 28 | Acc 90.79%\n → diffusion_model.single_blocks.7.linear2 | 32 → 28 | Acc 90.22%\n → diffusion_model.single_blocks.5.linear1 | 32 → 30 | Acc 90.03%\n → diffusion_model.single_blocks.8.linear1 | 32 → 30 | Acc 91.68%\n → diffusion_model.single_blocks.5.linear2 | 32 → 30 | Acc 92.90%\n → diffusion_model.single_blocks.8.linear2 | 32 → 28 | Acc 90.99%\n → diffusion_model.single_blocks.9.linear1 | 32 → 30 | Acc 91.27%\n → diffusion_model.single_blocks.6.linear2 | 32 → 30 | Acc 92.28%\n → diffusion_model.single_blocks.9.linear2 | 32 → 28 | Acc 90.49%\n[Thread 1 | GPU 1] ✅ Saved [B] | 158.0MB → 128.1MB\n[Thread 1 | GPU 1] 🏁 Worker completed\n → diffusion_model.single_blocks.7.linear2 | 32 → 30 | Acc 91.38%\n → diffusion_model.single_blocks.8.linear2 | 32 → 30 | Acc 91.63%\n → diffusion_model.single_blocks.9.linear2 | 32 → 30 | Acc 91.48%\n[Thread 0 | GPU 0] ✅ Saved [A] | 158.0MB → 131.5MB\n[Thread 0 | GPU 0] 🏁 Worker completed\n\n🎉 All LoRA pruning completed successfully!\n","output_type":"stream"}],"execution_count":8}]}
|
|
|
|
|
|