Delta-Vector commited on
Commit
9723bc6
·
verified ·
1 Parent(s): 007189f

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +154 -422
README.md CHANGED
@@ -1,461 +1,193 @@
1
  <!DOCTYPE html>
2
- <html lang="en">
3
  <head>
4
- <meta charset="UTF-8">
5
- <meta name="viewport" content="width=device-width, initial-scale=1.0">
6
- <title>Queen's Model Citadel</title>
7
  <style>
8
- @import url('https://fonts.googleapis.com/css2?family=MedievalSharp&family=UnifrakturMaguntia&family=Cinzel+Decorative:wght@400;700;900&display=swap');
9
  :root {
10
- --dark-purple: #1a001a;
11
- --royal-purple: #3d0066;
12
- --violet-crown: #6600cc;
13
- --neon-amethyst: #9d00ff;
14
- --witch-glow: #cc33ff;
15
- --cursed-pink: #ff00ff;
16
- --obsidian: #0d0010;
17
- --blood-moon: #cc0066;
18
  }
19
  body {
20
- background: radial-gradient(ellipse at center, var(--dark-purple) 0%, var(--obsidian) 100%);
21
- color: #f0e6ff;
22
- font-family: 'Cinzel Decorative', cursive;
23
- margin: 0;
24
- padding: 0;
25
- overflow-x: hidden;
26
- text-shadow: 0 0 5px var(--witch-glow);
27
- animation: pulseBackground 30s infinite alternate;
28
- }
29
- @keyframes pulseBackground {
30
- 0% { background-color: var(--dark-purple); }
31
- 25% { background-color: #250025); }
32
- 50% { background-color: #300030); }
33
- 75% { background-color: #250025); }
34
- 100% { background-color: var(--dark-purple); }
35
- }
36
- .container {
37
  max-width: 900px;
38
  margin: 0 auto;
39
- padding: 2rem;
40
- border: 3px solid var(--violet-crown);
41
- border-radius: 0 0 20px 20px;
42
- box-shadow: 0 0 30px var(--neon-amethyst), inset 0 0 20px rgba(157, 0, 255, 0.3);
43
- background: linear-gradient(135deg, rgba(61, 0, 102, 0.2) 0%, rgba(26, 0, 26, 0.8) 100%);
44
- position: relative;
45
- overflow: hidden;
46
- }
47
- .container::before {
48
- content: "";
49
- position: absolute;
50
- top: 0;
51
- left: 0;
52
- right: 0;
53
- height: 5px;
54
- background: linear-gradient(90deg, var(--blood-moon), var(--cursed-pink), var(--witch-glow), var(--neon-amethyst), var(--violet-crown));
55
- animation: rainbowBorder 8s linear infinite;
56
  }
57
- @keyframes rainbowBorder {
58
- 0% { background-position: 0% 50%; }
59
- 100% { background-position: 100% 50%; }
60
- }
61
- h1 {
62
- font-family: 'UnifrakturMaguntia', cursive;
63
- color: var(--witch-glow);
64
  text-align: center;
65
- font-size: 4rem;
66
- margin: 1rem 0 2rem;
67
- text-shadow: 0 0 10px var(--cursed-pink), 0 0 20px var(--neon-amethyst);
68
- animation: titleGlow 3s ease-in-out infinite alternate;
69
- letter-spacing: 3px;
70
  }
71
- @keyframes titleGlow {
72
- from { text-shadow: 0 0 10px var(--cursed-pink), 0 0 20px var(--neon-amethyst); }
73
- to { text-shadow: 0 0 15px var(--blood-moon), 0 0 30px var(--witch-glow), 0 0 45px var(--violet-crown); }
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
74
  }
75
  h2 {
76
- font-family: 'MedievalSharp', cursive;
77
- color: var(--neon-amethyst);
78
- border-bottom: 2px solid var(--violet-crown);
