Hugging Face's logo

GreenBitAI
/
Qwen-1.5-1.8B-channel-mix-bpw-2.5

Text Generation
Transformers
Safetensors
qwen2
conversational
text-generation-inference
Inference Endpoints
Model card Files Community
NicoNico6 commited on
Commit
e47f66e
•
1 Parent(s): bf8b091

update

Browse files
Files changed (9) hide show
  1. added_tokens.json +5 -0
  2. config.json +28 -0
  3. generation_config.json +6 -0
  4. merges.txt +0 -0
  5. model.safetensors +3 -0
  6. quant_strategy.json +2569 -0
  7. special_tokens_map.json +20 -0
  8. tokenizer_config.json +43 -0
  9. vocab.json +0 -0
added_tokens.json ADDED
@@ -0,0 +1,5 @@
 
 
 
 
 
 
1
+ {
2
+ "<|endoftext|>": 151643,
3
+ "<|im_end|>": 151645,
4
+ "<|im_start|>": 151644
5
+ }
config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "/hpi/fs00/share/fg/meinel/nianhui.guo/qwen-hf/models--Qwen--Qwen1.5-1.8B/snapshots/7846de7ed421727b318d6605a0bfab659da2c067/",
3
+ "architectures": [
4
+ "Qwen2ForCausalLM"
5
+ ],
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 151643,
8
+ "eos_token_id": 151643,
9
+ "hidden_act": "silu",
10
+ "hidden_size": 2048,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 5504,
13
+ "max_position_embeddings": 32768,
14
+ "max_window_layers": 21,
15
+ "model_type": "qwen2",
16
+ "num_attention_heads": 16,
17
+ "num_hidden_layers": 24,
18
+ "num_key_value_heads": 16,
19
+ "rms_norm_eps": 1e-06,
20
+ "rope_theta": 1000000.0,
21
+ "sliding_window": 32768,
22
+ "tie_word_embeddings": false,
23
+ "torch_dtype": "float16",
24
+ "transformers_version": "4.39.2",
25
+ "use_cache": true,
26
+ "use_sliding_window": false,
27
+ "vocab_size": 151936
28
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 151643,
4
+ "eos_token_id": 151643,
5
+ "transformers_version": "4.39.2"
6
+ }
merges.txt ADDED
The diff for this file is too large to render. See raw diff
 
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf63156d4f8b23027095236ec92105dbbc6f7f63ea94eec5f5a9319f1ad668e4
3
+ size 1657693952
quant_strategy.json ADDED
@@ -0,0 +1,2569 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "measurement": {
3
+ "model.layers.0": {
4
+ "accuracy": 0.7506313323974609,
5
+ "total_bits": 111884816,
6
+ "q_proj": {
7
+ "group_size": {
8
+ "4": 128,
9
+ "2": 128
10
+ },
11
+ "bits": [
12
+ 4,
13
+ 2
14
+ ],
15
+ "bits_prop": [
16
+ 0.01,
17
+ 0.99
18
+ ],
19
+ "scale_bits": 4
20
+ },
21
+ "k_proj": {
22
+ "group_size": {
23
+ "4": 128,
24
+ "2": 128
25
+ },
26
+ "bits": [
27
+ 4,
28
+ 2
29
+ ],
30
+ "bits_prop": [
31
+ 0.01,
32
+ 0.99
33
+ ],
34
+ "scale_bits": 4
35
+ },
36
+ "v_proj": {
37
+ "group_size": {
38
+ "4": 128,
39
+ "2": 128
40
+ },
41
+ "bits": [
42
+ 4,
43
+ 2
44
+ ],
45
+ "bits_prop": [
46
+ 0.1,
47
+ 0.9
48
+ ],
49
+ "scale_bits": 4
50
+ },
51
+ "o_proj": {
52
+ "group_size": {
53
+ "4": 128,
54
+ "2": 128
55
+ },
56
+ "bits": [
57
+ 4,
58
+ 2
59
+ ],
60
+ "bits_prop": [
61
+ 0.01,
62
+ 0.99
63
+ ],
64
+ "scale_bits": 4
65
+ },
66
+ "up_proj": {
67
+ "group_size": {
68
+ "4": 128,
69
+ "2": 128
70
+ },
71
+ "bits": [
72
+ 4,
73
+ 2
74
+ ],
75
+ "bits_prop": [
76
+ 0.01,
77
+ 0.99
78
+ ],
79
+ "scale_bits": 4
80
+ },
81
+ "gate_proj": {
82
+ "group_size": {
83
+ "4": 128,
84
+ "2": 128
85
+ },
86
+ "bits": [
87
+ 4,
88
+ 2
89
+ ],
90
+ "bits_prop": [
91
+ 0.01,
92
+ 0.99
93
+ ],
94
+ "scale_bits": 4
95
+ },
96
+ "down_proj": {
97
+ "group_size": {
98
+ "4": 128,
99
+ "2": 128
100
+ },
101
+ "bits": [
102
+ 4,
103
+ 2
104
+ ],
105
+ "bits_prop": [
106
+ 0.15,
107
+ 0.85
108
+ ],
109
+ "scale_bits": 4
110
+ }
111
+ },
112
+ "model.layers.1": {
113
+ "accuracy": 0.8626322746276855,
114
+ "total_bits": 108214800,
115
+ "q_proj": {
116
+ "group_size": {
117
+ "4": 128,
118
+ "2": 128
119
+ },
120
+ "bits": [
121
+ 4,
122
+ 2
123
+ ],
124
+ "bits_prop": [
125
+ 0.01,
126
+ 0.99
127
+ ],
128
+ "scale_bits": 4
129
+ },
130
+ "k_proj": {
131
+ "group_size": {
132
+ "4": 128,
133
+ "2": 128
134
+ },
135
+ "bits": [
136
+ 4,
137
+ 2
138
+ ],
139
+ "bits_prop": [
140
+ 0.01,
141
+ 0.99
142
+ ],
143
+ "scale_bits": 4
144
+ },
145
+ "v_proj": {
146
+ "group_size": {
147
+ "4": 128,
148
+ "2": 128
149
+ },
150
+ "bits": [
151
+ 4,
152
+ 2
153
+ ],
154
+ "bits_prop": [
155
+ 0.05,
156
+ 0.95
157
+ ],
158
+ "scale_bits": 4
159
+ },
160
+ "o_proj": {
161
+ "group_size": {
162
+ "4": 128,
163
+ "2": 128
164
+ },
165
+ "bits": [
166
+ 4,
167
+ 2
168
+ ],
169
+ "bits_prop": [
170
+ 0.02,
171
+ 0.98
172
+ ],
173
+ "scale_bits": 4
174
+ },
175
+ "up_proj": {
176
+ "group_size": {
177
+ "4": 128,
178
+ "2": 128
179
+ },
180
+ "bits": [
181
+ 4,
182
+ 2
183
+ ],
184
+ "bits_prop": [
185
+ 0.01,
186
+ 0.99
187
+ ],
188
+ "scale_bits": 4
189
+ },
190
+ "gate_proj": {
191
+ "group_size": {
192
+ "4": 128,
193
+ "2": 128
194
+ },
195
+ "bits": [
196
+ 4,
197
+ 2
198
+ ],
199
+ "bits_prop": [
200
+ 0.01,
201
+ 0.99
202
+ ],
203
+ "scale_bits": 4
204
+ },
205
+ "down_proj": {
206
+ "group_size": {
207
+ "4": 128,
208
+ "2": 128
209
+ },
210
+ "bits": [
211
+ 4,
212
+ 2
213
+ ],
214
+ "bits_prop": [
215
+ 0.02,
216
+ 0.98
217
+ ],
218
+ "scale_bits": 4
219
+ }
220
+ },
221
+ "model.layers.2": {
222
+ "accuracy": 0.8636598587036133,
223
+ "total_bits": 108214800,
224
+ "q_proj": {
225
+ "group_size": {
226
