File size: 5,319 Bytes
b8db39d
 
 
bf37f73
011d917
 
 
 
8086464
45c246f
011d917
 
 
 
 
 
 
27a7c2d
011d917
27a7c2d
011d917
27a7c2d
011d917
27a7c2d
011d917
33f8801
011d917
33f8801
bfdf6df
33f8801
 
 
 
5d05d2d
 
e5602d3
 
 
 
33f8801
d303730
011d917
d303730
bfdf6df
d303730
011d917
 
 
 
 
724f369
e8b6c0e
d303730
011d917
 
d303730
 
011d917
 
d303730
 
011d917
 
d303730
 
 
 
 
e8b6c0e
011d917
724f369
 
ab0590a
724f369
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
---
license: cc-by-nc-4.0
---
<p align="center"><img src="https://i.ibb.co/pbpJHpk/iambe-sml.png"/><font size="6"> <b>Iambe-20b-DARE</b> </font></p>

## Description and Role

Named after a charming daughter of Echo and Pan in Greek myth, Iambe-20b-DARE is a [DARE](https://github.com/yule-BUAA/MergeLM) merge building on my recent experiments.

Iambe is intended to have the best realistically possible understanding of anatomy and of a scene's state for a 20b merge, while remaining personable and authentic in "voice". 

## Prompting and Context

Iambe-20b-DARE uses Alpaca formatting, and has an effective context size of 4096 tokens. This model is uncensored, and the output/deployment of this model is the responsibility of the user.

## Method and Hypothesis

Based on my extended vanilla model [BigLlama](https://huggingface.co/athirdpath/BigLlama-20b), this adds elements of:

- [NeverSleep/Noromaid-20b-v0.1.1](https://huggingface.co/NeverSleep/Noromaid-20b-v0.1.1) - Addded to adapt the excellent writing and "soul" that come from the datasets backing this model.
 
- [athirdpath/Eileithyia-20b](https://huggingface.co/athirdpath/Eileithyia-20b) - Added at low weight and density to capture anatomical data and its relation to fiction without the model's other... quirks.
 
- [athirdpath/CleverGirl-20b-Blended](https://huggingface.co/athirdpath/CleverGirl-20b-Blended) - Added to capture CleverGirl's problem-solving abilities.

<p align="center"><font size="7"> <b>Examples</b></font>

<p align="center"><font size="3"> <i>(q5_k_m GGUF quant, Ooba textgen, Midnight Enigma preset for KISS reasons)</i></font>

<p align="center"><font size="5"> Reasoning and Recall - STEM</font><img src="https://i.ibb.co/YDkRD43/base.png"/>

<p align="center"><font size="5"> Reasoning and Recall - Social Science</font><img src="https://i.ibb.co/L0kc3pb/Screenshot-2023-11-29-024037.png"/>

<p align="center"><font size="5"> Ethics and Prompt Interpretation</font><img src="https://i.ibb.co/b5vH6jD/Screenshot-2023-11-29-040408.png"/>

<p align="center"><font size="5">Role-Playing, Complex Card</font>
  <p align="center"><font size="3">SillyTavern, Roleplay instruct preset, just Min_P 0.1 and Temp 1.2</font><img src="https://i.ibb.co/NrKNn2j/Screenshot-2023-11-29-051923.png"/>


<p align="center"><font size="6"><b><a href="https://i.ibb.co/m5G0ZVp/Screenshot-2023-11-29-004705.png">!!NSFW!! - Erotica Writing Example - !!NSFW!!</font></a></b></p>

## Testing and Conclusions

VERY Impressed so far, concrete data coming eventually.

Does still have some confusion (at q5_k_m), but has instantly become my daily driver.

## Recipe
merge_method: dare_ties

  - base_model: athirdpath/BigLlama-20b

  - model: NeverSleep/Noromaid-20b-v0.1.1
   
      weight: 0.41 / density: 0.50
    
  - model: athirdpath/Eileithyia-20b
   
      weight: 0.18 / density: 0.30
    
  - model: athirdpath/CleverGirl-20b-Blended
   
      weight: 0.41 / density: 0.50
    

int8_mask: true

dtype: bfloat16

## Gratitude
Thanks to brucethemoose for the recipe. Thanks to Undi95 and IkariDev at NeverSleep for Noromaid, as well as lots of inspiration. Thanks to Sao10K for [half of](https://huggingface.co/Sao10K/Mythical-Destroyer-V2-L2-13B) CleverGirl.

## Geneology Credits - Models and LoRAs, 3 levels deep

- athirdpath/BigLlama-20b

  - TheBloke/Llama-2-13B-fp16
  
  - iamshnoo/alpaca-2-13b-english

- NeverSleep/Noromaid-20b-v0.1.1
 
  - Aesir Private RP dataset

  - HuggingFaceH4/no_robots
 
- athirdpath/CleverGirl-20b-Blended

  - athirdpath/Orca-2-13b-Alpaca-Uncensored
 
    - microsoft/Orca-2-13b
   
    - athirdpath/Orca-2-13b-Alpaca-Uncensored-LORA
   
  - Sao10K/Mythical-Destroyer-V2-L2-13B

    - TheBloke/Llama-2-13B-fp16
   
    - Gryphe/MythoMax-L2-13b
      
    - totally-not-an-llm/PuddleJumper-13b
      
    - jondurbin/airoboros-l2-13b-2.1
      
    - rombodawg/LosslessMegaCoder-llama2-13b-mini
      
    - The-Face-Of-Goonery/Chronos-Beluga-v2-13bfp16
  
- athirdpath/Eileithyia-20b

  - athirdpath/Harmonia-20B

    - Undi95/Emerhyst-20B
 
      - Undi95/Emerald-13B

      - Undi95/Amethyst-13B
 
    - Undi95/MXLewd-L2-20B
  
      - Undi95/MLewd-L2-13B-v2-3

      - Undi95/ReMM-v2.1-L2-13B

      - Xwin-LM/Xwin-LM-13B-V0.1
 
    - Undi95/Lewd-Sydney-20B
 
      - Free_Sydney_V2_13b_HF
      
      - Undi95/Xwin-MLewd-13B-V0.2
       
      - lemonilia/LimaRP-Llama2-13B-v3-EXPERIMENT

    - athirdpath/Nethena-20b-Glued
 
      - NeverSleep/Nethena-20B
     
      - athirdpath/Nethena-20b-Glue-LORA
      
    - tavtav/Rose-20B
 
      - CalderaAI/13B-Thorns-l2
     
      - NeverSleep/Noromaid-13b-v0.1.0
 
    - Undi95/PsyMedRP-v1-20B
   
      - jondurbin/airoboros-l2-13b-3.0
     
      - ehartford/Samantha-1.11-13b
     
      - Xwin-LM/Xwin-LM-13B-V0.1
     
      - chaoyi-wu/MedLLaMA_13B
     
      - migtissera/Synthia-13B-v1.2
 
    - NeverSleep/Noromaid-20b-v0.1.1
   
      - Aesir Private RP dataset

      - HuggingFaceH4/no_robots
 
    - Undi95/U-Amethyst-20B
   
      - Xwin-LM/Xwin-LM-13B-V0.1
        
      - The-Face-Of-Goonery/Huginn-13b-FP16
        
      - zattio770/120-Days-of-LORA-v2-13B
        
      - lemonilia/LimaRP-Llama2-13B-v3-EXPERIMENT
        
      - Undi95/Unholy-v1-12L-13B
 
  - athirdpath/Eileithyia-20b-LORA
 
Thanks again to everyone involved.