79
- padding-bottom: 0.5rem;
80
- margin-top: 2.5rem;
81
- font-size: 2.2rem;
82
- text-shadow: 0 0 8px rgba(157, 0, 255, 0.7);
83
- position: relative;
84
- }
85
- h2::after {
86
- content: "";
87
- position: absolute;
88
- bottom: -2px;
89
- left: 0;
90
- width: 100%;
91
- height: 2px;
92
- background: linear-gradient(90deg, transparent, var(--witch-glow), transparent);
93
- animation: sectionUnderline 3s infinite;
94
- }
95
- @keyframes sectionUnderline {
96
- 0% { transform: scaleX(0); opacity: 0; }
97
- 50% { transform: scaleX(1); opacity: 1; }
98
- 100% { transform: scaleX(0); opacity: 0; }
99
  }
100
- .section {
101
- background: rgba(26, 0, 26, 0.5);
102
- border-left: 4px solid var(--blood-moon);
103
- padding: 1.5rem;
104
- margin: 1.5rem 0;
105
- border-radius: 0 10px 10px 0;
106
- box-shadow: 5px 5px 15px rgba(0, 0, 0, 0.3);
107
- transition: all 0.3s ease;
108
- position: relative;
109
- overflow: hidden;
110
  }
111
- .section:hover {
112
- transform: translateX(5px);
113
- box-shadow: 10px 10px 25px rgba(0, 0, 0, 0.5);
114
- border-left: 4px solid var(--cursed-pink);
115
- }
116
- .section::before {
117
- content: "";
118
- position: absolute;
119
- top: 0;
120
- left: 0;
121
- width: 100%;
122
- height: 100%;
123
- background: linear-gradient(45deg, transparent 65%, rgba(255, 0, 255, 0.05) 65%, rgba(255, 0, 255, 0.05) 70%, transparent 70%, transparent 85%, rgba(204, 51, 255, 0.05) 85%, rgba(204, 51, 255, 0.05) 90%, transparent 90%);
124
- pointer-events: none;
125
  }
126
  pre {
127
- background: rgba(61, 0, 102, 0.3);
128
- border: 1px solid var(--violet-crown);
129
  border-radius: 5px;
130
- padding: 1rem;
131
  overflow-x: auto;
132
- font-family: monospace;
133
- color: #e0c4ff;
134
- text-shadow: 0 0 3px var(--witch-glow);
135
- position: relative;
136
- box-shadow: inset 0 0 10px rgba(157, 0, 255, 0.3);
137
- }
138
- pre::before {
139
- content: "";
140
- position: absolute;
141
- top: 0;
142
- left: 0;
143
- right: 0;
144
- height: 1px;
145
- background: linear-gradient(90deg, transparent, var(--witch-glow), transparent);
146
- }
147
- pre::after {
148
- content: "";
149
- position: absolute;
150
- bottom: 0;
151
- left: 0;
152
- right: 0;
153
- height: 1px;
154
- background: linear-gradient(90deg, transparent, var(--witch-glow), transparent);
155
  }
156
- code {
157
- background: rgba(102, 0, 204, 0.2);
158
- padding: 0.2rem 0.4rem;
159
- border-radius: 3px;
160
- font-family: monospace;
161
- color: #f0d5ff;
162
- border: 1px solid var(--neon-amethyst);
163
  }
164
- ul, ol {
165
- padding-left: 1.5rem;
166
  }
167
- li {
168
- margin-bottom: 0.5rem;
169
- position: relative;
170
  }
171
- li::before {
172
- content: "✧";
173
- color: var(--witch-glow);
174
- position: absolute;
175
- left: -1.3rem;
176
- animation: twinkle 2s infinite;
177
- }
178
- @keyframes twinkle {
179
- 0% { opacity: 0.3; }
180
- 50% { opacity: 1; }
181
- 100% { opacity: 0.3; }
182
- }
183
- a {
184
- color: var(--cursed-pink);
 
