update changelog

Signed-off-by: Vladimir Mandic <mandic00@live.com>
This commit is contained in:
Vladimir Mandic
2026-08-26 16:13:12 +02:00
parent 4be7d3c761
commit b14ce1032b
2 changed files with 28 additions and 2 deletions
+12 -1
View File
@@ -1,6 +1,17 @@
# Change Log for SD.Next
## Highlights for 2026-08-26
## Update for 2026-08-27
- **LoRA**
- new calibration engine that allows lora to be applied with far smaller error when dealing with highly quantized models
- new apply engine that allows lora to be applied much faster
- *note*: calibration data is stored once calculated so it can be reused for future runs, location is `models/calibration` folder
- **Internal**
- attention mechanisms decision tree and apply method refactor
## Update for 2026-08-26
### Highlights for 2026-08-26
Time for a new release, *this is a large one*!
Main focus is improving video workflows which also brings full support for new [MiniMax H3](https://vladmandic.github.io/sdnext-docs/MiniMax) and [LTXVideo-2.5](https://vladmandic.github.io/sdnext-docs/LTX)
+16 -1
View File
@@ -21,6 +21,7 @@ compiled (hooks would break the graph) and everything is gated by the
"""
import os
from typing import Optional, TypedDict
import torch
@@ -28,9 +29,23 @@ from modules import paths, shared, script_callbacks
from modules.logger import log
class CaptureRecord(TypedDict):
m: torch.nn.Module
ss: Optional[torch.Tensor]
n: int
done: bool
class CaptureState(TypedDict):
model: Optional[str]
recs: dict[str, CaptureRecord]
handles: list[torch.utils.hooks.RemovableHandle]
complete: bool
TOKENS_DONE = 65536
calib_root = os.path.join(paths.models_path, 'calibration')
capture = {'model': None, 'recs': {}, 'handles': [], 'complete': False}
capture: CaptureState = {'model': None, 'recs': {}, 'handles': [], 'complete': False}
def enabled():