Skip to content
TheChatArchiveThe Chat Archive
Claudeclaude-3-7-sonnet-20250219-thinking-32k· AnthropicHugging Face LMArena Human Preference 140K

what drop rate to use in DareMerge tool for merging SDXL checkpoints

From a Chatbot Arena battleThe same person sent this prompt to two anonymous models at once. Compare it with Qwenqwen3-235b-a22b-no-thinking’s answer.
  1. User

    what drop rate to use in DareMerge tool for merging SDXL checkpoints

  2. Claude 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.

  3. User

    what magnitude mask threshold to use

  4. Claude 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.

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.