Update README.md
Browse files
README.md
CHANGED
@@ -1,134 +1,167 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
<!DOCTYPE html>
|
2 |
<html>
|
3 |
<head>
|
4 |
<style>
|
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 |
</style>
|
126 |
</head>
|
127 |
<body>
|
128 |
<div class="header">
|
129 |
<h1>Rei-12B</h1>
|
130 |
-
<p class="tagline">
|
131 |
-
<img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/nqMkoIsmScaTFHCFirGsc.png" alt="Rei Model" class="model-img">
|
132 |
</div>
|
133 |
<div class="card">
|
134 |
<h2>✨ Overview</h2>
|
@@ -164,18 +197,19 @@ Can I ask a question?<|im_end|>
|
|
164 |
<h2>⚙️ Training</h2>
|
165 |
<h3>Hparams</h3>
|
166 |
<ul>
|
167 |
-
<li>For Hparams for this model we experimented with
|
168 |
-
<li>Calculated via checking the model arch distribution's we divised 3 different values, knowing the weight
|
|
|
169 |
<li>If you would consult the graph, you'd notice a few things, First and foremost is that setting gradient-clip too high can be deterimental to the model as the logs and testing show that the model was overfit, Meanwhile setting it too low can also cause problems as the 1e-4 run appears to be underfit, The best one was by far the 0.001 clip which resulted in non-overfit, non-underfit model. </li>
|
170 |
</ul>
|
171 |
<h3>Configuration</h3>
|
172 |
<div class="details">
|
173 |
<details>
|
174 |
<summary>View Axolotl Config</summary>
|
175 |
-
<p>
|
176 |
</details>
|
177 |
</div>
|
178 |
-
<p>The model was trained for 2 epochs on
|
179 |
<div class="badge-container">
|
180 |
<a href="https://github.com/OpenAccess-AI-Collective/axolotl">
|
181 |
<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" class="badge">
|
@@ -183,11 +217,13 @@ Can I ask a question?<|im_end|>
|
|
183 |
</div>
|
184 |
</div>
|
185 |
<div class="card">
|
186 |
-
<h2>⚠️
|
187 |
-
<p><em>
|
|
|
|
|
188 |
</div>
|
189 |
<div class="footer">
|
190 |
-
<p>Rei-12B |
|
191 |
</div>
|
192 |
</body>
|
193 |
</html>
|
|
|
1 |
+
---
|
2 |
+
datasets:
|
3 |
+
- PocketDoc/Dans-Personamaxx-Logs
|
4 |
+
- anthracite-org/kalo-opus-instruct-22k-no-refusal
|
5 |
+
- lodrick-the-lafted/kalo-opus-instruct-3k-filtered
|
6 |
+
- anthracite-org/nopm_claude_writing_fixed
|
7 |
+
- anthracite-org/kalo_opus_misc_240827
|
8 |
+
- anthracite-org/kalo_misc_part2
|
9 |
+
- NewEden/Claude-Instruct-5K
|
10 |
+
- NewEden/Claude-Instruct-2.7K
|
11 |
+
base_model:
|
12 |
+
- NewEden/MistralAI-Nemo-Instruct-ChatML
|
13 |
+
pipeline_tag: text-generation
|
14 |
+
library_name: transformers
|
15 |
+
language:
|
16 |
+
- en
|
17 |
+
tags:
|
18 |
+
- roleplay
|
19 |
+
- finetune
|
20 |
+
- mistral
|
21 |
+
- magnum
|
22 |
+
- claude
|
23 |
+
- story-writing
|
24 |
+
---
|
25 |
<!DOCTYPE html>
|
26 |
<html>
|
27 |
<head>
|
28 |
<style>
|
29 |
+
:root {
|
30 |
+
--primary: #6e48aa;
|
31 |
+
--secondary: #9d50bb;
|
32 |
+
--accent: #4776e6;
|
33 |
+
--bg: #1a1a2e;
|
34 |
+
--card-bg: #2a2a3a;
|
35 |
+
--text: #ffffff;
|
36 |
