pipenetwork commited on
Commit
33e439f
·
verified ·
1 Parent(s): 77e8e38

Add files using upload-large-folder tool

Browse files
README.md ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ library_name: mlx
3
+ pipeline_tag: text-generation
4
+ tags: [mlx, hunyuan, hy3, reap, moe-pruning]
5
+ license: apache-2.0
6
+ base_model: tencent/Hy3
7
+ ---
8
+ # pipenetwork/Hy3-REAP75-MLX-4bit
9
+ REAP-pruned [tencent/Hy3](https://huggingface.co/tencent/Hy3) (Hunyuan-3): kept **48/192 routed
10
+ experts/layer** by REAP saliency, MLX 4-bit. First REAP-pruned Hy3. Requires mlx-lm hy_v3 (PR #1211):
11
+ `pip install git+https://github.com/kernelpool/mlx-lm.git@add-hy3-preview`
chat_template.jinja ADDED
@@ -0,0 +1,222 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {#- ----------‑‑‑ special token variables ‑‑‑---------- -#}
2
+ {%- set HYTK = ':opensource' %}
3
+ {%- set eos_token = '<|hy_eos{}|>'.format(HYTK) %}
4
+ {%- set bos_token = '<|hy_begin_of_sentence{}|>'.format(HYTK) %}
5
+ {%- set pad_token = '<|hy_pad{}|>'.format(HYTK) %}
6
+ {%- set user_token = '<|hy_User{}|>'.format(HYTK) %}
7
+ {%- set assistant_token = '<|hy_Assistant{}|>'.format(HYTK) %}
8
+ {%- set think_begin_token = '<think{}>'.format(HYTK) %}
9
+ {%- set think_end_token = '</think{}>'.format(HYTK) %}
10
+ {%- set toolcalls_begin_token = '<tool_calls{}>'.format(HYTK) %}
11
+ {%- set toolcalls_end_token = '</tool_calls{}>'.format(HYTK) %}
12
+ {%- set toolcall_begin_token = '<tool_call{}>'.format(HYTK) %}
13
+ {%- set toolcall_end_token = '</tool_call{}>'.format(HYTK) %}
14
+ {%- set toolsep_token = '<tool_sep{}>'.format(HYTK) %}
15
+ {%- set argkey_begin_token = '<arg_key{}>'.format(HYTK) %}
16
+ {%- set argkey_end_token = '</arg_key{}>'.format(HYTK) %}
17
+ {%- set argvalue_begin_token = '<arg_value{}>'.format(HYTK) %}
18
+ {%- set argvalue_end_token = '</arg_value{}>'.format(HYTK) %}
19
+ {%- set toolresponses_begin_token = '<tool_responses{}>'.format(HYTK) %}
20
+ {%- set toolresponses_end_token = '</tool_responses{}>'.format(HYTK) %}
21
+ {%- set toolresponse_begin_token = '<tool_response{}>'.format(HYTK) %}
22
+ {%- set toolresponse_end_token = '</tool_response{}>'.format(HYTK) %}
23
+ {%- set reasoning_mode_token = '<|reasoning_mode{}|>'.format(HYTK) %}
24
+
25
+ {#- ----------‑‑‑ hyperparameters variables ‑‑‑---------- -#}
26
+ {%- if not add_generation_prompt is defined %}
27
+ {%- set add_generation_prompt = false %}
28
+ {%- endif %}
29
+ {%- if not preserved_thinking is defined %}
30
+ {%- if not tools %}
31
+ {%- set preserved_thinking = false %}
32
+ {%- else %}
33
+ {%- set preserved_thinking = true %}
34
+ {%- endif %}
35
+ {%- endif %}
36
+ {%- if not is_training is defined %}
37
+ {%- set is_training = false %}
38
+ {%- endif %}
39
+
40
+ {%- if not reasoning_effort is defined %}
41
+ {%- set reasoning_effort = 'no_think' %}
42
+ {%- elif reasoning_effort not in ['high', 'low', 'no_think'] %}
43
+ {%- if reasoning_effort is none %}
44
+ {{- raise_exception('reasoning_effort error : None, should be no_think/low/high') }}
