Datasets:
Tags:
Not-For-All-Audiences
Upload build joycap sdnq.ipynb
Browse files- build joycap sdnq.ipynb +1 -1
build joycap sdnq.ipynb
CHANGED
|
@@ -1 +1 @@
|
|
| 1 |
-
{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"provenance":[
|
|
|
|
| 1 |
+
{"nbformat":4,"nbformat_minor":0,"metadata":{"colab":{"provenance":[],"gpuType":"T4","authorship_tag":"ABX9TyNs/3LbnNmhiqkujqJ3Qhts"},"kernelspec":{"name":"python3","display_name":"Python 3"},"language_info":{"name":"python"},"accelerator":"GPU"},"cells":[{"cell_type":"markdown","source":["# New"],"metadata":{"id":"c0yCIbvJnRAL"}},{"cell_type":"code","source":["#@markdown ## 1 - Setup environment (clean install)\n","\n","from google.colab import drive, userdata\n","\n","drive.mount('/content/drive')\n","\n","hf_token = userdata.get(\"HF_TOKEN\")\n","\n","if hf_token:\n"," from huggingface_hub import login\n"," login(token=hf_token)\n","\n","print(\"✅ Drive + HF login ready\")"],"metadata":{"id":"58JPv1RKnUoz"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["#@markdown ## 2 - Install correct dependencies (FIXED VERSIONS)\n","\n","#!pip uninstall -y transformers tokenizers -q\n","\n","!pip install -q \\\n"," transformers \\\n"," tokenizers \\\n"," accelerate \\\n"," huggingface_hub \\\n"," safetensors \\\n"," sdnq\n","\n","print(\"✅ Dependencies installed\")"],"metadata":{"id":"G9162XsznXiN"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["import torch\n","from PIL import Image\n","from transformers import AutoProcessor, LlavaForConditionalGeneration\n","\n","\n","#IMAGE_PATH = \"image.jpg\"\n","#PROMPT = \"Write a long descriptive caption for this image in a formal tone.\"\n","MODEL_NAME = \"fancyfeast/llama-joycaption-beta-one-hf-llava\"\n","\n","\n","# Load JoyCaption\n","# bfloat16 is the native dtype of the LLM used in JoyCaption (Llama 3.1)\n","# device_map=0 loads the model into the first GPU\n","processor = AutoProcessor.from_pretrained(MODEL_NAME)\n","llava_model = LlavaForConditionalGeneration.from_pretrained(\n"," MODEL_NAME,\n"," torch_dtype=\"bfloat16\",\n"," device_map=\"cpu\",\n"," trust_remote_code=True)\n","\n","#----#\n","llava_model.eval()\n","torch.cuda.empty_cache()\n","gc.collect()"],"metadata":{"id":"AKWy87W2u4zR"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["#@markdown ## 5 - Pipeline wrapper\n","\n","class JoyCaptionPipe:\n"," def __init__(self, model, processor):\n"," self.model = model\n"," self.processor = processor\n"," self.device = torch.device(\"cpu\")\n","\n"," def __call__(self, image, prompt=\"Describe this image\"):\n"," inputs = self.processor(\n"," text=prompt,\n"," images=image,\n"," return_tensors=\"pt\"\n"," )\n","\n"," with torch.no_grad():\n"," output = self.model.generate(\n"," **inputs,\n"," max_new_tokens=128\n"," )\n","\n"," return self.processor.batch_decode(\n"," output,\n"," skip_special_tokens=True\n"," )[0]\n","\n","\n","pipe = JoyCaptionPipe(model, processor)\n","\n","print(\"✅ Pipeline ready\")"],"metadata":{"id":"icfmFpUCnpvE"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["#@markdown ## 6 - SDNQ Configuration (CPU FIXED)\n","\n","from sdnq import SDNQConfig, sdnq_post_load_quant\n","import gc\n","\n","sdnq_config = SDNQConfig(\n"," weights_dtype=\"uin4\",\n"," quantized_matmul_dtype=\"int8\",\n"," group_size=0,\n"," svd_rank=32,\n"," svd_steps=8,\n"," use_svd=False,\n"," quant_conv=False,\n"," quant_embedding=False,\n"," use_quantized_matmul=True,\n"," use_dynamic_quantization=True,\n"," dequantize_fp32=True,\n"," non_blocking=False,\n"," add_skip_keys=True,\n","\n"," # ✅ FIXED (was CUDA → now CPU)\n"," quantization_device=\"cpu\",\n"," return_device=\"cpu\",\n","\n"," modules_to_not_convert=[\n"," \"correction_coefs\",\n"," \"prediction_coefs\",\n"," \"lm_head\",\n"," \"embedding_projection\"\n"," ],\n","\n"," modules_dtype_dict={\"int8\": [\"lm_head\"]},\n","\n"," modules_quant_config={\n"," \"embed_tokens_per_layer\": {\n"," \"quantization_device\": \"cpu\"\n"," }\n"," },\n",")\n","\n","print(\"⏳ Applying SDNQ quantization...\")\n","\n","pipe.model = sdnq_post_load_quant(\n"," pipe.model,\n"," **sdnq_config.