+
--highlight: #ff7e5f;
|
37 |
+
}
|
38 |
+
body {
|
39 |
+
font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
|
40 |
+
background-color: var(--bg);
|
41 |
+
color: var(--text);
|
42 |
+
line-height: 1.6;
|
43 |
+
max-width: 900px;
|
44 |
+
margin: 0 auto;
|
45 |
+
padding: 20px;
|
46 |
+
}
|
47 |
+
.header {
|
48 |
+
text-align: center;
|
49 |
+
margin-bottom: 30px;
|
50 |
+
border-bottom: 2px solid var(--secondary);
|
51 |
+
padding-bottom: 20px;
|
52 |
+
}
|
53 |
+
h1 {
|
54 |
+
color: var(--highlight);
|
55 |
+
font-size: 2.5em;
|
56 |
+
margin-bottom: 10px;
|
57 |
+
background: linear-gradient(90deg, var(--highlight), var(--secondary));
|
58 |
+
-webkit-background-clip: text;
|
59 |
+
background-clip: text;
|
60 |
+
color: transparent;
|
61 |
+
}
|
62 |
+
.tagline {
|
63 |
+
font-style: italic;
|
64 |
+
color: var(--secondary);
|
65 |
+
}
|
66 |
+
.model-img {
|
67 |
+
border-radius: 10px;
|
68 |
+
border: 3px solid var(--accent);
|
69 |
+
box-shadow: 0 0 20px rgba(157, 80, 187, 0.3);
|
70 |
+
max-width: 100%;
|
71 |
+
height: auto;
|
72 |
+
}
|
73 |
+
.card {
|
74 |
+
background-color: var(--card-bg);
|
75 |
+
border-radius: 8px;
|
76 |
+
padding: 20px;
|
77 |
+
margin: 20px 0;
|
78 |
+
box-shadow: 0 4px 15px rgba(157, 80, 187, 0.4);
|
79 |
+
border-left: 4px solid var(--accent);
|
80 |
+
color: var(--text);
|
81 |
+
}
|
82 |
+
h2 {
|
83 |
+
color: var(--highlight);
|
84 |
+
border-bottom: 1px solid var(--secondary);
|
85 |
+
padding-bottom: 5px;
|
86 |
+
}
|
87 |
+
h3 {
|
88 |
+
color: var(--accent);
|
89 |
+
}
|
90 |
+
code {
|
91 |
+
background-color: rgba(0, 50, 0, 0.5);
|
92 |
+
padding: 2px 5px;
|
93 |
+
border-radius: 3px;
|
94 |
+
font-family: 'Courier New', Courier, monospace;
|
95 |
+
color: #00ff00;
|
96 |
+
}
|
97 |
+
pre {
|
98 |
+
background-color: #0a1a0a;
|
99 |
+
padding: 15px;
|
100 |
+
border-radius: 5px;
|
101 |
+
overflow-x: auto;
|
102 |
+
border-left: 3px solid #00ff00;
|
103 |
+
color: #00ff00;
|
104 |
+
font-family: 'Courier New', Courier, monospace;
|
105 |
+
}
|
106 |
+
.badge-container {
|
107 |
+
display: flex;
|
108 |
+
justify-content: center;
|
109 |
+
margin: 20px 0;
|
110 |
+
}
|
111 |
+
.badge {
|
112 |
+
transition: transform 0.3s;
|
113 |
+
}
|
114 |
+
.badge:hover {
|
115 |
+
transform: scale(1.05);
|
116 |
+
}
|
117 |
+
.details {
|
118 |
+
background-color: #0a1a0a;
|
119 |
+
border-radius: 5px;
|
120 |
+
padding: 10px;
|
121 |
+
margin: 10px 0;
|
122 |
+
box-shadow: 0 4px 15px rgba(0, 255, 0, 0.15);
|
123 |
+
color: #00ff00;
|
124 |
+
font-family: 'Courier New', Courier, monospace;
|
125 |
+
border: 1px solid #00aa00;
|
126 |
+
}
|
127 |
+
.details summary {
|
128 |
+
cursor: pointer;
|
129 |
+
font-weight: bold;
|
130 |
+
color: #00ff00;
|
131 |
+
}
|
132 |
+
.quant-links {
|
133 |
+
display: flex;
|
134 |
+
gap: 20px;
|
135 |
+
justify-content: center;
|
136 |
+
flex-wrap: wrap;
|
137 |
+
}
|
138 |
+
.quant-link {
|
139 |
+