+ "4": 128,
227
+ "2": 128
228
+ },
229
+ "bits": [
230
+ 4,
231
+ 2
232
+ ],
233
+ "bits_prop": [
234
+ 0.01,
235
+ 0.99
236
+ ],
237
+ "scale_bits": 4
238
+ },
239
+ "k_proj": {
240
+ "group_size": {
241
+ "4": 128,
242
+ "2": 128
243
+ },
244
+ "bits": [
245
+ 4,
246
+ 2
247
+ ],
248
+ "bits_prop": [
249
+ 0.01,
250
+ 0.99
251
+ ],
252
+ "scale_bits": 4
253
+ },
254
+ "v_proj": {
255
+ "group_size": {
256
+ "4": 128,
257
+ "2": 128
258
+ },
259
+ "bits": [
260
+ 4,
261
+ 2
262
+ ],
263
+ "bits_prop": [
264
+ 0.05,
265
+ 0.95
266
+ ],
267
+ "scale_bits": 4
268
+ },
269
+ "o_proj": {
270
+ "group_size": {
271
+ "4": 128,
272
+ "2": 128
273
+ },
274
+ "bits": [
275
+ 4,
276
+ 2
277
+ ],
278
+ "bits_prop": [
279
+ 0.02,
280
+ 0.98
281
+ ],
282
+ "scale_bits": 4
283
+ },
284
+ "up_proj": {
285
+ "group_size": {
286
+ "4": 128,
287
+ "2": 128
288
+ },
289
+ "bits": [
290
+ 4,
291
+ 2
292
+ ],
293
+ "bits_prop": [
294
+ 0.01,
295
+ 0.99
296
+ ],
297
+ "scale_bits": 4
298
+ },
299
+ "gate_proj": {
300
+ "group_size": {
301
+ "4": 128,
302
+ "2": 128
303
+ },
304
+ "bits": [
305
+ 4,
306
+ 2
307
+ ],
308
+ "bits_prop": [
309
+ 0.01,
310
+ 0.99
311
+ ],
312
+ "scale_bits": 4
313
+ },
314
+ "down_proj": {
315
+ "group_size": {
316
+ "4": 128,
317
+ "2": 128
318
+ },
319
+ "bits": [
320
+ 4,
321
+ 2
322
+ ],
323
+ "bits_prop": [
324
+ 0.02,
325
+ 0.98
326
+ ],
327
+ "scale_bits": 4
328
+ }
329
+ },
330
+ "model.layers.3": {
331
+ "accuracy": 0.8567333221435547,
332
+ "total_bits": 109263376,
333
+ "q_proj": {
334
+ "group_size": {
335
+ "4": 128,
336
+ "2": 128
337
+ },
338
+ "bits": [
339
+ 4,
340
+ 2
341
+ ],
342
+ "bits_prop": [
343
+ 0.05,
344
+ 0.95
345
+ ],
346
+ "scale_bits": 4
347
+ },
348
+ "k_proj": {
349
+ "group_size": {
350
+ "4": 128,
351
+ "2": 128
352
+ },
353
+ "bits": [
354
+ 4,
355
+ 2
356
+ ],
357
+ "bits_prop": [
358
+ 0.05,
359
+ 0.95
360
+ ],
361
+ "scale_bits": 4
362
+ },
363
+ "v_proj": {
364
+ "group_size": {
365
+ "4": 128,
366
+ "2": 128
367
+ },
368
+ "bits": [
369
+ 4,
370
+ 2
371
+ ],
372
+ "bits_prop": [
373
+ 0.05,
374
+ 0.95
375
+ ],
376
+ "scale_bits": 4
377
+ },
378
+ "o_proj": {
379
+ "group_size": {
380
+ "4": 128,
381
+ "2": 128
382
+ },
383
+ "bits": [
384
+ 4,
385
+ 2
386
+ ],
387
+ "bits_prop": [
388
+ 0.05,
389
+ 0.95
390
+ ],
391
+ "scale_bits": 4
392
+ },
393
+ "up_proj": {
394
+ "group_size": {
395
+ "4": 128,
396
+ "2": 128
397
+ },
398
+ "bits": [
399
+ 4,
400
+ 2
401
+ ],
402
+ "bits_prop": [
403
+ 0.05,
404
+ 0.95
405
+ ],
406
+ "scale_bits": 4
407
+ },
408
+ "gate_proj": {
409
+ "group_size": {
410
+ "4": 128,
411
+ "2": 128
412
+ },
413
+ "bits": [
414
+ 4,
415
+ 2
416
+ ],
417
+ "bits_prop": [
418
+ 0.05,
419
+ 0.95
420
+ ],
421
+ "scale_bits": 4
422
+ },
423
+ "down_proj": {
424
+ "group_size": {
425
+ "4": 128,
426
+ "2": 128
427
+ },
428
+ "bits": [
429
+ 4,
430
+ 2
431
+ ],
432
+ "bits_prop": [
433
+ 0.05,
434
+ 0.95
435
+ ],
436
+ "scale_bits": 4
437
+ }
438
+ },
439
+ "model.layers.4": {
440
+ "accuracy": 0.8393054008483887,
441
+ "total_bits": 109263376,
442
+ "q_proj": {
443
+ "group_size": {
444
+ "4": 128,
445
+ "2": 128
446
+ },
447
+ "bits": [
448
+ 4,
449
+ 2
450
+ ],
451
+ "bits_prop": [
452
+ 0.05,
453
+ 0.95
454
+ ],
455
+ "scale_bits": 4
456
+ },
457
+ "k_proj": {
458
+ "group_size": {
459
+ "4": 128,
460
+ "2": 128
461
+ },
462
+ "bits": [
463
+ 4,
464
+ 2
465
+ ],
466
+ "bits_prop": [
467
+ 0.05,
468
+ 0.95
469
+ ],
470
+ "scale_bits": 4
471
+ },
472
+ "v_proj": {
473
+ "group_size": {
474
+ "4": 128,
475
+ "2": 128
476
+ },
477
+ "bits": [
478
+ 4,
479
+ 2
480
+ ],
481
+ "bits_prop": [
482
+ 0.05,
483
+ 0.95
484
+ ],
485
+ "scale_bits": 4
486
+ },
487
+ "o_proj": {
488
+ "group_size": {
489
+ "4": 128,
490
+ "2": 128
491
+ },
492
+ "bits": [
493
+ 4,
494
+ 2
495
+ ],
496
+ "bits_prop": [
497
+ 0.05,
498
+ 0.95
499
+ ],
500
+ "scale_bits": 4
501
+ },
502
+ "up_proj": {
503
+ "group_size": {
504
+ "4": 128,
505
+ "2": 128
506
+ },
507
+ "bits": [
508
+ 4,
509
+ 2
510
+ ],
511
+ "bits_prop": [
512
+ 0.05,
513
+ 0.95
514
+ ],
515
+ "scale_bits": 4
516
+ },
517
+ "gate_proj": {
518
+ "group_size": {
519
+ "4": 128,
520
+ "2": 128
521
+ },
522
+ "bits": [
523
+ 4,
524
+ 2
525
+ ],
526
+ "bits_prop": [
527
+ 0.05,
528
+ 0.95
529
+ ],
530
+ "scale_bits": 4
531
+ },
532
+ "down_proj": {
533
+ "group_size": {
534
+ "4": 128,
535
+ "2": 128
536
+ },
537
+ "bits": [
538
+ 4,
539
+ 2
540
+ ],
541
+ "bits_prop": [
542
+ 0.05,
543
+ 0.95
544
+ ],
545
+ "scale_bits": 4
546
+ }
547
+ },
548
+ "model.layers.5": {
549
+ "accuracy": 0.8224668502807617,
550
+ "total_bits": 112409104,
551
+ "q_proj": {
552
+ "group_size": {
553
+ "4": 128,
554
+ "2": 128
555
+ },
556
+ "bits": [
557
+ 4,
558
+ 2
559
+ ],
560
+ "bits_prop": [
561
+ 0.01,
562
+ 0.99
563
+ ],
564
+ "scale_bits": 4
565
+ },
566
+ "k_proj": {
567
+ "group_size": {
568
+ "4": 128,
569
+ "2": 128
570
+ },
571
+ "bits": [
572
+ 4,
573
+ 2
574
+ ],
575
+ "bits_prop": [
576
+ 0.01,
577
+ 0.99
578
+ ],
579
+ "scale_bits": 4
580
+ },
581
+ "v_proj": {
582
+ "group_size": {
583
+ "4": 128,
584
+ "2": 128
585
+ },
586
+ "bits": [
587
+ 4,
588
+ 2
589
+ ],
590
+ "bits_prop": [
591
+ 0.15,
592
+ 0.85
593
+ ],
594
+ "scale_bits": 4
595
+ },
596
