Maxed-Out-99
7fd86b156a
Multiple improvements and fixes across GGUF loading and model patching: - dequant.py: Fix index dtype for IQ4 dequant gathers by casting indices to int64 to avoid dtype issues. - loader.py: - Extend TXT_ARCH_LIST with gemma3. - Fix get_field to extract scalar values via .item(). - Add get_gguf_metadata to collect simple GGUF metadata (string/int/float/bool). - Change gguf_sd_loader to return (state_dict, extra) where extra includes arch_str and metadata. - Dequantize 1D BF16 tensors to float32 to avoid incorrect quantization for bias/1D params. - Add GEMMA3_SD_MAP and gemma3_norm_corrections to reverse a Gemma3-specific norm offset (apply -1.0 correction and dequantize if needed). - Add gguf_gemma3_tokenizer_loader to reconstruct a SentencePiece tokenizer from GGUF metadata. - Update gguf_clip_loader to handle gemma3: map keys, apply norm corrections, and produce tokenizer data. - Ensure gguf_mmproj_loader and other callers unpack the new gguf_sd_loader return value. - nodes.py: - Improve GGUFModelPatcher mmap handling: track named modules to unmap, add pin_weight_to_device to safely move modules when releasing mmap, clear tracking after release. - Pass GGUF metadata into comfy.sd.load_diffusion_model_state_dict when supported, and add error checks when loading fails. These changes add Gemma3 model/tokenizer support, fix dtype and BF16 edge cases, and improve low-memory mmap/unmap handling for safer weight pinning and loading.
ComfyUI-SmartModelLoaders-MXD
Smart, unified model loaders for ComfyUI that support both standard .safetensors and quantized .gguf formats — no switching nodes required.
Includes flexible UNET and CLIP loaders that work across models like SDXL, SD3, Flux, and more.
✅ Features
- 🧠 Unified loaders for
.safetensorsand.ggufformats - 🔀 Drop-in replacements for standard UNET and CLIP nodes
- 💪 Supports single, dual, triple, and quad CLIP configs
- ⚙️ Internal handling of GGUF logic with
GGMLOpsandGGUFModelPatcher - 🧼 Clean fallback to standard ComfyUI loading when needed
📦 Included Nodes
| Node Name | Purpose |
|---|---|
Smart UNET Loader MXD |
Loads UNET from .safetensors or .gguf |
Smart CLIP Loader MXD |
Loads a single CLIP model of any supported format |
Smart Dual CLIP Loader MXD |
Loads 2 CLIPs (ideal for SDXL, Flux, etc.) |
Smart Triple CLIP Loader MXD |
Loads 3 CLIPs (used in SD3 and similar setups) |
Smart Quad CLIP Loader MXD |
Loads 4 CLIPs (for advanced/experimental workflows) |
🧩 Installation
Clone directly into your ComfyUI custom nodes directory:
git clone https://github.com/Maxed-Out-99/ComfyUI-SmartModelLoaders-MXD.git
Install requirements:
pip install -r requirements.txt
🙏 Attribution This project is based on and extends:
city96/ComfyUI-GGUF Licensed under Apache 2.0 License
Modifications, restructuring, and additional loader support by Maxed-Out-99 (2025).