__dict__\n",")\n","\n","gc.collect()\n","\n","print(\"✅ SDNQ quantization complete\")"],"metadata":{"id":"57D6k_NJnu78"},"execution_count":null,"outputs":[]},{"cell_type":"markdown","source":["# SDNQ Joycaption"],"metadata":{"id":"PS-ywR2bxNv-"}},{"cell_type":"code","source":["from google.colab import drive, userdata\n","\n","drive.mount('/content/drive')\n","\n","hf_token = userdata.get(\"HF_TOKEN\")\n","\n","if hf_token:\n"," from huggingface_hub import login\n"," login(token=hf_token)\n","\n","print(\"✅ Drive + HF login ready\")"],"metadata":{"id":"9T7ZiDPbxUfE"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["!pip install -q \\\n"," transformers \\\n"," tokenizers \\\n"," accelerate \\\n"," huggingface_hub \\\n"," safetensors \\\n"," sdnq \\\n"," pillow"],"metadata":{"id":"UyuYdMNHxQvg"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["import torch\n","import gc\n","from PIL import Image\n","from transformers import AutoProcessor, LlavaForConditionalGeneration"],"metadata":{"id":"XJMl4aomxXAh"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["MODEL_NAME = \"fancyfeast/llama-joycaption-beta-one-hf-llava\"\n","\n","processor = AutoProcessor.from_pretrained(MODEL_NAME)\n","\n","llava_model = LlavaForConditionalGeneration.from_pretrained(\n"," MODEL_NAME,\n"," torch_dtype=torch.float16,\n"," device_map=\"cpu\", # CPU-first for SDNQ pipeline\n"," trust_remote_code=True\n",")\n","\n","llava_model.eval()\n","\n","gc.collect()\n","print(\"✅ Model loaded\")"],"metadata":{"id":"At8jlIzmxb7q"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["from sdnq import SDNQConfig, sdnq_post_load_quant"],"metadata":{"id":"z2vMUFvhxiSQ"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["sdnq_config = SDNQConfig(\n"," weights_dtype=\"uin4\",\n"," quantized_matmul_dtype=\"int8\",\n"," group_size=0,\n"," svd_rank=32,\n"," svd_steps=8,\n"," use_svd=False,\n"," quant_conv=False,\n"," quant_embedding=False,\n"," use_quantized_matmul=True,\n"," use_dynamic_quantization=True,\n"," dequantize_fp32=True,\n"," non_blocking=False,\n"," add_skip_keys=True,\n","\n"," quantization_device=\"cuda\",\n"," return_device=\"cpu\",\n","\n"," modules_to_not_convert=[\n"," \"correction_coefs\",\n"," \"prediction_coefs\",\n"," \"lm_head\",\n"," \"embedding_projection\"\n"," ],\n","\n"," modules_dtype_dict={\"int8\": [\"lm_head\"]},\n","\n"," modules_quant_config={\n"," \"embed_tokens_per_layer\": {\n"," \"quantization_device\": \"cpu\"\n"," }\n"," },\n",")\n","\n","print(\"⏳ Applying SDNQ quantization...\")"],"metadata":{"id":"gJ21Rz5kxnqA"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["llava_model = sdnq_post_load_quant(\n"," llava_model,\n"," **sdnq_config.__dict__\n",")\n","\n","gc.collect()\n","print(\"✅ SDNQ applied\")"],"metadata":{"id":"rMI_4a4fxw1f"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["class JoyCaptionPipe:\n"," def __init__(self, model, processor):\n"," self.model = model\n"," self.processor = processor\n","\n"," def __call__(self, image, prompt=\"Describe this image in detail.\", max_new_tokens=512):\n","\n"," convo = [\n"," {\"role\": \"system\", \"content\": \"You are a helpful image captioner.\"},\n"," {\"role\": \"user\", \"content\": prompt},\n"," ]\n","\n"," convo_string = self.processor.apply_chat_template(\n"," convo,\n"," tokenize=False,\n"," add_generation_prompt=True\n"," )\n","\n"," inputs = self.processor(\n"," text=[convo_string],\n"," images=[image],\n"," return_tensors=\"pt\"\n"," )\n","\n"," inputs = {k: v.to(\"cpu\") for k, v in inputs.items()}\n","\n"," with torch.no_grad():\n"," output = self.model.generate(\n"," **inputs,\n"," max_new_tokens=max_new_tokens,\n"," do_sample=True,\n"," temperature=0.6,\n"," top_p=0.9\n"," )\n","\n"," # Trim prompt tokens safely\n"," gen_tokens = output[0][inputs[\"input_ids\"].shape[1]:]\n","\n"," caption = self.processor.tokenizer.decode(\n"," gen_tokens,\n"," skip_special_tokens=True,\n"," clean_up_tokenization_spaces=False\n"," )\n","\n"," return caption.strip()"],"metadata":{"id":"mlz-YREhx93E"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["pipe = JoyCaptionPipe(llava_model, processor)\n","print(\"✅ Pipeline ready\")"],"metadata":{"id":"JpL1ZiGTyLMR"},"execution_count":null,"outputs":[]},{"cell_type":"code","source":["IMAGE_PATH = \"image.jpg\" #@param {type:'string'}\n","image = Image.open(IMAGE_PATH)\n","\n","caption = pipe(\n"," image,\n"," prompt=\"Write a long, formal, highly detailed description of this image.\"\n",")\n","\n","print(\"\\n🧾 CAPTION:\\n\")\n","print(caption)"],"metadata":{"cellView":"form","id":"Nsp_fFsmyUNQ"},"execution_count":null,"outputs":[]}]}
|