Views
No views yet
1merge_method: della_linear
2base_model: CultriX/Qwen2.5-14B-Wernickev3
3dtype: bfloat16
4parameters:
5 epsilon: 0.015 # Fine-grain scaling for precision.
6 lambda: 1.6 # Strong emphasis on top-performing models.
7 normalize: true # Stable parameter integration across models.
8adaptive_merge_parameters:
9 task_weights:
10 tinyArc: 1.75 # Logical reasoning.
11 tinyHellaswag: 1.65 # Contextual predictions.
12 tinyMMLU: 1.8 # Domain knowledge.
13 tinyTruthfulQA: 2.0 # Prioritize truthful reasoning.
14 tinyTruthfulQA_mc1: 1.85
15 tinyWinogrande: 1.9 # Advanced reasoning and predictions.
16 IFEval: 2.1 # Instruction-following and multitasking.
17 BBH: 1.9 # Complex reasoning.
18 MATH: 2.3 # Mathematical reasoning.
19 GPQA: 2.0 # Factual QA.
20 MUSR: 2.1 # Multi-step reasoning.
21 MMLU-PRO: 1.95 # Domain multitask performance.
22 smoothing_factor: 0.1 # Smooth blending across benchmarks.
23gradient_clipping:
24 CultriX/Qwen2.5-14B-Wernickev3: 0.9
25 djuna/Q2.5-Veltha-14B-0.5: 0.92
26 qingy2019/Qwen2.5-Math-14B-Instruct: 0.94
27 CultriX/SeQwence-14Bv1: 0.88
28 allknowingroger/QwenSlerp6-14B: 0.87
29 CultriX/Qwenfinity-2.5-14B: 0.85
30models:
31 - model: CultriX/Qwen2.5-14B-Wernickev3
32 parameters:
33 weight: 0.25
34 density: 0.72
35 - model: djuna/Q2.5-Veltha-14B-0.5
36 parameters:
37 weight: 0.22
38 density: 0.75
39 - model: qingy2019/Qwen2.5-Math-14B-Instruct
40 parameters:
41 weight: 0.18
42 density: 0.74
43 - model: CultriX/SeQwence-14Bv1
44 parameters:
45 weight: 0.15
46 density: 0.7
47 - model: allknowingroger/QwenSlerp6-14B
48 parameters:
49 weight: 0.12
50 density: 0.68
51 - model: CultriX/Qwenfinity-2.5-14B
52 parameters:
53 weight: 0.15
54 density: 0.7
55tokenizer_source: CultriX/Qwen2.5-14B-Wernickev3
56