+ "o_proj": {
597
+ "group_size": {
598
+ "4": 128,
599
+ "2": 128
600
+ },
601
+ "bits": [
602
+ 4,
603
+ 2
604
+ ],
605
+ "bits_prop": [
606
+ 0.01,
607
+ 0.99
608
+ ],
609
+ "scale_bits": 4
610
+ },
611
+ "up_proj": {
612
+ "group_size": {
613
+ "4": 128,
614
+ "2": 128
615
+ },
616
+ "bits": [
617
+ 4,
618
+ 2
619
+ ],
620
+ "bits_prop": [
621
+ 0.01,
622
+ 0.99
623
+ ],
624
+ "scale_bits": 4
625
+ },
626
+ "gate_proj": {
627
+ "group_size": {
628
+ "4": 128,
629
+ "2": 128
630
+ },
631
+ "bits": [
632
+ 4,
633
+ 2
634
+ ],
635
+ "bits_prop": [
636
+ 0.01,
637
+ 0.99
638
+ ],
639
+ "scale_bits": 4
640
+ },
641
+ "down_proj": {
642
+ "group_size": {
643
+ "4": 128,
644
+ "2": 128
645
+ },
646
+ "bits": [
647
+ 4,
648
+ 2
649
+ ],
650
+ "bits_prop": [
651
+ 0.15,
652
+ 0.85
653
+ ],
654
+ "scale_bits": 4
655
+ }
656
+ },
657
+ "model.layers.6": {
658
+ "accuracy": 0.9044299125671387,
659
+ "total_bits": 108214800,
660
+ "q_proj": {
661
+ "group_size": {
662
+ "4": 128,
663
+ "2": 128
664
+ },
665
+ "bits": [
666
+ 4,
667
+ 2
668
+ ],
669
+ "bits_prop": [
670
+ 0.01,
671
+ 0.99
672
+ ],
673
+ "scale_bits": 4
674
+ },
675
+ "k_proj": {
676
+ "group_size": {
677
+ "4": 128,
678
+ "2": 128
679
+ },
680
+ "bits": [
681
+ 4,
682
+ 2
683
+ ],
684
+ "bits_prop": [
685
+ 0.01,
686
+ 0.99
687
+ ],
688
+ "scale_bits": 4
689
+ },
690
+ "v_proj": {
691
+ "group_size": {
692
+ "4": 128,
693
+ "2": 128
694
+ },
695
+ "bits": [
696
+ 4,
697
+ 2
698
+ ],
699
+ "bits_prop": [
700
+ 0.05,
701
+ 0.95
702
+ ],
703
+ "scale_bits": 4
704
+ },
705
+ "o_proj": {
706
+ "group_size": {
707
+ "4": 128,
708
+ "2": 128
709
+ },
710
+ "bits": [
711
+ 4,
712
+ 2
713
+ ],
714
+ "bits_prop": [
715
+ 0.02,
716
+ 0.98
717
+ ],
718
+ "scale_bits": 4
719
+ },
720
+ "up_proj": {
721
+ "group_size": {
722
+ "4": 128,
723
+ "2": 128
724
+ },
725
+ "bits": [
726
+ 4,
727
+ 2
728
+ ],
729
+ "bits_prop": [
730
+ 0.01,
731
+ 0.99
732
+ ],
733
+ "scale_bits": 4
734
+ },
735
+ "gate_proj": {
736
+ "group_size": {
737
+ "4": 128,
738
+ "2": 128
739
+ },
740
+ "bits": [
741
+ 4,
742
+ 2
743
+ ],
744
+ "bits_prop": [
745
+ 0.01,
746
+ 0.99
747
+ ],
748
+ "scale_bits": 4
749
+ },
750
+ "down_proj": {
751
+ "group_size": {
752
+ "4": 128,
753
+ "2": 128
754
+ },
755
+ "bits": [
756
+ 4,
757
+ 2
758
+ ],
759
+ "bits_prop": [
760
+ 0.02,
761
+ 0.98
762
+ ],
763
+ "scale_bits": 4
764
+ }
765
+ },
766
+ "model.layers.7": {
767
+ "accuracy": 0.912956953048706,
768
+ "total_bits": 108214800,
769
+ "q_proj": {
770
+ "group_size": {
771
+ "4": 128,
772
+ "2": 128
773
+ },
774
+ "bits": [
775
+ 4,
776
+ 2
777
+ ],
778
+ "bits_prop": [
779
+ 0.01,
780
+ 0.99
781
+ ],
782
+ "scale_bits": 4
783
+ },
784
+ "k_proj": {
785
+ "group_size": {
786
+ "4": 128,
787
+ "2": 128
788
+ },
789
+ "bits": [
790
+ 4,
791
+ 2
792
+ ],
793
+ "bits_prop": [
794
+ 0.01,
795
+ 0.99
796
+ ],
797
+ "scale_bits": 4
798
+ },
799
+ "v_proj": {
800
+ "group_size": {
801
+ "4": 128,
802
+ "2": 128
803
+ },
804
+ "bits": [
805
+ 4,
806
+ 2
807
+ ],
808
+ "bits_prop": [
809
+ 0.05,
810
+ 0.95
811
+ ],
812
+ "scale_bits": 4
813
+ },
814
+ "o_proj": {
815
+ "group_size": {
816
+ "4": 128,
817
+ "2": 128
818
+ },
819
+ "bits": [
820
+ 4,
821
+ 2
822
+ ],
823
+ "bits_prop": [
824
+ 0.02,
825
+ 0.98
826
+ ],
827
+ "scale_bits": 4
828
+ },
829
+ "up_proj": {
830
+ "group_size": {
831
+ "4": 128,
832
+ "2": 128
833
+ },
834
+ "bits": [
835
+ 4,
836
+ 2
837
+ ],
838
+ "bits_prop": [
839
+ 0.01,
840
+ 0.99
841
+ ],
842
+ "scale_bits": 4
843
+ },
844
+ "gate_proj": {
845
+ "group_size": {
846
+ "4": 128,
847
+ "2": 128
848
+ },
849
+ "bits": [
850
+ 4,
851
+ 2
852
+ ],
853
+ "bits_prop": [
854
+ 0.01,
855
+ 0.99
856
+ ],
857
+ "scale_bits": 4
858
+ },
859
+ "down_proj": {
860
+ "group_size": {
861
+ "4": 128,
862
+ "2": 128
863
+ },
864
+ "bits": [
865
+ 4,
866
+ 2
867
+ ],
868
+ "bits_prop": [
869
+ 0.02,
870
+ 0.98
871
+ ],
872
+ "scale_bits": 4
873
+ }
874
+ },
875
+ "model.layers.8": {
876
+ "accuracy": 0.9059407711029053,
877
+ "total_bits": 108214800,
878
+ "q_proj": {
879
+ "group_size": {
880
+ "4": 128,
881
+ "2": 128
882
+ },
883
+ "bits": [
884
+ 4,
885
+ 2
886
+ ],
887
+ "bits_prop": [
888
+ 0.01,
889
+ 0.99
890
+ ],
891
+ "scale_bits": 4
892
+ },
893
+ "k_proj": {
894
+ "group_size": {
895
+ "4": 128,
896
+ "2": 128
897
+ },
898
+ "bits": [
899
+ 4,
900
+ 2
901
+ ],
902
+ "bits_prop": [
903
+ 0.01,
904
+ 0.99
905
+ ],
906
+ "scale_bits": 4
907
+ },
908
+ "v_proj": {
909
+ "group_size": {
910
+ "4": 128,
911
+ "2": 128
912
+ },
913
+ "bits": [
914
+ 4,
915
+ 2
916
+ ],
917
+ "bits_prop": [
918
+ 0.05,
919
+ 0.95
920
+ ],
921
+ "scale_bits": 4
922
+ },
923
+ "o_proj": {
924
+ "group_size": {
925
+ "4": 128,
926
+ "2": 128
927
+ },
928
+ "bits": [
929
+ 4,
930
+ 2
931
+ ],
932
+ "bits_prop": [
933
+ 0.02,
934
+ 0.98
935
+ ],
936
+ "scale_bits": 4
937
+ },
938
+ "up_proj": {
939
+ "group_size": {
940
+ "4": 128,
941
+ "2": 128
942
+ },
943
+ "bits": [
944
+ 4,
945
+ 2
946
+ ],
947
+ "bits_prop": [
948
+ 0.01,
949
+ 0.99
950
+ ],
951
+ "scale_bits": 4
952
+ },
953
+ "gate_proj": {
954
+ "group_size": {
955
+ "4": 128,
956
+ "2": 128
957
+ },
958
+ "bits": [
959
+ 4,
960
+ 2
961
+ ],
962
+ "bits_prop": [
963
+ 0.01,
964
+ 0.99
