what drop rate to use in DareMerge tool for merging SDXL checkpoints
- User
what drop rate to use in DareMerge tool for merging SDXL checkpoints
Assistant
For merging SDXL checkpoints with DareMerge, the recommended drop rate typically ranges between 0.2 to 0.3 (20-30%). This provides a good balance when combining models without losing too many features from either checkpoint.
The optimal rate depends on your specific goals:
- Lower rates (0.1-0.2) preserve more information from both models but might cause conflicts
- Higher rates (0.3-0.4) reduce conflicts but might lose unique features
- For more experimental merges, some users go as high as 0.5
Start with 0.25 (25%) as a baseline and adjust based on your results.
- User
what magnitude mask threshold to use
Assistant
For the magnitude mask threshold in DareMerge when merging SDXL checkpoints, a value between 0.3 and 0.5 is generally recommended.
More specifically:
- 0.3-0.4 works well for most standard merges
- 0.5 is good when you want more aggressive filtering
- Below 0.3 tends to introduce more noise but preserves more details
The threshold determines which weights are significant enough to include in the merge. Start with 0.4 and adjust based on your results.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.