Add README
Browse files
README.md
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model:
|
| 4 |
+
- vllm-sr/d3-flash
|
| 5 |
+
- vllm-sr/d3-lite
|
| 6 |
+
library_name: diffusion-single-file
|
| 7 |
+
tags:
|
| 8 |
+
- comfyui
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# d3 (ComfyUI)
|
| 12 |
+
|
| 13 |
+
The backbones of [vllm-sr/d3-flash](https://huggingface.co/vllm-sr/d3-flash) and [vllm-sr/d3-lite](https://huggingface.co/vllm-sr/d3-lite) as single files for ComfyUI. The weights are unchanged except for the format.
|
| 14 |
+
|
| 15 |
+
| File | Format |
|
| 16 |
+
| --- | --- |
|
| 17 |
+
| `d3_flash_bf16` / `d3_lite_bf16` | bf16, as released |
|
| 18 |
+
| `d3_flash_int8_convrot` / `d3_lite_int8_convrot` | int8 ConvRot |
|
| 19 |
+
|
| 20 |
+
The bf16 files only rename the keys to ComfyUI's Qwen3.5 layout (`model.` prefix). In the int8 files, only the language-model linear layers are quantized, in ComfyUI's `comfy_quant` format (ConvRot group 256 with per-row scales). Embeddings, norms and the vision encoder stay bf16.
|
| 21 |
+
|
| 22 |
+
Put the files in `ComfyUI/models/text_encoders` and use them with [ComfyUI-TypedDecision](https://github.com/nomadoor/ComfyUI-TypedDecision). The readout (`decision_config.json`, `readout.safetensors`) comes from the original repositories.
|