aashish1904 commited on
Commit
f39766c
1 Parent(s): 5c771aa

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +133 -0
README.md ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+
4
+ base_model:
5
+ - Locutusque/Apollo-0.4-Llama-3.1-8B
6
+ - maldv/badger-writer-llama-3-8b
7
+ - ArliAI/Llama-3.1-8B-ArliAI-Formax-v1.0
8
+ - arcee-ai/Llama-3.1-SuperNova-Lite
9
+ - v000000/L3-8B-Poppy-Moonfall-C
10
+ - Casual-Autopsy/Jamet-L3-Stheno-BlackOasis-8B
11
+ - SicariusSicariiStuff/Dusk_Rainbow
12
+ - ResplendentAI/Rawr_Llama3_8B
13
+ - ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
14
+ - maximalists/BRAG-Llama-3.1-8b-v0.1
15
+ - tohur/natsumura-storytelling-rp-1.0-llama-3.1-8b
16
+ library_name: transformers
17
+ tags:
18
+ - mergekit
19
+ - merge
20
+ - roleplay
21
+ - RP
22
+ - storytelling
23
+ license: cc-by-nc-4.0
24
+
25
+ ---
26
+
27
+ [![QuantFactory Banner](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeiuCm7c8lEwEJuRey9kiVZsRn2W-b4pWlu3-X534V3YmVuVc2ZL-NXg2RkzSOOS2JXGHutDuyyNAUtdJI65jGTo8jT9Y99tMi4H4MqL44Uc5QKG77B0d6-JfIkZHFaUA71-RtjyYZWVIhqsNZcx8-OMaA?key=xt3VSDoCbmTY7o-cwwOFwQ)](https://hf.co/QuantFactory)
28
+
29
+
30
+ # QuantFactory/L3.1-Aglow-Vulca-v0.1-8B-GGUF
31
+ This is quantized version of [kromeurus/L3.1-Aglow-Vulca-v0.1-8B](https://huggingface.co/kromeurus/L3.1-Aglow-Vulca-v0.1-8B) created using llama.cpp
32
+
33
+ # Original Model Card
34
+
35
+ ![image/png](https://cdn-uploads.huggingface.co/production/uploads/667eea5cdebd46a5ec4dcc3d/5zwye-BvUG51XaSHcW-vG.png)
36
+
37
+ ...I'm tired.
38
+
39
+ Behold: Aglow Vulca. Had a Theseus Paradox of whether to keep the original name since everything save three models were replaced with another model, but it still has a spirit of Ablaze Vulca so I just changed the preceding adjective.
40
+
41
+ This took over a month, way too much money, and half of what was remaining of my sanity. If I could verbalize what the hell I went through trying to get this model to work, this repo would be 32k tokens long kekw. After figuring out how fast v0.1 could crack, I'd gotten to work on a v0.2 to at least smooth out the problems. Simple right?
42
+
43
+ No. It was not. But, much pain and suffering later, I've come out with a beast of an 8B merge that can handle almost anything thrown at it.
44
+
45
+ I'd like to give a special thanks to those in the BackyardAI discord for helping me test (especially one person, you know who you are) and watch me go down an insane downward spiral. They made the image above and helped troubleshoot versions until the final model was created. This merge would have taken much longer and the final version would be poorer without them. I'm the most active in that server so if you have questions, please join and say hi.
46
+
47
+ For **Quants**, look to your right at the model tree next to 'Quantizations'.
48
+
49
+ ### Model Details & Recommended Settings
50
+
51
+ This is a story telling first model that is proficient in narrative driven RP. Does best with straightforward instructions. Any wishy-washy language will confuse it. As per usual with any of my models with Formax in it, it's pretty sensitive to instructs so choose your words wisely.
52
+
53
+ Once going though, it's able to generate detailed and human-ish outputs with lots of personality depending on the information given. Has a habit of matching the style and format of the input; style, spacing, grammar, etc. Can interweave details from the character persona, chat history, and user persona (if there is one) to create unique interactions and plot points. Leans more or less positive naturally but can be flipped if prompted correctly.