965
+ ],
966
+ "scale_bits": 4
967
+ },
968
+ "down_proj": {
969
+ "group_size": {
970
+ "4": 128,
971
+ "2": 128
972
+ },
973
+ "bits": [
974
+ 4,
975
+ 2
976
+ ],
977
+ "bits_prop": [
978
+ 0.02,
979
+ 0.98
980
+ ],
981
+ "scale_bits": 4
982
+ }
983
+ },
984
+ "model.layers.9": {
985
+ "accuracy": 0.9044547080993652,
986
+ "total_bits": 108214800,
987
+ "q_proj": {
988
+ "group_size": {
989
+ "4": 128,
990
+ "2": 128
991
+ },
992
+ "bits": [
993
+ 4,
994
+ 2
995
+ ],
996
+ "bits_prop": [
997
+ 0.01,
998
+ 0.99
999
+ ],
1000
+ "scale_bits": 4
1001
+ },
1002
+ "k_proj": {
1003
+ "group_size": {
1004
+ "4": 128,
1005
+ "2": 128
1006
+ },
1007
+ "bits": [
1008
+ 4,
1009
+ 2
1010
+ ],
1011
+ "bits_prop": [
1012
+ 0.01,
1013
+ 0.99
1014
+ ],
1015
+ "scale_bits": 4
1016
+ },
1017
+ "v_proj": {
1018
+ "group_size": {
1019
+ "4": 128,
1020
+ "2": 128
1021
+ },
1022
+ "bits": [
1023
+ 4,
1024
+ 2
1025
+ ],
1026
+ "bits_prop": [
1027
+ 0.05,
1028
+ 0.95
1029
+ ],
1030
+ "scale_bits": 4
1031
+ },
1032
+ "o_proj": {
1033
+ "group_size": {
1034
+ "4": 128,
1035
+ "2": 128
1036
+ },
1037
+ "bits": [
1038
+ 4,
1039
+ 2
1040
+ ],
1041
+ "bits_prop": [
1042
+ 0.02,
1043
+ 0.98
1044
+ ],
1045
+ "scale_bits": 4
1046
+ },
1047
+ "up_proj": {
1048
+ "group_size": {
1049
+ "4": 128,
1050
+ "2": 128
1051
+ },
1052
+ "bits": [
1053
+ 4,
1054
+ 2
1055
+ ],
1056
+ "bits_prop": [
1057
+ 0.01,
1058
+ 0.99
1059
+ ],
1060
+ "scale_bits": 4
1061
+ },
1062
+ "gate_proj": {
1063
+ "group_size": {
1064
+ "4": 128,
1065
+ "2": 128
1066
+ },
1067
+ "bits": [
1068
+ 4,
1069
+ 2
1070
+ ],
1071
+ "bits_prop": [
1072
+ 0.01,
1073
+ 0.99
1074
+ ],
1075
+ "scale_bits": 4
1076
+ },
1077
+ "down_proj": {
1078
+ "group_size": {
1079
+ "4": 128,
1080
+ "2": 128
1081
+ },
1082
+ "bits": [
1083
+ 4,
1084
+ 2
1085
+ ],
1086
+ "bits_prop": [
1087
+ 0.02,
1088
+ 0.98
1089
+ ],
1090
+ "scale_bits": 4
1091
+ }
1092
+ },
1093
+ "model.layers.10": {
1094
+ "accuracy": 0.8973643779754639,
1095
+ "total_bits": 109263376,
1096
+ "q_proj": {
1097
+ "group_size": {
1098
+ "4": 128,
1099
+ "2": 128
1100
+ },
1101
+ "bits": [
1102
+ 4,
1103
+ 2
1104
+ ],
1105
+ "bits_prop": [
1106
+ 0.05,
1107
+ 0.95
1108
+ ],
1109
+ "scale_bits": 4
1110
+ },
1111
+ "k_proj": {
1112
+ "group_size": {
1113
+ "4": 128,
1114
+ "2": 128
1115
+ },
1116
+ "bits": [
1117
+ 4,
1118
+ 2
1119
+ ],
1120
+ "bits_prop": [
1121
+ 0.05,
1122
+ 0.95
1123
+ ],
1124
+ "scale_bits": 4
1125
+ },
1126
+ "v_proj": {
1127
+ "group_size": {
1128
+ "4": 128,
1129
+ "2": 128
1130
+ },
1131
+ "bits": [
1132
+ 4,
1133
+ 2
1134
+ ],
1135
+ "bits_prop": [
1136
+ 0.05,
1137
+ 0.95
1138
+ ],
1139
+ "scale_bits": 4
1140
+ },
1141
+ "o_proj": {
1142
+ "group_size": {
1143
+ "4": 128,
1144
+ "2": 128
1145
+ },
1146
+ "bits": [
1147
+ 4,
1148
+ 2
1149
+ ],
1150
+ "bits_prop": [
1151
+ 0.05,
1152
+ 0.95
1153
+ ],
1154
+ "scale_bits": 4
1155
+ },
1156
+ "up_proj": {
1157
+ "group_size": {
1158
+ "4": 128,
1159
+ "2": 128
1160
+ },
1161
+ "bits": [
1162
+ 4,
1163
+ 2
1164
+ ],
1165
+ "bits_prop": [
1166
+ 0.05,
1167
+ 0.95
1168
+ ],
1169
+ "scale_bits": 4
1170
+ },
1171
+ "gate_proj": {
1172
+ "group_size": {
1173
+ "4": 128,
1174
+ "2": 128
1175
+ },
1176
+ "bits": [
1177
+ 4,
1178
+ 2
1179
+ ],
1180
+ "bits_prop": [
1181
+ 0.05,
1182
+ 0.95
1183
+ ],
1184
+ "scale_bits": 4
1185
+ },
1186
+ "down_proj": {
1187
+ "group_size": {
1188
+ "4": 128,
1189
+ "2": 128
1190
+ },
1191
+ "bits": [
1192
+ 4,
1193
+ 2
1194
+ ],
1195
+ "bits_prop": [
1196
+ 0.05,
1197
+ 0.95
1198
+ ],
1199
+ "scale_bits": 4
1200
+ }
1201
+ },
1202
+ "model.layers.11": {
1203
+ "accuracy": 0.899599552154541,
1204
+ "total_bits": 112409104,
1205
+ "q_proj": {
1206
+ "group_size": {
1207
+ "4": 128,
1208
+ "2": 128
1209
+ },
1210
+ "bits": [
1211
+ 4,
1212
+ 2
1213
+ ],
1214
+ "bits_prop": [
1215
+ 0.01,
1216
+ 0.99
1217
+ ],
1218
+ "scale_bits": 4
1219
+ },
1220
+ "k_proj": {
1221
+ "group_size": {
1222
+ "4": 128,
1223
+ "2": 128
1224
+ },
1225
+ "bits": [
1226
+ 4,
1227
+ 2
1228
+ ],
1229
+ "bits_prop": [
1230
+ 0.01,
1231
+ 0.99
1232
+ ],
1233
+ "scale_bits": 4
1234
+ },
1235
+ "v_proj": {
1236
+ "group_size": {
1237
+ "4": 128,
1238
+ "2": 128
1239
+ },
1240
+ "bits": [
1241
+ 4,
1242
+ 2
1243
+ ],
1244
+ "bits_prop": [
1245
+ 0.15,
1246
+ 0.85
1247
+ ],
1248
+ "scale_bits": 4
1249
+ },
1250
+ "o_proj": {
1251
+ "group_size": {
1252
+ "4": 128,
1253
+ "2": 128
1254
+ },
1255
+ "bits": [
1256
+ 4,
1257
+ 2
1258
+ ],
1259
+ "bits_prop": [
1260
+ 0.01,
1261
+ 0.99
1262
+ ],
1263
+ "scale_bits": 4
1264
+ },
1265
+ "up_proj": {
1266
+ "group_size": {
1267
+ "4": 128,
1268
+ "2": 128
1269
+ },
1270
+ "bits": [
1271
+ 4,
1272
+ 2
1273
+ ],
1274
+ "bits_prop": [
1275
+ 0.01,
1276
+ 0.99
1277
+ ],
1278
+ "scale_bits": 4
1279
+ },
1280
+ "gate_proj": {
1281
+ "group_size": {
1282
+ "4": 128,
1283
+ "2": 128
1284
+ },
1285
+ "bits": [
1286
+ 4,
1287
+ 2
1288
+ ],
1289
+ "bits_prop": [
1290
+ 0.01,
1291
+ 0.99
1292
+ ],
1293
+ "scale_bits": 4
1294
+ },
1295
+ "down_proj": {
1296
+ "group_size": {
1297
+ "4": 128,
1298
+ "2": 128
1299
+ },
1300
+ "bits": [
1301
+ 4,
1302
+ 2
1303
+ ],
1304
+ "bits_prop": [
1305
+ 0.15,
1306
+ 0.85
1307
+ ],
1308
+ "scale_bits": 4
1309
+ }
1310
+ },
1311