45
+ {%- else %}
46
+ {{- raise_exception('reasoning_effort error : ' + reasoning_effort + ', should be no_think/low/high') }}
47
+ {%- endif %}
48
+ {%- endif %}
49
+
50
+ {%- if fallback_strategy is defined and fallback_strategy == 'reasoning_toolcall_retry' %}
51
+ {%- set reasoning_effort = 'high' %}
52
+ {%- set add_generation_prompt = false %}
53
+ {%- endif %}
54
+ {%- if not raw_last_assistant is defined %}
55
+ {%- set raw_last_assistant = false %}
56
+ {%- endif %}
57
+
58
+ {%- macro visible_text(content) -%}
59
+ {%- if content is string -%}
60
+ {{- content }}
61
+ {%- elif content is iterable and content is not mapping -%}
62
+ {%- for item in content -%}
63
+ {%- if item is mapping and item.type == 'text' -%}
64
+ {{- item.text }}
65
+ {%- elif item is string -%}
66
+ {{- item }}
67
+ {%- endif -%}
68
+ {%- endfor -%}
69
+ {%- elif content is none -%}
70
+ {{- '' }}
71
+ {%- else -%}
72
+ {{- content }}
73
+ {%- endif -%}
74
+ {%- endmacro -%}
75
+
76
+ {%- set ns = namespace(last_user_index=-1) %}
77
+ {%- set sp_ns = namespace(system_prompt='', is_first_sp=true) %}
78
+ {%- for message in messages %}
79
+ {%- if message['role'] == 'system' %}
80
+ {%- set sp_ns.system_prompt = sp_ns.system_prompt + visible_text(message['content']) %}
81
+ {%- endif %}
82
+ {%- if message['role'] == 'user' %}
83
+ {%- set ns.last_user_index = loop.index0 %}
84
+ {%- endif %}
85
+ {%- endfor %}
86
+ {%- if reasoning_effort is defined and reasoning_effort is string and reasoning_effort != '' and not tools %}
87
+ {%- set sp_ns.system_prompt = sp_ns.system_prompt + reasoning_mode_token + 'reasoning_effort:' + reasoning_effort %}
88
+ {%- endif %}
89
+ {{- bos_token }}
90
+ {{- sp_ns.system_prompt }}
91
+ {%- if tools %}
92
+ {%- if sp_ns.system_prompt != '' %}
93
+ {{- '\n\n# Tools\n\nYou may call one or more functions to assist with the user query.' }}
94
+ {%- else %}
95
+ {{- '# Tools\n\nYou may call one or more functions to assist with the user query.' }}
96
+ {%- endif %}
97
+ {{- '\n\nYou are provided with function signatures within <tools></tools> XML tags:' }}
98
+ {{- '\n<tools>\n' }}
99
+ {%- for tool in tools %}
100
+ {%- if loop.index0 > 0 %}
101
+ {{- '\n' }}
102
+ {%- endif %}
103
+ {{- tool | tojson }}
104
+ {%- endfor %}
105
+ {{- '\n</tools>\n\n' }}
106
+ {{- 'For function call returns, you should first print ' + toolcalls_begin_token + '\n' }}
107
+ {{- 'For each function call, you should return object like:\n' }}
108
+ {{- toolcall_begin_token + '{function-name}' + toolsep_token + '\n' }}
109
+ {{- argkey_begin_token + '{arg-key-1}' + argkey_end_token + '\n' }}
110
+ {{- argvalue_begin_token + '{arg-value-1}' + argvalue_end_token + '\n' }}
111
+ {{- argkey_begin_token + '{arg-key-2}' + argkey_end_token + '\n' }}
112
+ {{- argvalue_begin_token + '{arg-value-2}' + argvalue_end_token + '\n' }}
113
+ {{- '...\n' }}
114
+ {{- toolcall_end_token + '\n' }}
115
+ {%- if reasoning_effort is defined and reasoning_effort is string and reasoning_effort != '' %}
116
