Nohobby/Q2.5-Qwetiapin-32B

Qwetiapin

There's no 'I' in 'brain damage'

Overview

An attempt to make QwentileSwap write even better by merging it with RP-Ink. And DeepSeek, because why not. However, I screwed up the first merge step by accidentally setting an extremely high epsilon value. Step2 wasn't planned, but due to a wonky tensor size mismatch error, I couldn't merge Step1 into QwentileSwap using sce, so I just threw in some random model. And that did, in fact, solve the issue.

The result? Well, it's usable, I guess. The slop is reduced, more details are brought up, but said details sometimes get messed up. It's fixed by a few swipes and there's a chance that it's caused by my sampler settings, but uhh I'll just leave them as they are.

Prompt format: ChatML

Settings: This kinda works but I'm weird

Quants

Static | Imatrix

Merge Details

Merging Steps

Step1

dtype: bfloat16
tokenizer_source: base
merge_method: della_linear
parameters:
  density: 0.5
  epsilon: 0.4 #was supposed to be 0.04 
  lambda: 1.1
base_model: allura-org/Qwen2.5-32b-RP-Ink
models:
  - model: deepseek-ai/DeepSeek-R1-Distill-Qwen-32B
    parameters:
      weight:
        - filter: v_proj
          value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0]
        - filter: o_proj
          value: [1, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1]
        - filter: up_proj
          value: [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
        - filter: gate_proj
          value: [0, 0, 1, 1, 1, 1, 1, 1, 1, 0, 0]
        - filter: down_proj
          value: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
        - value: 0
  - model: allura-org/Qwen2.5-32b-RP-Ink
    parameters:
      weight:
        - filter: v_proj
          value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1]
        - filter: o_proj
          value: [0, 1, 0, 1, 1, 1, 1, 1, 0, 0, 0]
        - filter: up_proj
          value: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
        - filter: gate_proj
          value: [1, 1, 0, 0, 0, 0, 0, 0, 0, 1, 1]
        - filter: down_proj
          value: [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1]
        - value: 1

Step2

models:
  - model: Aryanne/QwentileSwap
    parameters:
      weight: [1.0, 0.9, 0.8, 0.9, 1.0]
  - model: Daemontatox/Cogito-Ultima
    parameters:
      weight: [0, 0.1, 0.2, 0.1, 0]
merge_method: nuslerp
parameters:
  nuslerp_row_wise: true
dtype: bfloat16
tokenizer_source: base

Step3

models:
  - model: Step2
  - model: Step1
merge_method: sce
base_model: Step2
parameters:
  select_topk:
    - value: [0.3, 0.35, 0.4, 0.35, 0.2]
dtype: bfloat16