Add files using upload-large-folder tool

README.md

@@ -0,0 +1,150 @@
+---
+base_model:
+  - Qwen/Qwen-Image
+base_model_relation: quantized
+tags:
+- dfloat11
+- df11
+- lossless compression
+- 70% size, 100% accuracy
+---
+
+# DFloat11 Compressed Model: `Qwen/Qwen-Image`
+
+This is a **DFloat11 losslessly compressed** version of the original `Qwen/Qwen-Image` model. It reduces model size by **32%** compared to the original BFloat16 model, while maintaining **bit-identical outputs** and supporting **efficient GPU inference**.
+
+🔥🔥🔥 Thanks to DFloat11 compression, Qwen-Image can now run on **a single 32GB GPU**, or on **a single 16GB GPU with CPU offloading**, while maintaining full model quality. 🔥🔥🔥
+
+### 📊 Performance Comparison
+
+| Model                                     | Model Size | Peak GPU Memory (1328x1328 image generation) | Generation Time (A100 GPU) |
+|-------------------------------------------|------------|----------------------------------------------|----------------------------|
+| Qwen-Image (BFloat16)                     | ~41 GB     | OOM                                          | -                          |
+| Qwen-Image (DFloat11)                     | 28.42 GB   | 29.74 GB                                     | 100 seconds                |
+| Qwen-Image (DFloat11 + GPU Offloading)    | 28.42 GB   | 16.68 GB                                     | 260 seconds                |
+
+### 🔧 How to Use
+
+1. Install or upgrade the DFloat11 pip package *(installs the CUDA kernel automatically; requires a CUDA-compatible GPU and PyTorch installed)*:
+
+    ```bash
+    pip install -U dfloat11[cuda12]
+    ```
+
+2. Install or upgrade diffusers:
+
+    ```bash
+    pip install git+https://github.com/huggingface/diffusers
+    ```
+
+3. Save the following code to a Python file `qwen_image.py`:
+
+    ```python
+    from diffusers import DiffusionPipeline, QwenImageTransformer2DModel
+    import torch
+    from transformers.modeling_utils import no_init_weights
+    from dfloat11 import DFloat11Model
+    import argparse
+
+    def parse_args():
+        parser = argparse.ArgumentParser(description='Generate images using Qwen-Image model')
+        parser.add_argument('--cpu_offload', action='store_true', help='Enable CPU offloading')
+        parser.add_argument('--prompt', type=str, default='A coffee shop entrance features a chalkboard sign reading "Qwen Coffee 😊 $2 per cup," with a neon light beside it displaying "通义千问". Next to it hangs a poster showing a beautiful Chinese woman, and beneath the poster is written "π≈3.1415926-53589793-23846264-33832795-02384197".',
+                            help='Text prompt for image generation')
+        parser.add_argument('--negative_prompt', type=str, default=' ',
+                            help='Negative prompt for image generation')
+        parser.add_argument('--aspect_ratio', type=str, default='16:9', choices=['1:1', '16:9', '9:16', '4:3', '3:4'],
+                            help='Aspect ratio of generated image')
+        parser.add_argument('--num_inference_steps', type=int, default=50,
+                            help='Number of denoising steps')
+        parser.add_argument('--true_cfg_scale', type=float, default=4.0,
+                            help='Classifier free guidance scale')
+        parser.add_argument('--seed', type=int, default=42,
+                            help='Random seed for generation')
+        parser.add_argument('--output', type=str, default='example.png',
+                            help='Output image path')
+        parser.add_argument('--language', type=str, default='en', choices=['en', 'zh'],
+                            help='Language for positive magic prompt')
+        return parser.parse_args()
+
+    args = parse_args()
+
+    model_name = "Qwen/Qwen-Image"
+
+    with no_init_weights():
+        transformer = QwenImageTransformer2DModel.from_config(
+            QwenImageTransformer2DModel.load_config(
+                model_name, subfolder="transformer",
+            ),
+        ).to(torch.bfloat16)
+
+    DFloat11Model.from_pretrained(
+        "DFloat11/Qwen-Image-DF11",
+        device="cpu",
+        cpu_offload=args.cpu_offload,
+        bfloat16_model=transformer,
+    )
+
+    pipe = DiffusionPipeline.from_pretrained(
+        model_name,
+        transformer=transformer,
+        torch_dtype=torch.bfloat16,
+    )
+    pipe.enable_model_cpu_offload()
+
+    positive_magic = {
+        "en": "Ultra HD, 4K, cinematic composition.", # for english prompt,
+        "zh": "超清，4K，电影级构图" # for chinese prompt,
+    }
+
+    # Generate with different aspect ratios
+    aspect_ratios = {
+        "1:1": (1328, 1328),
+        "16:9": (1664, 928),
+        "9:16": (928, 1664),
+        "4:3": (1472, 1140),
+        "3:4": (1140, 1472),
+    }
+
+    width, height = aspect_ratios[args.aspect_ratio]
+
+    image = pipe(
+        prompt=args.prompt + positive_magic[args.language],
+        negative_prompt=args.negative_prompt,
+        width=width,
+        height=height,
+        num_inference_steps=args.num_inference_steps,
+        true_cfg_scale=args.true_cfg_scale,
+        generator=torch.Generator(device="cuda").manual_seed(args.seed)
+    ).images[0]
+
+    image.save(args.output)
+
+    max_memory = torch.cuda.max_memory_allocated()
+    print(f"Max memory: {max_memory / (1000 ** 3):.2f} GB")
+    ```
+
+4. To run without CPU offloading (32GB VRAM required):
+    ```bash
+    python qwen_image.py
+    ```
+
+    To run with CPU offloading (16GB VRAM required):
+    ```bash
+    python qwen_image.py --cpu_offload
+    ```
+
+
+### 🔍 How It Works
+
+We apply **Huffman coding** to losslessly compress the exponent bits of BFloat16 model weights, which are highly compressible (their 8 bits carry only ~2.6 bits of actual information). To enable fast inference, we implement a highly efficient CUDA kernel that performs on-the-fly weight decompression directly on the GPU.
+
+The result is a model that is **~32% smaller**, delivers **bit-identical outputs**, and achieves performance **comparable to the original** BFloat16 model.
+
+Learn more in our [research paper](https://arxiv.org/abs/2504.11651).
+
+### 📄 Learn More
+
+* **Paper**: [70% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float](https://arxiv.org/abs/2504.11651)
+* **GitHub**: [https://github.com/LeanModels/DFloat11](https://github.com/LeanModels/DFloat11)
+* **HuggingFace**: [https://huggingface.co/DFloat11](https://huggingface.co/DFloat11)

config.json

@@ -0,0 +1,28 @@
+{
+  "dfloat11_config": {
+    "bytes_per_thread": 8,
+    "pattern_dict": {
+      "transformer_blocks\\.\\d+": [
+        "img_mod.1",
+        "attn.to_q",
+        "attn.to_k",
+        "attn.to_v",
+        "attn.add_k_proj",
+        "attn.add_v_proj",
+        "attn.add_q_proj",
+        "attn.to_out.0",
+        "attn.to_add_out",
+        "img_mlp.net.0.proj",
+        "img_mlp.net.2",
+        "txt_mod.1",
+        "txt_mlp.net.0.proj",
+        "txt_mlp.net.2"
+      ]
+    },
+    "threads_per_block": [
+      512
+    ],
+    "version": "0.3.1"
+  },
+  "model_type": "qwen2_5_vl"
+}

diffusion_pytorch_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37837fc3e638dbb9296584e8a417fa8d624fc637e2efb5902ee3cb1f903ddbcd
+size 28423288808