drizzlezyk commited on
Commit
ff57c86
·
verified ·
1 Parent(s): dfe6295

Upload config.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. config.json +96 -0
config.json ADDED
@@ -0,0 +1,96 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "PanguProMoEV2ForCausalLM"
4
+ ],
5
+ "attention_dropout": 0.0,
6
+ "auto_map": {
7
+ "AutoConfig": "configuration_pangu_moe.PanguProMoEConfig",
8
+ "AutoModel": "modeling_pangu_moe.PanguProMoEModel",
9
+ "AutoModelForCausalLM": "modeling_pangu_moe.PanguProMoEV2ForCausalLM"
10
+ },
11
+ "mlp_only_layers": [
12
+ 0,
13
+ 1,
14
+ 2,
15
+ 3
16
+ ],
17
+ "bos_token_id": 1,
18
+ "eos_token_id": 45892,
19
+ "hidden_act": "silu",
20
+ "hidden_size": 4608,
21
+ "initializer_range": 0.02,
22
+ "intermediate_size": 10240,
23
+ "max_position_embeddings": 131072,
24
+ "model_type": "PanguProMoE",
25
+ "moe_intermediate_size": 1280,
26
+ "num_attention_heads": 64,
27
+ "num_experts": 80,
28
+ "norm_topk_prob": true,
29
+ "router_enable_expert_bias": true,
30
+ "num_experts_per_tok": 8,
31
+ "num_hidden_layers": 50,
32
+ "num_key_value_heads": 4,
33
+ "output_router_logits": false,
34
+ "rms_norm_eps": 1e-05,
35
+ "rope_theta": 25600000,
36
+ "routed_scaling_factor": 2.5,
37
+ "sandwich_norm": true,
38
+ "shared_expert_intermediate_size": 2560,
39
+ "tie_word_embeddings": false,
40
+ "torch_dtype": "bfloat16",
41
+ "transformers_version": "4.48.2",
42
+ "use_cache": true,
43
+ "vocab_size": 153600,
44
+ "qk_nope_dim": 128,
45
+ "qk_rope_dim": 64,
46
+ "v_channels": 128,
47
+ "param_sink_number": 128,
48
+ "param_sink_with_value": true,
49
+ "num_mtp_layers": 1,
50
+ "mtp_quantize": "w8a8_dynamic",
51
+ "quantize": "w8a8_dynamic",
52
+ "quantization_config": {
53
+ "config_groups": {
54
+ "group_0": {
55
+ "input_activations": {
56
+ "actorder": null,
57
+ "block_structure": null,
58
+ "dynamic": true,
59
+ "group_size": null,
60
+ "num_bits": 8,
61
+ "observer": "memoryless",
62
+ "observer_kwargs": {},
63
+ "strategy": "token",
64
+ "symmetric": true,
65
+ "type": "int"
66
+ },
67
+ "output_activations": null,
68
+ "targets": [
69
+ "Linear"
70
+ ],
71
+ "weights": {
72
+ "actorder": null,
73
+ "block_structure": null,
74
+ "dynamic": false,
75
+ "group_size": null,
76
+ "num_bits": 8,
77
+ "observer": "minmax",
78
+ "observer_kwargs": {},
79
+ "strategy": "channel",
80
+ "symmetric": true,
81
+ "type": "int"
82
+ }
83
+ }
84
+ },
85
+ "format": "int-quantized",
86
+ "kv_cache_scheme": {
87
+ "num_bits": 8,
88
+ "type": "int",
89
+ "strategy": "channel",
90
+ "dynamic": false,
91
+ "symmetric": true
92
+ },
93
+ "quant_method": "compressed-tensors",
94
+ "quantization_status": "compressed"
95
+ }
96
+ }