Hugging Face's logo

SvalTek
/
ColdBrew-Calypso

Safetensors
llama
Merge
mergekit
lazymergekit
SvalTek/ColdBrew-Aphid
Model card Files Community
ColdBrew-Calypso
File size: 3,008 Bytes 
ba23a1c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
---
base_model:
- SvalTek/ColdBrew-Aphid
tags:
- merge
- mergekit
- lazymergekit
- SvalTek/ColdBrew-Aphid
---

# ColdBrew-Calypso

ColdBrew-Calypso is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [SvalTek/ColdBrew-Aphid](https://huggingface.co./SvalTek/ColdBrew-Aphid)

## 🧩 Configuration

```yaml
const_tag:
  - &SFU 0.7071067812  # Upper scaling factor (1/sqrt(2))
  - &SFL 0.85          # Lower scaling factor
  - &SFMLP 0.8         # MLP scaling factor

attenuate-env-lower: &attenuated_env_lower
  parameters:
    scale:
      - filter: q_proj
        value: *SFL
      - filter: k_proj
        value: *SFL
      - filter: v_proj
        value: *SFL
      - filter: o_proj
        value: *SFL
      - filter: gate_proj
        value: *SFMLP
      - filter: up_proj
        value: *SFMLP
      - filter: down_proj
        value: *SFMLP

attenuate-env-upper: &attenuated_env_upper
  parameters:
    scale:
      - filter: q_proj
        value: *SFU
      - filter: k_proj
        value: *SFU
      - filter: v_proj
        value: *SFU
      - filter: o_proj
        value: *SFU
      - filter: gate_proj
        value: *SFMLP
      - filter: up_proj
        value: *SFMLP
      - filter: down_proj
        value: *SFMLP

slices:
  # Preserve input layers
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [0, 12]

  # Early Expansion (Duplicate Twice)
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [13, 15]
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [13, 15]
        <<: *attenuated_env_lower
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [13, 15]
        <<: *attenuated_env_lower

  # Late Expansion (Duplicate Twice)
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [15, 18]
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [15, 18]
        <<: *attenuated_env_upper
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [15, 18]
        <<: *attenuated_env_upper

  # Preserve output layers
  - sources:
      - model: SvalTek/ColdBrew-Aphid
        layer_range: [19, 27]

merge_method: passthrough
dtype: bfloat16
parameters:
  normalize: true
  int8_mask: true
```

## 💻 Usage

```python
!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "SvalTek/ColdBrew-Calypso"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```