win10/SVD-Qwen3-Coder-Next-Thinking

Qwen-3-next-coderrr

This is a merge of pre-trained language models created using mergekit.

Merge Details

Merge Method

This model was merged using the Geometric Manifold Merge (PRISM) merge method using F:\Huihui-Qwen3-Coder-Next-abliterated as a base.

Models Merged

The following models were included in the merge:

• F:\Huihui-Qwen3-Next-80B-A3B-Thinking-abliterated

Configuration

The following YAML configuration was used to produce this model:

# PRISM: Geometric Manifold Merge
# Projected Riemannian Interpolation for Structure-aware Merging
#
# This method decomposes weight tensors via SVD and merges each component
# on its natural Riemannian manifold:
#   - Singular vectors: SLERP on the Stiefel manifold
#   - Singular values: Geometric interpolation on R_+
#   - Normalization weights: Log-Euclidean (geometric mean)
#
# Parameters:
#   svd_frac (float, 0-1): Fraction of min(m,n) for SVD rank (default: 0.25)
#   max_svd_rank (int): Maximum SVD rank cap (default: 512)
#   weight (per-model): Relative weight of each model

models:
  - model: F:\Huihui-Qwen3-Coder-Next-abliterated
    parameters:
      weight: 0.5
  - model: F:\Huihui-Qwen3-Next-80B-A3B-Thinking-abliterated
    parameters:
      weight: 0.5

merge_method: geometric
base_model: F:\Huihui-Qwen3-Coder-Next-abliterated
parameters:
  svd_frac: 0.25
  max_svd_rank: 2048
dtype: bfloat16
tokenizer:
  source: "base"