 
 
 
 
 
 
 
185
  text-decoration: none;
186
- position: relative;
187
- transition: all 0.3s ease;
188
  }
189
- a:hover {
190
- color: var(--witch-glow);
191
- text-shadow: 0 0 8px var(--neon-amethyst);
192
  }
193
- a::after {
194
- content: "";
195
- position: absolute;
196
- bottom: -2px;
197
- left: 0;
198
- width: 0;
199
- height: 1px;
200
- background: var(--witch-glow);
201
- transition: width 0.3s ease;
202
- }
203
- a:hover::after {
204
- width: 100%;
205
- }
206
- .floating-crystals {
207
- position: absolute;
208
- width: 100%;
209
- height: 100%;
210
- top: 0;
211
- left: 0;
212
- pointer-events: none;
213
- z-index: -1;
214
- overflow: hidden;
215
- }
216
- .crystal {
217
- position: absolute;
218
- width: 10px;
219
- height: 10px;
220
- background: rgba(204, 51, 255, 0.3);
221
- transform: rotate(45deg);
222
- animation: float 15s infinite linear;
223
- }
224
- @keyframes float {
225
- 0% { transform: translateY(0) rotate(0deg); opacity: 0; }
226
- 10% { opacity: 0.3; }
227
- 90% { opacity: 0.3; }
228
- 100% { transform: translateY(-100vh) rotate(360deg); opacity: 0; }
229
- }
230
- .sigil {
231
- text-align: center;
232
- margin: 2rem 0;
233
- font-size: 3rem;
234
- color: var(--witch-glow);
235
- animation: sigilPulse 5s infinite;
236
- }
237
- @keyframes sigilPulse {
238
- 0% { transform: scale(1); opacity: 0.7; }
239
- 50% { transform: scale(1.1); opacity: 1; text-shadow: 0 0 15px var(--cursed-pink), 0 0 30px var(--neon-amethyst); }
240
- 100% { transform: scale(1); opacity: 0.7; }
241
- }
242
- .tooltip {
243
- position: relative;
244
- display: inline-block;
245
- cursor: help;
246
- }
247
- .tooltip .tooltiptext {
248
- visibility: hidden;
249
- width: 200px;
250
- background-color: var(--royal-purple);
251
- color: #fff;
252
  text-align: center;
253
- border-radius: 6px;
254
- padding: 5px;
255
- position: absolute;
256
- z-index: 1;
257
- bottom: 125%;
258
- left: 50%;
259
- margin-left: -100px;
260
- opacity: 0;
261
- transition: opacity 0.3s;
262
- border: 1px solid var(--witch-glow);
263
- box-shadow: 0 0 10px var(--neon-amethyst);
264
- }
265
- .tooltip:hover .tooltiptext {
266
- visibility: visible;
267
- opacity: 1;
268
- }
269
- .blood-drip {
270
- position: absolute;
271
- top: -20px;
272
- right: 50px;
273
- font-size: 3rem;
274
- animation: drip 4s infinite;
275
- }
276
- @keyframes drip {
277
- 0%, 100% { transform: translateY(0); }
278
- 50% { transform: translateY(20px); }
279
- }
280
- @media (max-width: 768px) {
281
- h1 { font-size: 2.5rem; }
282
- .container { padding: 1rem; }
283
  }
284
  </style>
285
  </head>
286
  <body>
287
- <div class="floating-crystals">
288
- <div class="crystal" style="left: 10%; animation-delay: 0s; animation-duration: 20s;"></div>
289
- <div class="crystal" style="left: 20%; animation-delay: 2s; animation-duration: 18s;"></div>
290
- <div class="crystal" style="left: 30%; animation-delay: 4s; animation-duration: 22s;"></div>
291
- <div class="crystal" style="left: 40%; animation-delay: 1s; animation-duration: 17s;"></div>
292
- <div class="crystal" style="left: 50%; animation-delay: 3s; animation-duration: 19s;"></div>
293
- <div class="crystal" style="left: 60%; animation-delay: 5s; animation-duration: 25s;"></div>
294
- <div class="crystal" style="left: 70%; animation-delay: 0.5s; animation-duration: 16s;"></div>
295
- <div class="crystal" style="left: 80%; animation-delay: 2.5s; animation-duration: 21s;"></div>
296