+ "model.layers.12": {
1312
+ "accuracy": 0.8899784088134766,
1313
+ "total_bits": 112409104,
1314
+ "q_proj": {
1315
+ "group_size": {
1316
+ "4": 128,
1317
+ "2": 128
1318
+ },
1319
+ "bits": [
1320
+ 4,
1321
+ 2
1322
+ ],
1323
+ "bits_prop": [
1324
+ 0.01,
1325
+ 0.99
1326
+ ],
1327
+ "scale_bits": 4
1328
+ },
1329
+ "k_proj": {
1330
+ "group_size": {
1331
+ "4": 128,
1332
+ "2": 128
1333
+ },
1334
+ "bits": [
1335
+ 4,
1336
+ 2
1337
+ ],
1338
+ "bits_prop": [
1339
+ 0.01,
1340
+ 0.99
1341
+ ],
1342
+ "scale_bits": 4
1343
+ },
1344
+ "v_proj": {
1345
+ "group_size": {
1346
+ "4": 128,
1347
+ "2": 128
1348
+ },
1349
+ "bits": [
1350
+ 4,
1351
+ 2
1352
+ ],
1353
+ "bits_prop": [
1354
+ 0.15,
1355
+ 0.85
1356
+ ],
1357
+ "scale_bits": 4
1358
+ },
1359
+ "o_proj": {
1360
+ "group_size": {
1361
+ "4": 128,
1362
+ "2": 128
1363
+ },
1364
+ "bits": [
1365
+ 4,
1366
+ 2
1367
+ ],
1368
+ "bits_prop": [
1369
+ 0.01,
1370
+ 0.99
1371
+ ],
1372
+ "scale_bits": 4
1373
+ },
1374
+ "up_proj": {
1375
+ "group_size": {
1376
+ "4": 128,
1377
+ "2": 128
1378
+ },
1379
+ "bits": [
1380
+ 4,
1381
+ 2
1382
+ ],
1383
+ "bits_prop": [
1384
+ 0.01,
1385
+ 0.99
1386
+ ],
1387
+ "scale_bits": 4
1388
+ },
1389
+ "gate_proj": {
1390
+ "group_size": {
1391
+ "4": 128,
1392
+ "2": 128
1393
+ },
1394
+ "bits": [
1395
+ 4,
1396
+ 2
1397
+ ],
1398
+ "bits_prop": [
1399
+ 0.01,
1400
+ 0.99
1401
+ ],
1402
+ "scale_bits": 4
1403
+ },
1404
+ "down_proj": {
1405
+ "group_size": {
1406
+ "4": 128,
1407
+ "2": 128
1408
+ },
1409
+ "bits": [
1410
+ 4,
1411
+ 2
1412
+ ],
1413
+ "bits_prop": [
1414
+ 0.15,
1415
+ 0.85
1416
+ ],
1417
+ "scale_bits": 4
1418
+ }
1419
+ },
1420
+ "model.layers.13": {
1421
+ "accuracy": 0.8856725692749023,
1422
+ "total_bits": 112409104,
1423
+ "q_proj": {
1424
+ "group_size": {
1425
+ "4": 128,
1426
+ "2": 128
1427
+ },
1428
+ "bits": [
1429
+ 4,
1430
+ 2
1431
+ ],
1432
+ "bits_prop": [
1433
+ 0.01,
1434
+ 0.99
1435
+ ],
1436
+ "scale_bits": 4
1437
+ },
1438
+ "k_proj": {
1439
+ "group_size": {
1440
+ "4": 128,
1441
+ "2": 128
1442
+ },
1443
+ "bits": [
1444
+ 4,
1445
+ 2
1446
+ ],
1447
+ "bits_prop": [
1448
+ 0.01,
1449
+ 0.99
1450
+ ],
1451
+ "scale_bits": 4
1452
+ },
1453
+ "v_proj": {
1454
+ "group_size": {
1455
+ "4": 128,
1456
+ "2": 128
1457
+ },
1458
+ "bits": [
1459
+ 4,
1460
+ 2
1461
+ ],
1462
+ "bits_prop": [
1463
+ 0.15,
1464
+ 0.85
1465
+ ],
1466
+ "scale_bits": 4
1467
+ },
1468
+ "o_proj": {
1469
+ "group_size": {
1470
+ "4": 128,
1471
+ "2": 128
1472
+ },
1473
+ "bits": [
1474
+ 4,
1475
+ 2
1476
+ ],
1477
+ "bits_prop": [
1478
+ 0.01,
1479
+ 0.99
1480
+ ],
1481
+ "scale_bits": 4
1482
+ },
1483
+ "up_proj": {
1484
+ "group_size": {
1485
+ "4": 128,
1486
+ "2": 128
1487
+ },
1488
+ "bits": [
1489
+ 4,
1490
+ 2
1491
+ ],
1492
+ "bits_prop": [
1493
+ 0.01,
1494
+ 0.99
1495
+ ],
1496
+ "scale_bits": 4
1497
+ },
1498
+ "gate_proj": {
1499
+ "group_size": {
1500
+ "4": 128,
1501
+ "2": 128
1502
+ },
1503
+ "bits": [
1504
+ 4,
1505
+ 2
1506
+ ],
1507
+ "bits_prop": [
1508
+ 0.01,
1509
+ 0.99
1510
+ ],
1511
+ "scale_bits": 4
1512
+ },
1513
+ "down_proj": {
1514
+ "group_size": {
1515
+ "4": 128,
1516
+ "2": 128
1517
+ },
1518
+ "bits": [
1519
+ 4,
1520
+ 2
1521
+ ],
1522
+ "bits_prop": [
1523
+ 0.15,
1524
+ 0.85
1525
+ ],
1526
+ "scale_bits": 4
1527
+ }
1528
+ },
1529
+ "model.layers.14": {
1530
+ "accuracy": 0.8767843246459961,
1531
+ "total_bits": 112409104,
1532
+ "q_proj": {
1533
+ "group_size": {
1534
+ "4": 128,
1535
+ "2": 128
1536
+ },
1537
+ "bits": [
1538
+ 4,
1539
+ 2
1540
+ ],
1541
+ "bits_prop": [
1542
+ 0.01,
1543
+ 0.99
1544
+ ],
1545
+ "scale_bits": 4
1546
+ },
1547
+ "k_proj": {
1548
+ "group_size": {
1549
+ "4": 128,
1550
+ "2": 128
1551
+ },
1552
+ "bits": [
1553
+ 4,
1554
+ 2
1555
+ ],
1556
+ "bits_prop": [
1557
+ 0.01,
1558
+ 0.99
1559
+ ],
1560
+ "scale_bits": 4
1561
+ },
1562
+ "v_proj": {
1563
+ "group_size": {
1564
+ "4": 128,
1565
+ "2": 128
1566
+ },
1567
+ "bits": [
1568
+ 4,
1569
+ 2
1570
+ ],
1571
+ "bits_prop": [
1572
+ 0.15,
1573
+ 0.85
1574
+ ],
1575
+ "scale_bits": 4
1576
+ },
1577
+ "o_proj": {
1578
+ "group_size": {
1579
+ "4": 128,
1580
+ "2": 128
1581
+ },
1582
+ "bits": [
1583
+ 4,
1584
+ 2
1585
+ ],
1586
+ "bits_prop": [
1587
+ 0.01,
1588
+ 0.99
1589
+ ],
1590
+ "scale_bits": 4
1591
+ },
1592
+ "up_proj": {
1593
+ "group_size": {
1594
+ "4": 128,
1595
+ "2": 128
1596
+ },
1597
+ "bits": [
1598
+ 4,
1599
+ 2
1600
+ ],
1601
+ "bits_prop": [
1602
+ 0.01,
1603
+ 0.99
1604
+ ],
1605
+ "scale_bits": 4
1606
+ },
1607
+ "gate_proj": {
1608
+ "group_size": {
1609
+ "4": 128,
1610
+ "2": 128
1611
+ },
1612
+ "bits": [
1613
+ 4,
1614
+ 2
1615
+ ],
1616
+ "bits_prop": [
1617
+ 0.01,
1618
+ 0.99
1619
+ ],
1620
+ "scale_bits": 4
1621
+ },
1622
+ "down_proj": {
1623
+ "group_size": {
1624
+ "4": 128,
1625
+ "2": 128
1626
+ },
1627
+ "bits": [
1628
+ 4,
1629
+ 2
1630
+ ],
1631
+ "bits_prop": [
1632
+ 0.15,
1633
+ 0.85
1634
+ ],
1635
+ "scale_bits": 4
1636
+ }
1637
+ },
1638
+ "model.layers.15": {
1639
+ "accuracy": 0.8738350868225098,
1640
+ "total_bits": 112409104,
1641
+ "q_proj": {
1642
+ "group_size": {
1643
+ "4": 128,
1644
+ "2": 128
1645
+ },
1646
+ "bits": [
1647
+ 4,
1648
+ 2
1649
+ ],
1650
+ "bits_prop": [
1651