54
+
55
+ Being a Llama 3.1 model, it's still subject to the normal pros and cons of L3/L3.1 but I'd like to think I tamed some of it. Keep the temp on the lower end since there is a low chance it might freak out. If it does, swipe/regen the chat or delete the afflicted output and try again.
56
+
57
+ Rec. Settings:
58
+ ```
59
+ Template: Llama 3
60
+ Token Count: 128k Max
61
+ Temperature: 1.2
62
+ Min P: 0.1
63
+ Repeat Penalty: 1.05
64
+ Repeat Penalty Tokens: 256
65
+ ```
66
+
67
+ ### Merge Theory
68
+
69
+ Where to begin. The general though process was still the roughly same as the Ablaze, making one very smart model and another more creative focused model. This time, I merged Formax and RPmax in separately instead of doing one merge since they have different focuses.
70
+
71
+ 'Apollobulk' is the smarts, having the storytelling capabilities from badger writer, instruct following from Formax (duh) and the smarts of Super Nova. Apollo 0.4 was use as an RP temper to keep the overall model aligned with RP. Apollo 2.0 wasn't used as it skewed the merge too far towards inconsistent narratives.
72
+
73
+ 'Reshape' is the creative end, taking some inspo from the Ablaze's creative center. First created 'Darkened' as the main influence over the final writing style of Aglow. Poppy Moonfall C had the personality I was looking for but the smarts (though not important was still necessary) so the other three were added to round out it's overall capabilities while being very creative. Plopping that atop RPmax (For excellent unique RP interactions), BRAG (serious recall), and Natsumura (a great Storytelling/RP base) and model stock it, you get a really solid model on its own.
74
+
75
+ Slap the two components together in a simple gradient dare_linear merge and boom; this unit of an 8B model. As of writing and releasing this model, mergekit is fucked for me (one of its dependencies has broken L3 merging) so I can't test any other methods atm. If there is a better final merge method, I'll be uploading a v0.2 once the bug is fixed.
76
+
77
+ This time around, everything was done with [DavidAU]()'s High Quality method, merging with float32 at all steps. Made a significant difference in nuanced understanding of text.
78
+
79
+ ### Config
80
+
81
+ ```yaml
82
+ models:
83
+ - model: Locutusque/Apollo-0.4-Llama-3.1-8B
84
+ - model: maldv/badger-writer-llama-3-8b
85
+ - model: ArliAI/Llama-3.1-8B-ArliAI-Formax-v1.0
86
+ base_model: arcee-ai/Llama-3.1-SuperNova-Lite
87
+ parameters:
88
+ int8_mask: true
89
+ merge_method: model_stock
90
+ dtype: float32
91
+ tokenizer_source: base
92
+ name: apollobulk
93
+ ---
94
+ models:
95
+ - model: v000000/L3-8B-Poppy-Moonfall-C
96
+ - model: Casual-Autopsy/Jamet-L3-Stheno-BlackOasis-8B
97
+ - model: SicariusSicariiStuff/Dusk_Rainbow
98
+ base_model: ResplendentAI/Rawr_Llama3_8B
99
+ parameters:
100
+ int8_mask: true
101
+ merge_method: model_stock
102
+ dtype: float32
103
+ tokenizer_source: base
104
+ name: darkened
105
+ ---
106
+ models:
107
+ - model: darkened
108
+ - model: ArliAI/Llama-3.1-8B-ArliAI-RPMax-v1.1
109
+ - model: maximalists/BRAG-Llama-3.1-8b-v0.1
110
+ base_model: tohur/natsumura-storytelling-rp-1.0-llama-3.1-8b
111
+ parameters:
112
+ int8_mask: true
113
+ merge_method: model_stock
114
+ dtype: float32
115
+ tokenizer_source: base
116
+ name: reshape
117
+ ---
118
+ models:
119
+ - model: reshape
120
+ parameters:
121
+ weight: [0.1, 0.9]
122
+ - model: apollobulk
123
+ parameters:
124
+ weight: [0.9, 0.1]
125
+ base_model: reshape
126
+ tokenizer_source: base
127
+ parameters:
128
+ normalize: false
129
+ int8_mask: true
130
+ merge_method: dare_linear
131
+ dtype: float32
132
+ name: vulca
133
+ ```