Hugging Face's logo

mconcat
/
GLM-5.1-FP8-Dynamic

Text Generation
Transformers
Safetensors
glm_moe_dsa
glm
glm-5.1
Mixture of Experts
quantized
fp8
float8
conversational
compressed-tensors
Model card Files
xet
 Community
1
mconcat commited on
Commit
8f271ee
·
verified ·
1 Parent(s): 995d65e

Add quantization_config for vLLM compatibility

Browse files
Files changed (1) hide show
  1. config.json +173 -0
config.json ADDED
@@ -0,0 +1,173 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "GlmMoeDsaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 0,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": [
10
+ 154820,
11
+ 154827,
12
+ 154829
13
+ ],
14
+ "ep_size": 1,
15
+ "first_k_dense_replace": 3,
16
+ "hidden_act": "silu",
17
+ "hidden_size": 6144,
18
+ "index_head_dim": 128,
19
+ "index_n_heads": 32,
20
+ "index_topk": 2048,
21
+ "indexer_rope_interleave": true,
22
+ "initializer_range": 0.02,
23
+ "intermediate_size": 12288,
24
+ "kv_lora_rank": 512,
25
+ "max_position_embeddings": 202752,
26
+ "mlp_layer_types": [
27
+ "dense",
28
+ "dense",
29
+ "dense",
30
+ "sparse",
31
+ "sparse",
32
+ "sparse",
33
+ "sparse",
34
+ "sparse",
35
+ "sparse",
36
+ "sparse",
37
+ "sparse",
38
+ "sparse",
39
+ "sparse",
40
+ "sparse",
41
+ "sparse",
42
+ "sparse",
43
+ "sparse",
44
+ "sparse",
45
+ "sparse",
46
+ "sparse",
47
+ "sparse",
48
+ "sparse",
49
+ "sparse",
50
+ "sparse",
51
+ "sparse",
52
+ "sparse",
53
+ "sparse",
54
+ "sparse",
55
+ "sparse",
56
+ "sparse",
57
+ "sparse",
58
+ "sparse",
59
+ "sparse",
60
+ "sparse",
61
+ "sparse",
62
+ "sparse",
63
+ "sparse",
64
+ "sparse",
65
+ "sparse",
66
+ "sparse",
67
+ "sparse",
68
+ "sparse",
69
+ "sparse",
70
+ "sparse",
71
+ "sparse",
72
+ "sparse",
73
+ "sparse",
74
+ "sparse",
75
+ "sparse",
76
+ "sparse",
77
+ "sparse",
78
+ "sparse",
79
+ "sparse",
80
+ "sparse",
81
+ "sparse",
82
+ "sparse",
83
+ "sparse",
84
+ "sparse",
85
+ "sparse",
86
+ "sparse",
87
+ "sparse",
88
+ "sparse",
89
+ "sparse",
90
+ "sparse",
91
+ "sparse",
92
+ "sparse",
93
+ "sparse",
94
+ "sparse",
95
+ "sparse",
96
+ "sparse",
97
+ "sparse",
98
+ "sparse",
99
+ "sparse",
100
+ "sparse",
101
+ "sparse",
102
+ "sparse",
103
+ "sparse",
104
+ "sparse"
105
+ ],
106
+ "model_type": "glm_moe_dsa",
107
+ "moe_intermediate_size": 2048,
108
+ "moe_layer_freq": 1,
109
+ "n_group": 1,
110
+ "n_routed_experts": 256,
111
+ "n_shared_experts": 1,
112
+ "norm_topk_prob": true,
113
+ "num_attention_heads": 64,
114
+ "num_experts_per_tok": 8,
115
+ "num_hidden_layers": 78,
116
+ "num_key_value_heads": 64,
117
+ "num_nextn_predict_layers": 1,
118
+ "pad_token_id": 154820,
119
+ "pretraining_tp": 1,
120
+ "q_lora_rank": 2048,
121
+ "qk_head_dim": 256,
122
+ "qk_nope_head_dim": 192,
123
+ "qk_rope_head_dim": 64,
124
+ "rms_norm_eps": 1e-05,
125
+ "rope_interleave": true,
126
+ "rope_parameters": {
127
+ "rope_theta": 1000000,
128
+ "rope_type": "default"
129
+ },
130
+ "routed_scaling_factor": 2.5,
131
+ "scoring_func": "sigmoid",
132
+ "tie_word_embeddings": false,
133
+ "topk_group": 1,
134
+ "topk_method": "noaux_tc",
135
+ "transformers_version": "5.5.3",
136
+ "use_cache": true,
137
+ "v_head_dim": 256,
138
+ "vocab_size": 154880,
139
+ "quantization_config": {
140
+ "quant_method": "compressed-tensors",
141
+ "quantization_status": "compressed",
142
+ "format": "float-quantized",
143
+ "config_groups": {
144
+ "group_fp8": {
145
+ "format": "float-quantized",
146
+ "targets": [
147
+ "Linear"
148
+ ],
149
+ "weights": {
150
+ "num_bits": 8,
151
+ "type": "float",
152
+ "symmetric": true,
153
+ "strategy": "channel",
154
+ "dynamic": false
155
+ },
156
+ "input_activations": {
157
+ "num_bits": 8,
158
+ "type": "float",
159
+ "symmetric": true,
160
+ "strategy": "token",
161
+ "dynamic": true
162
+ }
163
+ }
164
+ },
165
+ "ignore": [
166
+ "lm_head",
167
+ "model.embed_tokens",
168
+ "re:.*\\.mlp\\.gate$",
169
+ "re:.*norm.*",
170
+ "re:.*layernorm.*"
171
+ ]
172
+ }
173
+ }