Hugging Face's logo

CultriX
/
Qwen2.5-14B-Ultima

Text Generation
Transformers
Safetensors
qwen2
mergekit
Merge
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
Qwen2.5-14B-Ultima
File size: 5,047 Bytes 
fbcc099
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
 
---
base_model:
- sometimesanotion/Lamarck-14B-v0.7-rc4
- sthenno/tempesthenno-ppo-ckpt40
library_name: transformers
tags:
- mergekit
- merge

---
# merge

This is a merge of pre-trained language models created using [mergekit](https://github.com/cg123/mergekit).

## Merge Details
### Merge Method

This model was merged using the [SLERP](https://en.wikipedia.org/wiki/Slerp) merge method.

### Models Merged

The following models were included in the merge:
* [sometimesanotion/Lamarck-14B-v0.7-rc4](https://huggingface.co/sometimesanotion/Lamarck-14B-v0.7-rc4)
* [sthenno/tempesthenno-ppo-ckpt40](https://huggingface.co/sthenno/tempesthenno-ppo-ckpt40)

### Configuration

The following YAML configuration was used to produce this model:

```yaml
# =============================================================================
# SuperMerge-14B-Simple
#
# This configuration merges only two components:
#   - Base Model: Provides stable foundational features.
#       Model: sometimesanotion/Lamarck-14B-v0.7-rc4
#
#   - Reasoning Module: Drives enhanced mid-layer reasoning.
#       Model: sthenno/tempesthenno-ppo-ckpt40
#
# The merge is performed using slerp with a V-shaped interpolation curve.
# Weighting across each 8-layer slice is tuned to balance core feature
# preservation with advanced reasoning.
# =============================================================================
name: SuperMerge-14B-Simple
merge_method: slerp
base_model: sometimesanotion/Lamarck-14B-v0.7-rc4
tokenizer_source: base
dtype: float32
out_dtype: bfloat16
parameters:
  int8_mask: true         # Optimize memory usage.
  normalize: true         # Ensure weights are on a comparable scale.
  rescale: false          # No additional rescaling necessary.
  # Interpolation curve for 6 slices (48 layers total):
  # Maintains a V-shaped emphasis for mid-layer processing.
  t: [0.1, 0.35, 0.85, 0.85, 0.35, 0.1]
slices:
  # ---------------------------------------------------------------------------
  # Slice 1 (Layers 0-8):
  #   - Early layers: nearly pure base model with minimal PPO influence.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [0, 8]
        parameters:
          weight: 0.95
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [0, 8]
        parameters:
          weight: 0.05

  # ---------------------------------------------------------------------------
  # Slice 2 (Layers 8-16):
  #   - Blend base with stronger PPO contributions to boost reasoning.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [8, 16]
        parameters:
          weight: 0.4
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [8, 16]
        parameters:
          weight: 0.6

  # ---------------------------------------------------------------------------
  # Slice 3 (Layers 16-24):
  #   - Mid-layer: Prioritize advanced reasoning by increasing the PPO share.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [16, 24]
        parameters:
          weight: 0.3
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [16, 24]
        parameters:
          weight: 0.7

  # ---------------------------------------------------------------------------
  # Slice 4 (Layers 24-32):
  #   - Continue the focus on reasoning with PPO while still retaining base traits.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [24, 32]
        parameters:
          weight: 0.35
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [24, 32]
        parameters:
          weight: 0.65

  # ---------------------------------------------------------------------------
  # Slice 5 (Layers 32-40):
  #   - Re-stabilize the network with a stronger base model contribution.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [32, 40]
        parameters:
          weight: 0.6
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [32, 40]
        parameters:
          weight: 0.4

  # ---------------------------------------------------------------------------
  # Slice 6 (Layers 40-48):
  #   - Final output layers: Maintain fluency with the base model augmented by PPO.
  # ---------------------------------------------------------------------------
  - sources:
      - model: sometimesanotion/Lamarck-14B-v0.7-rc4
        layer_range: [40, 48]
        parameters:
          weight: 0.6
      - model: sthenno/tempesthenno-ppo-ckpt40
        layer_range: [40, 48]
        parameters:
          weight: 0.4

```