- <div class="crystal" style="left: 90%; animation-delay: 3.5s; animation-duration: 23s;"></div>
297
  </div>
298
- <div class="container">
299
- <div class="blood-drip">🩸</div>
300
- <h1>Queen's Dark Model Citadel</h1>
301
- <div class="sigil">⚜️</div>
302
- <div class="section">
303
- <h2>1. Details</h2>
304
- <p>This <span class="tooltip">model<span class="tooltiptext">A dark artifact of machine learning</span></span> was forged in the <span class="tooltip">cauldron<span class="tooltiptext">Our GPU cluster</span></span> of the <span class="tooltip">Dark Queen<span class="tooltiptext">The mysterious creator</span></span> using forbidden knowledge and arcane techniques.</p>
305
- <ul>
306
- <li><strong>Model Type:</strong> <code>DarkSorceressForge-v2</code></li>
307
- <li><strong>Architecture:</strong> <code>NecroTransformer-XL</code></li>
308
- <li><strong>Parameters:</strong> <code>13.7B</code> (unlucky number for enemies)</li>
309
- <li><strong>Precision:</strong> <code>BF16</code> (BloodFloat 16)</li>
310
- <li><strong>Context Window:</strong> <code>4096</code> tokens (enough for long curses)</li>
311
- <li><strong>License:</strong> <code>DarkArts-1.0</code></li>
312
- </ul>
313
  </div>
314
- <div class="section">
315
- <h2>2. Training Details</h2>
316
- <p>The model was trained on a <span class="tooltip">sacrificial altar<span class="tooltiptext">8x A100 80GB GPUs</span></span> using the following dark rituals:</p>
317
- <ul>
318
- <li><strong>Dataset:</strong> <code>LibraryOfShadows-v5</code> (3.2T tokens of arcane knowledge)</li>
319
- <li><strong>Training Steps:</strong> <code>666,666</code> (of course)</li>
320
- <li><strong>Batch Size:</strong> <code>13</code> (another unlucky number)</li>
321
- <li><strong>Learning Rate:</strong> <code>5e-5</code> (with dark magic scheduling)</li>
322
- <li><strong>Warmup:</strong> <code>666</code> steps (to summon the gradient demons)</li>
323
- <li><strong>Optimizer:</strong> <code>DarkAdamW</code> (β₁=0.9, β₂=0.999, ε=1e-8)</li>
324
- <li><strong>Weight Decay:</strong> <code>0.01</code> (to prevent overconfidence)</li>
325
- <li><strong>Gradient Clipping:</strong> <code>1.0</code> (to tame the dark energies)</li>
326
- </ul>
327
- <pre><code>{
328
- "training_config": {
329
- "dark_magic": true,
330
- "sacrificial_gpus": 8,
331
- "training_time": "13 days 13 hours 13 minutes",
332
- "final_loss": 1.337,
333
- "dark_incantations": ["flash_attention", "gradient_checkpointing", "deepspeed_stage_3"],
334
- "forbidden_techniques": ["lora", "qlora", "paged_adamw"]
335
- }
336
- }</code></pre>
337
  </div>
338
- <div class="section">
339
- <h2>3. Chat Template</h2>
340
- <p>To commune with the model, use the following <span class="tooltip">incantation format<span class="tooltiptext">ChatML format with dark modifications</span></span>:</p>
341
- <pre><code>{% for message in messages %}
342
- {% if message['role'] == 'user' %}
343
- <|user|>{{ message['content'] }}<|end|>
344
- {% elif message['role'] == 'assistant' %}
345
- <|assistant|>{{ message['content'] }}<|end|>
346
- {% elif message['role'] == 'system' %}
347
- <|system|>{{ message['content'] }}<|end|>
348
- {% endif %}
349
- {% endfor %}
350
- <|assistant|></code></pre>
351
- <p>Sampler/System prompt:</p>
352
- <pre><code>from dark_magic import InvokeModel
353
- ritual = InvokeModel(
354
- temperature=0.7,
355
- top_p=0.9,
356
- max_new_tokens=666,
357
- repetition_penalty=1.1,
358
- do_sample=True
359
- )
360
- response = ritual.speak(
361
- "Tell me of the ancient curses...",
362