+ 0.01,
1652
+ 0.99
1653
+ ],
1654
+ "scale_bits": 4
1655
+ },
1656
+ "k_proj": {
1657
+ "group_size": {
1658
+ "4": 128,
1659
+ "2": 128
1660
+ },
1661
+ "bits": [
1662
+ 4,
1663
+ 2
1664
+ ],
1665
+ "bits_prop": [
1666
+ 0.01,
1667
+ 0.99
1668
+ ],
1669
+ "scale_bits": 4
1670
+ },
1671
+ "v_proj": {
1672
+ "group_size": {
1673
+ "4": 128,
1674
+ "2": 128
1675
+ },
1676
+ "bits": [
1677
+ 4,
1678
+ 2
1679
+ ],
1680
+ "bits_prop": [
1681
+ 0.15,
1682
+ 0.85
1683
+ ],
1684
+ "scale_bits": 4
1685
+ },
1686
+ "o_proj": {
1687
+ "group_size": {
1688
+ "4": 128,
1689
+ "2": 128
1690
+ },
1691
+ "bits": [
1692
+ 4,
1693
+ 2
1694
+ ],
1695
+ "bits_prop": [
1696
+ 0.01,
1697
+ 0.99
1698
+ ],
1699
+ "scale_bits": 4
1700
+ },
1701
+ "up_proj": {
1702
+ "group_size": {
1703
+ "4": 128,
1704
+ "2": 128
1705
+ },
1706
+ "bits": [
1707
+ 4,
1708
+ 2
1709
+ ],
1710
+ "bits_prop": [
1711
+ 0.01,
1712
+ 0.99
1713
+ ],
1714
+ "scale_bits": 4
1715
+ },
1716
+ "gate_proj": {
1717
+ "group_size": {
1718
+ "4": 128,
1719
+ "2": 128
1720
+ },
1721
+ "bits": [
1722
+ 4,
1723
+ 2
1724
+ ],
1725
+ "bits_prop": [
1726
+ 0.01,
1727
+ 0.99
1728
+ ],
1729
+ "scale_bits": 4
1730
+ },
1731
+ "down_proj": {
1732
+ "group_size": {
1733
+ "4": 128,
1734
+ "2": 128
1735
+ },
1736
+ "bits": [
1737
+ 4,
1738
+ 2
1739
+ ],
1740
+ "bits_prop": [
1741
+ 0.15,
1742
+ 0.85
1743
+ ],
1744
+ "scale_bits": 4
1745
+ }
1746
+ },
1747
+ "model.layers.16": {
1748
+ "accuracy": 0.8765683174133301,
1749
+ "total_bits": 112409104,
1750
+ "q_proj": {
1751
+ "group_size": {
1752
+ "4": 128,
1753
+ "2": 128
1754
+ },
1755
+ "bits": [
1756
+ 4,
1757
+ 2
1758
+ ],
1759
+ "bits_prop": [
1760
+ 0.01,
1761
+ 0.99
1762
+ ],
1763
+ "scale_bits": 4
1764
+ },
1765
+ "k_proj": {
1766
+ "group_size": {
1767
+ "4": 128,
1768
+ "2": 128
1769
+ },
1770
+ "bits": [
1771
+ 4,
1772
+ 2
1773
+ ],
1774
+ "bits_prop": [
1775
+ 0.01,
1776
+ 0.99
1777
+ ],
1778
+ "scale_bits": 4
1779
+ },
1780
+ "v_proj": {
1781
+ "group_size": {
1782
+ "4": 128,
1783
+ "2": 128
1784
+ },
1785
+ "bits": [
1786
+ 4,
1787
+ 2
1788
+ ],
1789
+ "bits_prop": [
1790
+ 0.15,
1791
+ 0.85
1792
+ ],
1793
+ "scale_bits": 4
1794
+ },
1795
+ "o_proj": {
1796
+ "group_size": {
1797
+ "4": 128,
1798
+ "2": 128
1799
+ },
1800
+ "bits": [
1801
+ 4,
1802
+ 2
1803
+ ],
1804
+ "bits_prop": [
1805
+ 0.01,
1806
+ 0.99
1807
+ ],
1808
+ "scale_bits": 4
1809
+ },
1810
+ "up_proj": {
1811
+ "group_size": {
1812
+ "4": 128,
1813
+ "2": 128
1814
+ },
1815
+ "bits": [
1816
+ 4,
1817
+ 2
1818
+ ],
1819
+ "bits_prop": [
1820
+ 0.01,
1821
+ 0.99
1822
+ ],
1823
+ "scale_bits": 4
1824
+ },
1825
+ "gate_proj": {
1826
+ "group_size": {
1827
+ "4": 128,
1828
+ "2": 128
1829
+ },
1830
+ "bits": [
1831
+ 4,
1832
+ 2
1833
+ ],
1834
+ "bits_prop": [
1835
+ 0.01,
1836
+ 0.99
1837
+ ],
1838
+ "scale_bits": 4
1839
+ },
1840
+ "down_proj": {
1841
+ "group_size": {
1842
+ "4": 128,
1843
+ "2": 128
1844
+ },
1845
+ "bits": [
1846
+ 4,
1847
+ 2
1848
+ ],
1849
+ "bits_prop": [
1850
+ 0.15,
1851
+ 0.85
1852
+ ],
1853
+ "scale_bits": 4
1854
+ }
1855
+ },
1856
+ "model.layers.17": {
1857
+ "accuracy": 0.883051872253418,
1858
+ "total_bits": 120273424,
1859
+ "q_proj": {
1860
+ "group_size": {
1861
+ "4": 128,
1862
+ "2": 128
1863
+ },
1864
+ "bits": [
1865
+ 4,
1866
+ 2
1867
+ ],
1868
+ "bits_prop": [
1869
+ 0.05,
1870
+ 0.95
1871
+ ],
1872
+ "scale_bits": 4
1873
+ },
1874
+ "k_proj": {
1875
+ "group_size": {
1876
+ "4": 128,
1877
+ "2": 128
1878
+ },
1879
+ "bits": [
1880
+ 4,
1881
+ 2
1882
+ ],
1883
+ "bits_prop": [
1884
+ 0.05,
1885
+ 0.95
1886
+ ],
1887
+ "scale_bits": 4
1888
+ },
1889
+ "v_proj": {
1890
+ "group_size": {
1891
+ "4": 128,
1892
+ "2": 128
1893
+ },
1894
+ "bits": [
1895
+ 4,
1896
+ 2
1897
+ ],
1898
+ "bits_prop": [
1899
+ 0.4,
1900
+ 0.6
1901
+ ],
1902
+ "scale_bits": 4
1903
+ },
1904
+ "o_proj": {
1905
+ "group_size": {
1906
+ "4": 128,
1907
+ "2": 128
1908
+ },
1909
+ "bits": [
1910
+ 4,
1911
+ 2
1912
+ ],
1913
+ "bits_prop": [
1914
+ 0.05,
1915
+ 0.95
1916
+ ],
1917
+ "scale_bits": 4
1918
+ },
1919
+ "up_proj": {
1920
+ "group_size": {
1921
+ "4": 128,
1922
+ "2": 128
1923
+ },
1924
+ "bits": [
1925
+ 4,
1926
+ 2
1927
+ ],
1928
+ "bits_prop": [
1929
+ 0.05,
1930
+ 0.95
1931
+ ],
1932
+ "scale_bits": 4
1933
+ },
1934
+ "gate_proj": {
1935
+ "group_size": {
1936
+ "4": 128,
1937
+ "2": 128
1938
+ },
1939
+ "bits": [
1940
+ 4,
1941
+ 2
1942
+ ],
1943
+ "bits_prop": [
1944
+ 0.05,
1945
+ 0.95
1946
+ ],
1947
+ "scale_bits": 4
1948
+ },
1949
+ "down_proj": {
1950
+ "group_size": {
1951
+ "4": 128,
1952
+ "2": 128
1953
+ },
1954
+ "bits": [
1955
+ 4,
1956
+ 2
1957
+ ],
1958
+ "bits_prop": [
1959
+ 0.4,
1960
+ 0.6
1961
+ ],
1962
+ "scale_bits": 4
1963
+ }
1964
+ },
1965
+ "model.layers.18": {
1966
+ "accuracy": 0.9032173156738281,
1967
+ "total_bits": 142490128,
1968
+ "q_proj": {
1969
+ "group_size": {
1970
+ "4": 128,
1971
+ "2": 128
1972
+ },
1973
+ "bits": [
1974
+ 4,
1975
+ 2
1976
+ ],
1977
+ "bits_prop": [
1978
+ 0.1,
1979
+ 0.9
1980
+ ],
1981
+ "scale_bits": 4
1982
+ },
1983
+ "k_proj": {
1984
+ "group_size": {
1985
+ "4": 128,
1986
+ "2": 128
1987
+ },
1988
+ "bits": [
1989
+ 4,
1990
+ 2
1991
+ ],
1992
+ "bits_prop": [
1993
+ 0.1,
1994
+ 0.9
1995
+ ],
1996
