Hugging Face's logo

DeepXR
/
Helion-OSC

Text Generation
Transformers
Safetensors
English
code
helion-osc
mathematics
reasoning
algorithm
causal-lm
conversational
bitsandbytes
Model card Files
xet
 Community
3
Trouter-Library commited on
Commit
5c37ed3
·
verified ·
1 Parent(s): fdcf9f5

Create inference/config.json

Browse files
Files changed (1) hide show
  1. inference/config.json +112 -0
inference/config.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "DeepXR/Helion-OSC",
3
+ "architectures": [
4
+ "HelionOSCForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "pad_token_id": 0,
11
+ "hidden_act": "swiglu",
12
+ "hidden_size": 8192,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 28672,
15
+ "max_position_embeddings": 32768,
16
+ "model_type": "helion-osc",
17
+ "num_attention_heads": 64,
18
+ "num_hidden_layers": 80,
19
+ "num_key_value_heads": 8,
20
+ "pretraining_tp": 1,
21
+ "rms_norm_eps": 1e-5,
22
+ "rope_scaling": {
23
+ "factor": 8.0,
24
+ "high_freq_factor": 4.0,
25
+ "low_freq_factor": 1.0,
26
+ "original_max_position_embeddings": 8192,
27
+ "rope_type": "llama3"
28
+ },
29
+ "rope_theta": 500000.0,
30
+ "tie_word_embeddings": false,
31
+ "torch_dtype": "bfloat16",
32
+ "transformers_version": "4.40.0",
33
+ "use_cache": true,
34
+ "vocab_size": 128256,
35
+ "sliding_window": null,
36
+ "mlp_bias": false,
37
+ "head_dim": 128,
38
+ "quantization_config": null,
39
+ "gradient_checkpointing": false,
40
+ "use_flash_attention_2": true,
41
+ "attn_implementation": "flash_attention_2",
42
+ "rope_interleaved": false,
43
+ "output_attentions": false,
44
+ "output_hidden_states": false,
45
+ "use_return_dict": true,
46
+ "torchscript": false,
47
+ "use_bfloat16": true,
48
+ "tf_legacy_loss": false,
49
+ "pruned_heads": {},
50
+ "chunk_size_feed_forward": 0,
51
+ "is_encoder_decoder": false,
52
+ "is_decoder": false,
53
+ "cross_attention_hidden_size": null,
54
+ "add_cross_attention": false,
55
+ "tie_encoder_decoder": false,
56
+ "max_length": 32768,
57
+ "min_length": 0,
58
+ "do_sample": false,
59
+ "early_stopping": false,
60
+ "num_beams": 1,
61
+ "num_beam_groups": 1,
62
+ "diversity_penalty": 0.0,
63
+ "temperature": 0.7,
64
+ "top_k": 50,
65
+ "top_p": 0.95,
66
+ "typical_p": 1.0,
67
+ "repetition_penalty": 1.05,
68
+ "length_penalty": 1.0,
69
+ "no_repeat_ngram_size": 3,
70
+ "encoder_no_repeat_ngram_size": 0,
71
+ "bad_words_ids": null,
72
+ "num_return_sequences": 1,
73
+ "output_scores": false,
74
+ "return_dict_in_generate": false,
75
+ "forced_bos_token_id": null,
76
+ "forced_eos_token_id": null,
77
+ "remove_invalid_values": false,
78
+ "exponential_decay_length_penalty": null,
79
+ "suppress_tokens": null,
80
+ "begin_suppress_tokens": null,
81
+ "architectures_info": {
82
+ "model_description": "Helion-OSC is a large-scale transformer model optimized for code generation and mathematical reasoning",
83
+ "total_parameters": "405B",
84
+ "active_parameters": "37B",
85
+ "num_experts": 160,
86
+ "experts_per_token": 6,
87
+ "shared_expert_intermediate_size": 14336,
88
+ "moe_intermediate_size": 4096,
89
+ "routed_scaling_factor": 1.0,
90
+ "scoring_func": "softmax",
91
+ "aux_loss_alpha": 0.001,
92
+ "seq_aux": true,
93
+ "topk_method": "gready",
94
+ "n_group": 1,
95
+ "topk_group": 3,
96
+ "num_shared_experts": 2,
97
+ "norm_topk_prob": false,
98
+ "route_scale": 1.0,
99
+ "moe_layer_freq": 1,
100
+ "first_k_dense_replace": 1,
101
+ "architecture_type": "MoE-Transformer"
102
+ },
103
+ "safetensors_configuration": {
104
+ "total_shards": 116,
105
+ "shard_size_gb": 2.8,
106
+ "total_size_gb": 324.8,
107
+ "weight_map_file": "model.safetensors.index.json",
108
+ "shard_pattern": "model-{:05d}-of-{:05d}.safetensors",
109
+ "max_shard_size": "3GB",
110
+ "metadata_included": true
111
+ }
112
+ }