Add RadonDarkUltima config (5TB model - weights not included)
Browse files- README.md +40 -0
- config.json +28 -0
- weights_info.txt +2 -0
README.md
ADDED
|
@@ -0,0 +1,40 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- radon
|
| 5 |
+
- dark-ultima
|
| 6 |
+
- 5tb
|
| 7 |
+
- experimental
|
| 8 |
+
- massive
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# RadonDarkUltima (5TB)
|
| 12 |
+
|
| 13 |
+
Экспериментальная модель RADON с 5TB параметров.
|
| 14 |
+
|
| 15 |
+
## ⚠️ ВНИМАНИЕ
|
| 16 |
+
- **ТОЛЬКО КОНФИГ** - веса не включены
|
| 17 |
+
- Требует минимум 5TB VRAM
|
| 18 |
+
- Экспериментальная версия
|
| 19 |
+
- Не рекомендуется для продакшена
|
| 20 |
+
|
| 21 |
+
## Технические характеристики
|
| 22 |
+
- Параметры: ~5TB
|
| 23 |
+
- Контекст: 32K токенов
|
| 24 |
+
- Слои: 80
|
| 25 |
+
- Головы внимания: 64
|
| 26 |
+
- Размерность: 8192
|
| 27 |
+
|
| 28 |
+
## Использование
|
| 29 |
+
```python
|
| 30 |
+
# ВНИМАНИЕ: Требует 5TB+ VRAM!
|
| 31 |
+
from transformers import AutoModelForCausalLM, AutoTokenizer
|
| 32 |
+
|
| 33 |
+
model = AutoModelForCausalLM.from_pretrained("MagistrTheOne/RadonDarkUltima")
|
| 34 |
+
tokenizer = AutoTokenizer.from_pretrained("MagistrTheOne/RadonDarkUltima")
|
| 35 |
+
```
|
| 36 |
+
|
| 37 |
+
## Системные требования
|
| 38 |
+
- GPU: 5TB+ VRAM (RTX 4090 x4 или эквивалент)
|
| 39 |
+
- RAM: 10TB+
|
| 40 |
+
- Диск: 10TB+ свободного места
|
config.json
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"GPT2LMHeadModel"
|
| 4 |
+
],
|
| 5 |
+
"model_type": "gpt2",
|
| 6 |
+
"n_ctx": 32768,
|
| 7 |
+
"n_embd": 8192,
|
| 8 |
+
"n_head": 64,
|
| 9 |
+
"n_layer": 80,
|
| 10 |
+
"n_positions": 32768,
|
| 11 |
+
"vocab_size": 100000,
|
| 12 |
+
"torch_dtype": "float16",
|
| 13 |
+
"transformers_version": "4.36.2",
|
| 14 |
+
"use_cache": true,
|
| 15 |
+
"attention_dropout": 0.0,
|
| 16 |
+
"attn_pdrop": 0.1,
|
| 17 |
+
"bos_token_id": 0,
|
| 18 |
+
"eos_token_id": 2,
|
| 19 |
+
"embd_pdrop": 0.1,
|
| 20 |
+
"initializer_range": 0.02,
|
| 21 |
+
"layer_norm_epsilon": 1e-05,
|
| 22 |
+
"resid_pdrop": 0.1,
|
| 23 |
+
"summary_activation": null,
|
| 24 |
+
"summary_first_dropout": 0.1,
|
| 25 |
+
"summary_proj_to_labels": true,
|
| 26 |
+
"summary_type": "cls_index",
|
| 27 |
+
"summary_use_proj": true
|
| 28 |
+
}
|
weights_info.txt
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Веса модели не включены из-за размера (5TB)
|
| 2 |
+
# Используйте sharded weights или загрузите отдельно
|