+ "scale_bits": 4
1997
+ },
1998
+ "v_proj": {
1999
+ "group_size": {
2000
+ "4": 128
2001
+ },
2002
+ "bits": [
2003
+ 4
2004
+ ],
2005
+ "bits_prop": [
2006
+ 1.0
2007
+ ],
2008
+ "scale_bits": 4
2009
+ },
2010
+ "o_proj": {
2011
+ "group_size": {
2012
+ "4": 128,
2013
+ "2": 128
2014
+ },
2015
+ "bits": [
2016
+ 4,
2017
+ 2
2018
+ ],
2019
+ "bits_prop": [
2020
+ 0.1,
2021
+ 0.9
2022
+ ],
2023
+ "scale_bits": 4
2024
+ },
2025
+ "up_proj": {
2026
+ "group_size": {
2027
+ "4": 128,
2028
+ "2": 128
2029
+ },
2030
+ "bits": [
2031
+ 4,
2032
+ 2
2033
+ ],
2034
+ "bits_prop": [
2035
+ 0.1,
2036
+ 0.9
2037
+ ],
2038
+ "scale_bits": 4
2039
+ },
2040
+ "gate_proj": {
2041
+ "group_size": {
2042
+ "4": 128,
2043
+ "2": 128
2044
+ },
2045
+ "bits": [
2046
+ 4,
2047
+ 2
2048
+ ],
2049
+ "bits_prop": [
2050
+ 0.1,
2051
+ 0.9
2052
+ ],
2053
+ "scale_bits": 4
2054
+ },
2055
+ "down_proj": {
2056
+ "group_size": {
2057
+ "4": 128
2058
+ },
2059
+ "bits": [
2060
+ 4
2061
+ ],
2062
+ "bits_prop": [
2063
+ 1.0
2064
+ ],
2065
+ "scale_bits": 4
2066
+ }
2067
+ },
2068
+ "model.layers.19": {
2069
+ "accuracy": 0.9173600673675537,
2070
+ "total_bits": 151271952,
2071
+ "q_proj": {
2072
+ "group_size": {
2073
+ "4": 128,
2074
+ "2": 128
2075
+ },
2076
+ "bits": [
2077
+ 4,
2078
+ 2
2079
+ ],
2080
+ "bits_prop": [
2081
+ 0.25,
2082
+ 0.75
2083
+ ],
2084
+ "scale_bits": 4
2085
+ },
2086
+ "k_proj": {
2087
+ "group_size": {
2088
+ "4": 128,
2089
+ "2": 128
2090
+ },
2091
+ "bits": [
2092
+ 4,
2093
+ 2
2094
+ ],
2095
+ "bits_prop": [
2096
+ 0.25,
2097
+ 0.75
2098
+ ],
2099
+ "scale_bits": 4
2100
+ },
2101
+ "v_proj": {
2102
+ "group_size": {
2103
+ "4": 128
2104
+ },
2105
+ "bits": [
2106
+ 4
2107
+ ],
2108
+ "bits_prop": [
2109
+ 1.0
2110
+ ],
2111
+ "scale_bits": 4
2112
+ },
2113
+ "o_proj": {
2114
+ "group_size": {
2115
+ "4": 128,
2116
+ "2": 128
2117
+ },
2118
+ "bits": [
2119
+ 4,
2120
+ 2
2121
+ ],
2122
+ "bits_prop": [
2123
+ 0.25,
2124
+ 0.75
2125
+ ],
2126
+ "scale_bits": 4
2127
+ },
2128
+ "up_proj": {
2129
+ "group_size": {
2130
+ "4": 128,
2131
+ "2": 128
2132
+ },
2133
+ "bits": [
2134
+ 4,
2135
+ 2
2136
+ ],
2137
+ "bits_prop": [
2138
+ 0.25,
2139
+ 0.75
2140
+ ],
2141
+ "scale_bits": 4
2142
+ },
2143
+ "gate_proj": {
2144
+ "group_size": {
2145
+ "4": 128,
2146
+ "2": 128
2147
+ },
2148
+ "bits": [
2149
+ 4,
2150
+ 2
2151
+ ],
2152
+ "bits_prop": [
2153
+ 0.25,
2154
+ 0.75
2155
+ ],
2156
+ "scale_bits": 4
2157
+ },
2158
+ "down_proj": {
2159
+ "group_size": {
2160
+ "4": 128
2161
+ },
2162
+ "bits": [
2163
+ 4
2164
+ ],
2165
+ "bits_prop": [
2166
+ 1
2167
+ ],
2168
+ "scale_bits": 4
2169
+ }
2170
+ },
2171
+ "model.layers.20": {
2172
+ "accuracy": 0.9381833076477051,
2173
+ "total_bits": 168835600,
2174
+ "q_proj": {
2175
+ "group_size": {
2176
+ "4": 128,
2177
+ "2": 128
2178
+ },
2179
+ "bits": [
2180
+ 4,
2181
+ 2
2182
+ ],
2183
+ "bits_prop": [
2184
+ 0.5,
2185
+ 0.5
2186
+ ],
2187
+ "scale_bits": 4
2188
+ },
2189
+ "k_proj": {
2190
+ "group_size": {
2191
+ "4": 128,
2192
+ "2": 128
2193
+ },
2194
+ "bits": [
2195
+ 4,
2196
+ 2
2197
+ ],
2198
+ "bits_prop": [
2199
+ 0.5,
2200
+ 0.5
2201
+ ],
2202
+ "scale_bits": 4
2203
+ },
2204
+ "v_proj": {
2205
+ "group_size": {
2206
+ "4": 128
2207
+ },
2208
+ "bits": [
2209
+ 4
2210
+ ],
2211
+ "bits_prop": [
2212
+ 1.0
2213
+ ],
2214
+ "scale_bits": 4
2215
+ },
2216
+ "o_proj": {
2217
+ "group_size": {
2218
+ "4": 128,
2219
+ "2": 128
2220
+ },
2221
+ "bits": [
2222
+ 4,
2223
+ 2
2224
+ ],
2225
+ "bits_prop": [
2226
+ 0.5,
2227
+ 0.5
2228
+ ],
2229
+ "scale_bits": 4
2230
+ },
2231
+ "up_proj": {
2232
+ "group_size": {
2233
+ "4": 128,
2234
+ "2": 128
2235
+ },
2236
+ "bits": [
2237
+ 4,
2238
+ 2
2239
+ ],
2240
+ "bits_prop": [
2241
+ 0.5,
2242
+ 0.5
2243
+ ],
2244
+ "scale_bits": 4
2245
+ },
2246
+ "gate_proj": {
2247
+ "group_size": {
2248
+ "4": 128,
2249
+ "2": 128
2250
+ },
2251
+ "bits": [
2252
+ 4,
2253
+ 2
2254
+ ],
2255
+ "bits_prop": [
2256
+ 0.5,
2257
+ 0.5
2258
+ ],
2259
+ "scale_bits": 4
2260
+ },
2261
+ "down_proj": {
2262
+ "group_size": {
2263
+ "4": 128
2264
+ },
2265
+ "bits": [
2266
+ 4
2267
+ ],
2268
+ "bits_prop": [
2269
+ 1.0
2270
+ ],
2271
+ "scale_bits": 4
2272
+ }
2273
+ },
2274
+ "model.layers.21": {
2275
+ "accuracy": 0.9543092250823975,
2276
+ "total_bits": 186399248,
2277
+ "q_proj": {
2278
+ "group_size": {
2279
+ "4": 128,
2280
+ "2": 128
2281
+ },
2282
+ "bits": [
2283
+ 4,
2284
+ 2
2285
+ ],
2286
+ "bits_prop": [
2287
+ 0.75,
2288
+ 0.25
2289
+ ],
2290
+ "scale_bits": 4
2291
+ },
2292
+ "k_proj": {
2293
+ "group_size": {
2294
+ "4": 128,
2295
+ "2": 128
2296
+ },
2297
+ "bits": [
2298
+ 4,
2299
+ 2
2300
+ ],
2301
+ "bits_prop": [
2302
+ 0.75,
2303
+ 0.25
2304
+ ],
2305
+ "scale_bits": 4
2306
+ },
2307
+ "v_proj": {
2308
+ "group_size": {
2309
+ "4": 128
2310
+ },
2311
+ "bits": [
2312
+ 4
2313
+ ],
2314
+ "bits_prop": [
2315
+ 1.0
2316
+ ],
2317
+ "scale_bits": 4
2318
+ },
2319
+ "o_proj": {
2320
+ "group_size": {
2321
+ "4": 128,
2322
+ "2": 128
2323
+ },
2324
+ "bits": [
2325
+ 4,
2326
+ 2
2327
+ ],
2328
+ "bits_prop": [
2329
+ 0.75,
2330
+ 0.25
2331
+ ],
2332
+ "scale_bits": 4
2333
+ },
2334
+ "up_proj": {
2335
+ "group_size": {
2336
+ "4": 128,
2337
+ "2": 128
2338
