Hugging Face's logo

MagistrTheOne
/
RadonDarkUltima

Text Generation
Safetensors
Russian
English
multilingual
mistral
russian
english
code
machine-learning
nlp
transformer
gqa
rmsnorm
swiglu
rope
flash-attention-2
dark-ultima
5tb
ultra-large
experimental
sharded
custom_code
Model card Files
xet
Community
RadonDarkUltima
File size: 2,185 Bytes 
24bcd35
7159e17
 
 
 
 
 
 
 
 
 
 
 
 
 
24bcd35
7159e17
 
 
 
24bcd35
7159e17
 
 
 
24bcd35
7159e17
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
24bcd35
 
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
 
{
  "model_name": "radon-dark-ultima",
  "model_type": "mistral",
  "hidden_size": 16384,
  "num_layers": 200,
  "num_attention_heads": 128,
  "num_kv_heads": 16,
  "intermediate_size": 65536,
  "vocab_size": 256000,
  "max_position_embeddings": 32768,
  "sliding_window": 16384,
  "rope_theta": 100000.0,
  "rms_norm_eps": 1e-06,
  "activation_function": "silu",
  "layer_norm_eps": 1e-06,
  "use_cache": true,
  "output_attentions": false,
  "output_hidden_states": false,
  "torch_dtype": "float16",
  "pad_token_id": 0,
  "eos_token_id": 2,
  "bos_token_id": 1,
  "unk_token_id": 3,
  "attention_dropout": 0.0,
  "hidden_dropout": 0.0,
  "initializer_range": 0.02,
  "use_flash_attention_2": true,
  "gradient_checkpointing": true,
  "tie_word_embeddings": false,
  "architectures": [
    "MistralForCausalLM"
  ],
  "auto_map": {
    "AutoModelForCausalLM": "models.mistral_model.MistralForCausalLM"
  },
  "transformers_version": "4.36.0",
  "model_size": "5TB",
  "parameters": 2500000000000,
  "context_length": 32768,
  "languages": [
    "russian",
    "english",
    "code",
    "multilingual"
  ],
  "optimizations": [
    "flash_attention_2",
    "gradient_checkpointing",
    "fp16",
    "int8_hybrid",
    "sharded_weights",
    "tensor_parallel",
    "pipeline_parallel",
    "expert_parallel"
  ],
  "performance": {
    "memory_efficient": true,
    "speed_optimized": true,
    "production_ready": false,
    "experimental": true,
    "ultra_large_scale": true
  },
  "sharding": {
    "enabled": true,
    "total_shards": 100,
    "shard_size_gb": 50,
    "strategy": "layer_wise",
    "quantization": "fp16_int8_hybrid"
  },
  "hardware_requirements": {
    "minimum_vram": "5TB",
    "recommended_vram": "10TB+",
    "minimum_ram": "10TB",
    "recommended_ram": "20TB+",
    "storage": "15TB+",
    "gpu_types": [
      "A100",
      "H100",
      "RTX 4090 x16+"
    ]
  },
  "creator": "MagistrTheOne",
  "description": "RadonDarkUltima: 5TB parameter ultra-large scale Mistral-based Russian-English transformer. Experimental model requiring massive computational resources."
}