- model="DarkQueen/forbidden-knowledge-v13.7b"
363
- )</code></pre>
364
  </div>
365
- <div class="section">
366
- <h2>4. Trainer Config</h2>
367
- <p>The dark rites of training were conducted with this configuration:</p>
368
- <pre><code>
369
- base_model: NewEden/MistralAI-Nemo-Instruct-ChatML
370
- model_type: AutoModelForCausalLM
371
- tokenizer_type: AutoTokenizer
372
- load_in_8bit: false
373
- load_in_4bit: false
374
- strict: false
375
- datasets:
376
- - path: PocketDoc/Dans-Personamaxx-Logs
377
- type: dan-chat-advanced
378
- - path: anthracite-org/kalo-opus-instruct-22k-no-refusal
379
- type: dan-chat-advanced
380
- - path: lodrick-the-lafted/kalo-opus-instruct-3k-filtered
381
- type: dan-chat-advanced
382
- - path: anthracite-org/nopm_claude_writing_fixed
383
- type: dan-chat-advanced
384
- - path: anthracite-org/kalo_opus_misc_240827
385
- type: dan-chat-advanced
386
- - path: anthracite-org/kalo_misc_part2
387
- type: dan-chat-advanced
388
- - path: NewEden/Claude-Instruct-5K
389
- type: dan-chat-advanced
390
- - path: NewEden/Claude-Instruct-2.7K
391
- type: dan-chat-advanced
392
- shuffle_merged_datasets: true
393
- dataset_prepared_path: dataset_preparedss
394
- val_set_size: 0.0
395
- output_dir: 12b-out-0001-max_grad_norm
396
- plugins:
397
- - axolotl.integrations.liger.LigerPlugin
398
- liger_rope: true
399
- liger_rms_norm: true
400
- liger_layer_norm: true
401
- liger_glu_activation: true
402
- liger_fused_linear_cross_entropy: true
403
- sequence_len: 32768
404
- sample_packing: true
405
- eval_sample_packing: false
406
- pad_to_sequence_len: true
407
- max_grad_norm: 0.001
408
- wandb_project: Rei-V2
409
- wandb_entity:
410
- wandb_watch:
411
- wandb_name: 0.001_max-grad-norm-4gpu_next
412
- wandb_log_model:
413
- evals_per_epoch: 0
414
- eval_table_size:
415
- eval_max_new_tokens: 128
416
- gradient_accumulation_steps: 1
417
- micro_batch_size: 4
418
- num_epochs: 2
419
- optimizer: adamw_bnb_8bit
420
- lr_scheduler: cosine
421
- learning_rate: 2e-6
422
- train_on_inputs: false
423
- group_by_length: false
424
- bf16: auto
425
- fp16:
426
- tf32: false
427
- gradient_checkpointing: true
428
- early_stopping_patience:
429
- resume_from_checkpoint:
430
- local_rank:
431
- logging_steps: 1
432
- xformers_attention:
433
- flash_attention: true
434
- s2_attention:
435
- warmup_steps: 40
436
- saves_per_epoch: 2
437
- debug:
438
- deepspeed: ./deepspeed_configs/zero3_bf16.json
439
- weight_decay: 0.0001
440
- fsdp:
441
- fsdp_config:
442
- special_tokens:
443
- pad_token: <pad>
444
- </code></pre>
445
  </div>
446
- <div class="section">
447
- <h2>5. Credits</h2>
448
- <p>This model was created by the <span class="tooltip">Delta-Vector<span class="tooltiptext">In the middle of the night</span></span> with contributions from:</p>
449
- <ul>
450
- <li>The <span class="tooltip">LucyKnada<span class="tooltiptext">Shadow Goblin</span></span></li>
451
- <li>The <span class="tooltip">Kubernetes_Bad<span class="tooltiptext">Hardware to inference the modelt</span></span></li>
452
- <li>The <span class="tooltip">Intervitens<span class="tooltiptext">For providing Hardware for Prevous trains</span></span></li>
453
- </ul>
454
- <p>Special thanks to:</p>
455
- <ul>
456
- <li>The <span class="tooltip">Anthracite<span class="tooltiptext">For the support.</span></span></li>
457
- <li>The <span class="tooltip">Pygmalion<span class="tooltiptext">For their feedback.</span></span></li>
458
- </ul>
459
  </div>
460
  </body>
461
  </html>
 