background: linear-gradient(135deg, var(--primary), var(--secondary));
|
140 |
+
color: #ff0000;
|
141 |
+
padding: 10px 20px;
|
142 |
+
border-radius: 5px;
|
143 |
+
text-decoration: none;
|
144 |
+
font-weight: bold;
|
145 |
+
transition: transform 0.3s, box-shadow 0.3s;
|
146 |
+
border: 3px solid #ff0000;
|
147 |
+
}
|
148 |
+
.quant-link:hover {
|
149 |
+
transform: translateY(-3px);
|
150 |
+
box-shadow: 0 5px 15px rgba(157, 80, 187, 0.4);
|
151 |
+
}
|
152 |
+
.footer {
|
153 |
+
text-align: center;
|
154 |
+
margin-top: 40px;
|
155 |
+
font-size: 0.9em;
|
156 |
+
color: var(--secondary);
|
157 |
+
}
|
158 |
</style>
|
159 |
</head>
|
160 |
<body>
|
161 |
<div class="header">
|
162 |
<h1>Rei-12B</h1>
|
163 |
+
<p class="tagline">Another prototype Magnum...</p>
|
164 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/nqMkoIsmScaTFHCFirGsc.png" alt="Rei Model" class="model-img" width="500px">
|
165 |
</div>
|
166 |
<div class="card">
|
167 |
<h2>✨ Overview</h2>
|
|
|
197 |
<h2>⚙️ Training</h2>
|
198 |
<h3>Hparams</h3>
|
199 |
<ul>
|
200 |
+
<li>For Hparams for this model we experimented with Grad clipping otherwise known as max_grad_norm</li>
|
201 |
+
<li>Calculated via checking the model arch distribution's we divised 3 different values, knowing the weight distribution for Mistral is 0.1. </li>
|
202 |
+
<img src="https://cdn-uploads.huggingface.co/production/uploads/66c26b6fb01b19d8c3c2467b/2AWCq2cNBbq90VnqkIx79.png" width="500px" />
|
203 |
<li>If you would consult the graph, you'd notice a few things, First and foremost is that setting gradient-clip too high can be deterimental to the model as the logs and testing show that the model was overfit, Meanwhile setting it too low can also cause problems as the 1e-4 run appears to be underfit, The best one was by far the 0.001 clip which resulted in non-overfit, non-underfit model. </li>
|
204 |
</ul>
|
205 |
<h3>Configuration</h3>
|
206 |
<div class="details">
|
207 |
<details>
|
208 |
<summary>View Axolotl Config</summary>
|
209 |
+
<p>https://wandb.ai/new-eden/Rei-V2/artifacts/axolotl-config/config-7hvbucx9/v0/files/axolotl_config_pw8f0c6u.yml</p>
|
210 |
</details>
|
211 |
</div>
|
212 |
+
<p>The model was trained for 2 epochs on 8x <a href="https://www.nvidia.com/en-us/data-center/h200/" style="color: var(--accent);">NVIDIA H200s</a> GPUs generously provided by @Kalomaze</p>
|
213 |
<div class="badge-container">
|
214 |
<a href="https://github.com/OpenAccess-AI-Collective/axolotl">
|
215 |
<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" class="badge">
|
|
|
217 |
</div>
|
218 |
</div>
|
219 |
<div class="card">
|
220 |
+
<h2>⚠️ Credits</h2>
|
221 |
+
<p><em>
|
222 |
+
I'd like to thank, Ruka/Sama twinkman | LucyKnada | Kubernetes Bad | PocketDoc | Tav | Trappu | And the rest of Anthracite/Pygmalion for testing, feedback, and support.
|
223 |
+
</em></p>
|
224 |
</div>
|
225 |
<div class="footer">
|
226 |
+
<p>Rei-12B | V2</p>
|
227 |
</div>
|
228 |
</body>
|
229 |
</html>
|