From b724cd7c5739399dedd5a8aee94165eab7af0dc0 Mon Sep 17 00:00:00 2001 From: Disty0 Date: Tue, 27 May 2025 21:21:42 +0300 Subject: [PATCH] Update changelog --- CHANGELOG.md | 3 +++ modules/shared.py | 2 +- 2 files changed, 4 insertions(+), 1 deletion(-) diff --git a/CHANGELOG.md b/CHANGELOG.md index 81486e925..07f916d52 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -9,10 +9,13 @@ - `INT8` -> `uint8` - `INT4_SYM` -> `int4` - `INT4` -> `uint4` + - Add `float8_e4m3fn` and `float8_e5m2` support - Set the default quant mode to `pre` - Use per token input quant with int8 matmul - Implement better layer hijacks - Fix Conv quant + - Fix lora weight change + - Use device_map = gpu - **IPEX** - Disabe Dynamic Attention by default on PyTorch 2.7 - Remove GradScaler hijack and use torch.amp.GradScaler instead diff --git a/modules/shared.py b/modules/shared.py index 8fed9c635..cdb71534d 100644 --- a/modules/shared.py +++ b/modules/shared.py @@ -525,7 +525,7 @@ options_templates.update(options_section(('quantization', "Quantization Settings "sdnq_quantize_sep": OptionInfo("

SDNQ: SDNext Quantization

", "", gr.HTML), "sdnq_quantize_weights": OptionInfo([], "Quantization enabled", gr.CheckboxGroup, {"choices": ["Model", "Transformer", "VAE", "TE", "Video", "LLM", "ControlNet"], "visible": native}), "sdnq_quantize_mode": OptionInfo("pre", "Quantization mode", gr.Dropdown, {"choices": ['pre', 'post'], "visible": native}), - "sdnq_quantize_weights_mode": OptionInfo("int8", "Quantization type", gr.Dropdown, {"choices": ['int8', 'uint4', "float8_e4m3fn", "float8_e5m2", 'uint8', 'int4'], "visible": native}), + "sdnq_quantize_weights_mode": OptionInfo("int8", "Quantization type", gr.Dropdown, {"choices": ['int8', 'uint4', "float8_e4m3fn", 'uint8', 'int4', "float8_e5m2"], "visible": native}), "sdnq_quantize_weights_group_size": OptionInfo(0, "Group size", gr.Slider, {"minimum": -1, "maximum": 4096, "step": 1, "visible": native}), "sdnq_quantize_conv_layers": OptionInfo(False, "Quantize the convolutional layers", gr.Checkbox, {"visible": native}), "sdnq_decompress_fp32": OptionInfo(False, "Decompress using full precision", gr.Checkbox, {"visible": native}),