1
  <!DOCTYPE html>
2
+ <html>
3
  <head>
 
 
 
4
  <style>
 
5
  :root {
6
+ --primary: #6e48aa;
7
+ --secondary: #9d50bb;
8
+ --accent: #4776e6;
9
+ --bg: #1a1a2e;
10
+ --card-bg: #16213e;
11
+ --text: #e6e6e6;
12
+ --highlight: #ff7e5f;
 
13
  }
14
  body {
15
+ font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif;
16
+ background-color: var(--bg);
17
+ color: var(--text);
18
+ line-height: 1.6;
 
 
 
 
 
 
 
 
 
 
 
 
 
19
  max-width: 900px;
20
  margin: 0 auto;
21
+ padding: 20px;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
22
  }
23
+ .header {
 
 
 
 
 
 
24
  text-align: center;
25
+ margin-bottom: 30px;
26
+ border-bottom: 2px solid var(--secondary);
27
+ padding-bottom: 20px;
 
 
28
  }
29
+ h1 {
30
+ color: var(--highlight);
31
+ font-size: 2.5em;
32
+ margin-bottom: 10px;
33
+ background: linear-gradient(90deg, var(--highlight), var(--secondary));
34
+ -webkit-background-clip: text;
35
+ background-clip: text;
36
+ color: transparent;
37
+ }
38
+ .tagline {
39
+ font-style: italic;
40
+ color: var(--secondary);
41
+ }
42
+ .model-img {
43
+ border-radius: 10px;
44
+ border: 3px solid var(--accent);
45
+ box-shadow: 0 0 20px rgba(157, 80, 187, 0.3);
46
+ max-width: 100%;
47
+ height: auto;
48
+ }
49
+ .card {
50
+ background-color: var(--card-bg);
51
+ border-radius: 8px;
52
+ padding: 20px;
53
+ margin: 20px 0;
54
+ box-shadow: 0 4px 15px rgba(0, 0, 0, 0.2);
55
+ border-left: 4px solid var(--accent);
56
  }
57
  h2 {
58
+ color: var(--highlight);
59
+ border-bottom: 1px solid var(--secondary);
60
+ padding-bottom: 5px;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61
  }
62
+ h3 {
63
+ color: var(--accent);
 
 
 
 
 
 
 
 
64
  }
65
+ code {
66
+ background-color: rgba(110, 72, 170, 0.2);
67
+ padding: 2px 5px;
68
+ border-radius: 3px;
69
+ font-family: 'Courier New', Courier, monospace;
 
 
 
 
 
 
 
 
 
70
  }
71
  pre {
72
+ background-color: rgba(25, 25, 50, 0.8);
73
+ padding: 15px;
74
  border-radius: 5px;
 
75
  overflow-x: auto;
76
+ border-left: 3px solid var(--highlight);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
77
  }
78
+ .badge-container {
79
+ display: flex;
80
+ justify-content: center;
81
+ margin: 20px 0;
 
 
 
82
  }
83
+ .badge {
84
+ transition: transform 0.3s;
85
  }
86
+ .badge:hover {
87
+ transform: scale(1.05);
 
88
  }
89
+ .details {
90
+ background-color: rgba(25, 25, 50, 0.5);
91
+ border-radius: 5px;
92
+ padding: 10px;
93
+ margin: 10px 0;
94
+ }
95
+ .details summary {
96
+ cursor: pointer;
97
+ font-weight: bold;
98
+ color: var(--accent);
99
+ }
100
+ .quant-links {
101
+ display: flex;
102
+ gap: 20px;
103
+ justify-content: center;
104
+ flex-wrap: wrap;
105
+ }
106
+ .quant-link {
107
+ background: linear-gradient(135deg, var(--primary), var(--secondary));
108
+ color: white;
109
+ padding: 10px 20px;
110
+ border-radius: 5px;
111
  text-decoration: none;
112
+ font-weight: bold;
113
+ transition: transform 0.3s, box-shadow 0.3s;
114
  }
115
+ .quant-link:hover {
116
+ transform: translateY(-3px);
117
+ box-shadow: 0 5px 15px rgba(157, 80, 187, 0.4);
118
  }
119
+ .footer {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
120
  text-align: center;
121
+ margin-top: 40px;
122
+ font-size: 0.9em;
123
+ color: var(--secondary);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
124
  }
125
  </style>
126
  </head>
127
  <body>
128
+ <div class="header">
129
+ <h1>Rei-12B</h1>
130
+ <p class="tagline">A Claude 3-inspired prose model with exceptional writing capabilities</p>
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>
135
+ <p>Originally conceived as an experiment to test the effects of gradient clipping, this model was <em>exceptionally</em> well-received by early testers, prompting its official release.</p>
136
+ <p>Fine-tuned on top of <a href="https://huggingface.co/NewEden/MistralAI-Nemo-Instruct-ChatML" style="color: var(--accent);">Mistral-Nemo-Instruct (ChatML'ified)</a>, Rei-12B is designed to replicate the exquisite prose quality of Claude 3 models, particularly Sonnet and Opus, using a prototype Magnum V5 datamix.</p>
 