+ },
2339
+ "bits": [
2340
+ 4,
2341
+ 2
2342
+ ],
2343
+ "bits_prop": [
2344
+ 0.75,
2345
+ 0.25
2346
+ ],
2347
+ "scale_bits": 4
2348
+ },
2349
+ "gate_proj": {
2350
+ "group_size": {
2351
+ "4": 128,
2352
+ "2": 128
2353
+ },
2354
+ "bits": [
2355
+ 4,
2356
+ 2
2357
+ ],
2358
+ "bits_prop": [
2359
+ 0.75,
2360
+ 0.25
2361
+ ],
2362
+ "scale_bits": 4
2363
+ },
2364
+ "down_proj": {
2365
+ "group_size": {
2366
+ "4": 128
2367
+ },
2368
+ "bits": [
2369
+ 4
2370
+ ],
2371
+ "bits_prop": [
2372
+ 1.0
2373
+ ],
2374
+ "scale_bits": 4
2375
+ }
2376
+ },
2377
+ "model.layers.22": {
2378
+ "accuracy": 0.9537525177001953,
2379
+ "total_bits": 186399248,
2380
+ "q_proj": {
2381
+ "group_size": {
2382
+ "4": 128,
2383
+ "2": 128
2384
+ },
2385
+ "bits": [
2386
+ 4,
2387
+ 2
2388
+ ],
2389
+ "bits_prop": [
2390
+ 0.75,
2391
+ 0.25
2392
+ ],
2393
+ "scale_bits": 4
2394
+ },
2395
+ "k_proj": {
2396
+ "group_size": {
2397
+ "4": 128,
2398
+ "2": 128
2399
+ },
2400
+ "bits": [
2401
+ 4,
2402
+ 2
2403
+ ],
2404
+ "bits_prop": [
2405
+ 0.75,
2406
+ 0.25
2407
+ ],
2408
+ "scale_bits": 4
2409
+ },
2410
+ "v_proj": {
2411
+ "group_size": {
2412
+ "4": 128
2413
+ },
2414
+ "bits": [
2415
+ 4
2416
+ ],
2417
+ "bits_prop": [
2418
+ 1.0
2419
+ ],
2420
+ "scale_bits": 4
2421
+ },
2422
+ "o_proj": {
2423
+ "group_size": {
2424
+ "4": 128,
2425
+ "2": 128
2426
+ },
2427
+ "bits": [
2428
+ 4,
2429
+ 2
2430
+ ],
2431
+ "bits_prop": [
2432
+ 0.75,
2433
+ 0.25
2434
+ ],
2435
+ "scale_bits": 4
2436
+ },
2437
+ "up_proj": {
2438
+ "group_size": {
2439
+ "4": 128,
2440
+ "2": 128
2441
+ },
2442
+ "bits": [
2443
+ 4,
2444
+ 2
2445
+ ],
2446
+ "bits_prop": [
2447
+ 0.75,
2448
+ 0.25
2449
+ ],
2450
+ "scale_bits": 4
2451
+ },
2452
+ "gate_proj": {
2453
+ "group_size": {
2454
+ "4": 128,
2455
+ "2": 128
2456
+ },
2457
+ "bits": [
2458
+ 4,
2459
+ 2
2460
+ ],
2461
+ "bits_prop": [
2462
+ 0.75,
2463
+ 0.25
2464
+ ],
2465
+ "scale_bits": 4
2466
+ },
2467
+ "down_proj": {
2468
+ "group_size": {
2469
+ "4": 128
2470
+ },
2471
+ "bits": [
2472
+ 4
2473
+ ],
2474
+ "bits_prop": [
2475
+ 1.0
2476
+ ],
2477
+ "scale_bits": 4
2478
+ }
2479
+ },
2480
+ "model.layers.23": {
2481
+ "accuracy": 0.9632000923156738,
2482
+ "total_bits": 203962896,
2483
+ "q_proj": {
2484
+ "group_size": {
2485
+ "4": 128
2486
+ },
2487
+ "bits": [
2488
+ 4
2489
+ ],
2490
+ "bits_prop": [
2491
+ 1
2492
+ ],
2493
+ "scale_bits": 4
2494
+ },
2495
+ "k_proj": {
2496
+ "group_size": {
2497
+ "4": 128
2498
+ },
2499
+ "bits": [
2500
+ 4
2501
+ ],
2502
+ "bits_prop": [
2503
+ 1
2504
+ ],
2505
+ "scale_bits": 4
2506
+ },
2507
+ "v_proj": {
2508
+ "group_size": {
2509
+ "4": 128
2510
+ },
2511
+ "bits": [
2512
+ 4
2513
+ ],
2514
+ "bits_prop": [
2515
+ 1
2516
+ ],
2517
+ "scale_bits": 4
2518
+ },
2519
+ "o_proj": {
2520
+ "group_size": {
2521
+ "4": 128
2522
+ },
2523
+ "bits": [
2524
+ 4
2525
+ ],
2526
+ "bits_prop": [
2527
+ 1
2528
+ ],
2529
+ "scale_bits": 4
2530
+ },
2531
+ "up_proj": {
2532
+ "group_size": {
2533
+ "4": 128
2534
+ },
2535
+ "bits": [
2536
+ 4
2537
+ ],
2538
+ "bits_prop": [
2539
+ 1
2540
+ ],
2541
+ "scale_bits": 4
2542
+ },
2543
+ "gate_proj": {
2544
+ "group_size": {
2545
+ "4": 128
2546
+ },
2547
+ "bits": [
2548
+ 4
2549
+ ],
2550
+ "bits_prop": [
2551
+ 1
2552
+ ],
2553
+ "scale_bits": 4
2554
+ },
2555
+ "down_proj": {
2556
+ "group_size": {
2557
+ "4": 128
2558
+ },
2559
+ "bits": [
2560
+ 4
2561
+ ],
2562
+ "bits_prop": [
2563
+ 1
2564
+ ],
2565
+ "scale_bits": 4
2566
+ }
2567
+ }
2568
+ }
2569
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "additional_special_tokens": [
3
+ "<|im_start|>",
4
+ "<|im_end|>"
5
+ ],
6
+ "eos_token": {
7
+ "content": "<|endoftext|>",
8
+ "lstrip": false,
9
+ "normalized": false,
10
+ "rstrip": false,
11
+ "single_word": false
12
+ },
13
+ "pad_token": {
14
+ "content": "<|endoftext|>",
15
+ "lstrip": false,
16
+ "normalized": false,
17
+ "rstrip": false,
18
+ "single_word": false
19
+ }
20
+ }
tokenizer_config.json ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "added_tokens_decoder": {
4
+ "151643": {
5
+ "content": "<|endoftext|>",
6
+ "lstrip": false,
7
+ "normalized": false,
8
+ "rstrip": false,
9
+ "single_word": false,
10
+ "special": true
11
+ },
12
+ "151644": {
13
+ "content": "<|im_start|>",
14
+ "lstrip": false,
15
+ "normalized": false,
16
+ "rstrip": false,
17
+ "single_word": false,
18
+ "special": true
19
+ },
20
+ "151645": {
21
+ "content": "<|im_end|>",
22
+ "lstrip": false,
23
+ "normalized": false,
24
+ "rstrip": false,
25
+ "single_word": false,
26
+ "special": true
27
+ }
28
+ },
29
+ "additional_special_tokens": [
30
+ "<|im_start|>",
31
+ "<|im_end|>"
32
+ ],
33
+ "bos_token": null,
34
+ "chat_template": "{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system\nYou are a helpful assistant<|im_end|>\n' }}{% endif %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
35
+ "clean_up_tokenization_spaces": false,
36
+ "eos_token": "<|endoftext|>",
37
+ "errors": "replace",
38
+ "model_max_length": 32768,
39
+ "pad_token": "<|endoftext|>",
40
+ "split_special_tokens": false,
41
+ "tokenizer_class": "Qwen2Tokenizer",
42
+ "unk_token": null
43
+ }
vocab.json ADDED
The diff for this file is too large to render. See raw diff