A DARE-TIES merge combining the strengths of two fine-tuned Qwen 3.5 27B variants — one distilled from Opus-style reasoning, the other from GLM 5.1 — into a single unified model.

Source Models

Model	Role	Density	Weight
Qwopus3.5-27B-v3.5	Donor A	0.60	0.55
Qwen3.5-27B-GLM5.1-Distill-v1	Donor B	0.50	0.45
unsloth/Qwen3.5-27B	Base	—	—

Merge Configuration

yaml

models:
  - model: Jackrong/Qwopus3.5-27B-v3.5
    parameters:
      density: 0.6
      weight: 0.55
  - model: Jackrong/Qwen3.5-27B-GLM5.1-Distill-v1
    parameters:
      density: 0.5
      weight: 0.45
merge_method: dare_ties
base_model: unsloth/Qwen3.5-27B
parameters:
  normalize: true
  int8_mask: true
dtype: bfloat16

Quick Start (4-bit QLoRA-ready)

python

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
import torch

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)

model_id = "rodrigomt/Qwen-3.5-Opus-GLM-27B"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto",
)