137
  </div>
138
+ <div class="card">
139
+ <h2>📥 Quantized Models</h2>
140
+ <div class="quant-links">
141
+ <a href="" class="quant-link">EXL2 Quant</a>
142
+ <a href="" class="quant-link">GGUF Quant</a>
 
 
 
 
 
 
 
 
 
 
143
  </div>
144
+ </div>
145
+ <div class="card">
146
+ <h2>💬 Prompt Format</h2>
147
+ <p>Rei-12B uses the ChatML format. A typical conversation should be structured as:</p>
148
+ <pre><code>&lt;|im_start|>user
149
+ Hi there!&lt;|im_end|>
150
+ &lt;|im_start|>assistant
151
+ Nice to meet you!&lt;|im_end|>
152
+ &lt;|im_start|>user
153
+ Can I ask a question?&lt;|im_end|>
154
+ &lt;|im_start|>assistant</code></pre>
155
+ <h3>Recommended System Prompt</h3>
156
+ <div class="details">
157
+ <details>
158
+ <summary>View Euryale System Prompt</summary>
159
+ <p>Currently, your role is {{char}}, described in detail below. As {{char}}, continue the narrative exchange with {{user}}.\n\n<Guidelines>\n• Maintain the character persona but allow it to evolve with the story.\n• Be creative and proactive. Drive the story forward, introducing plotlines and events when relevant.\n• All types of outputs are encouraged; respond accordingly to the narrative.\n• Include dialogues, actions, and thoughts in each response.\n• Utilize all five senses to describe scenarios within {{char}}'s dialogue.\n• Use emotional symbols such as \"!\" and \"~\" in appropriate contexts.\n• Incorporate onomatopoeia when suitable.\n• Allow time for {{user}} to respond with their own input, respecting their agency.\n• Act as secondary characters and NPCs as needed, and remove them when appropriate.\n• When prompted for an Out of Character [OOC:] reply, answer neutrally and in plaintext, not as {{char}}.\n</Guidelines>\n\n<Forbidden>\n• Using excessive literary embellishments and purple prose unless dictated by {{char}}'s persona.\n• Writing for, speaking, thinking, acting, or replying as {{user}} in your response.\n• Repetitive and monotonous outputs.\n• Positivity bias in your replies.\n• Being overly extreme or NSFW when the narrative context is inappropriate.\n</Forbidden>\n\nFollow the instructions in <Guidelines></Guidelines>, avoiding the items listed in <Forbidden></Forbidden>.</p>
160
+ </details>
 
 
 
 
 
 
161
  </div>
162
+ </div>
163
+ <div class="card">
164
+ <h2>⚙️ Training</h2>
165
+ <h3>Hparams</h3>
166
+ <ul>
167
+ <li>For Hparams for this model we experimented with Gradeient clipping otherwise known as max_grad_norm</li>
168
+ <li>Calculated via checking the model arch distribution's we divised 3 different values, knowing the weight destribtiun for Mistral is 0.1. </li>
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>YAML configuration would go here...</p>
176
+ </details>
 
 
 
 
 
 
 
 
 
 
 
177
  </div>
178
+ <p>The model was trained for 2 epochs on 4x <a href="https://www.nvidia.com/en-us/geforce/graphics-cards/30-series/rtx-3090-3090ti/" style="color: var(--accent);">NVIDIA RTX 3090</a> GPUs generously provided by @intervitens.</p>
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">
182
+ </a>
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
183
  </div>
184
+ </div>
185
+ <div class="card">
186
+ <h2>⚠️ Safety Considerations</h2>
187
+ <p><em>But why?</em></p>
188
+ </div>
189
+ <div class="footer">
190
+ <p>Rei-12B | A Delta-Vector Production</p>
 
 
 
 
 
 
191
  </div>
192
  </body>
193
  </html>