+ {{- 'At the end of function call returns, you should print ' + toolcalls_end_token + reasoning_mode_token + 'reasoning_effort:' + reasoning_effort }}
117
+ {%- else %}
118
+ {{- 'At the end of function call returns, you should print ' + toolcalls_end_token }}
119
+ {%- endif %}
120
+ {%- endif %}
121
+
122
+ {%- set prev_ns = namespace(is_tool=false, is_tool_first=true) %}
123
+ {%- set last_ns = namespace(last_is_assistant=false) %}
124
+ {%- for message in messages %}
125
+ {%- if message['role'] == 'user' %}
126
+ {%- if prev_ns.is_tool %}
127
+ {{- toolresponses_end_token }}
128
+ {%- endif %}
129
+ {{- user_token + visible_text(message['content']) }}
130
+ {%- set prev_ns.is_tool = false %}
131
+ {%- endif %}
132
+ {%- if message['role'] == 'assistant' %}
133
+ {%- if is_training %}
134
+ {%- if 'reasoning_content' in message and message['reasoning_content'] is string %}
135
+ {%- set rc = message['reasoning_content'] %}
136
+ {%- elif 'reasoning' in message and message['reasoning'] is string %}
137
+ {%- set rc = message['reasoning'] %}
138
+ {%- else %}
139
+ {%- set rc = none %}
140
+ {%- endif %}
141
+ {%- if rc is not none %}
142
+ {%- set content = think_begin_token + rc + think_end_token + visible_text(message['content']) %}
143
+ {%- else %}
144
+ {%- set content = think_begin_token + think_end_token + visible_text(message['content']) %}
145
+ {%- endif %}
146
+ {%- else %}
147
+ {%- if ((preserved_thinking is defined and preserved_thinking) or loop.index0 > ns.last_user_index) %}
148
+ {%- if 'reasoning_content' in message and message['reasoning_content'] is string %}
149
+ {%- set rc = message['reasoning_content'] %}
150
+ {%- elif 'reasoning' in message and message['reasoning'] is string %}
151
+ {%- set rc = message['reasoning'] %}
152
+ {%- else %}
153
+ {%- set rc = none %}
154
+ {%- endif %}
155
+ {%- if rc is not none %}
156
+ {%- set content = think_begin_token + rc + think_end_token + visible_text(message['content']) %}
157
+ {%- else %}
158
+ {%- set content = think_begin_token + think_end_token + visible_text(message['content']) %}
159
+ {%- endif %}
160
+ {%- else %}
161
+ {%- set content = think_begin_token + think_end_token + visible_text(message['content']) %}
162
+ {%- endif %}
163
+ {%- endif %}
164
+ {%- if prev_ns.is_tool %}
165
+ {{- toolresponses_end_token }}
166
+ {%- endif %}
167
+ {{- assistant_token }}
168
+ {%- if message['tool_calls'] is defined and message['tool_calls'] %}
169
+ {%- set prev_ns.is_tool_first = true %}
170
+ {{- content }}
171
+ {{- toolcalls_begin_token + '\n' }}
172
+ {%- for tool in message['tool_calls'] %}
173
+ {%- set arguments = tool['function']['arguments'] %}
174
+ {{- toolcall_begin_token + tool['function']['name'] + toolsep_token + '\n' }}
175
+ {%- for key, value in arguments.items() %}
176
+ {{- argkey_begin_token + key + argkey_end_token + '\n' }}
177
+ {%- if value is not string %}
178
+ {%- set value = value | tojson(ensure_ascii=False) %}
179
+ {%- endif %}
180
+ {{- argvalue_begin_token + value + argvalue_end_token + '\n' }}
181
+ {%- endfor %}
182
+ {{- toolcall_end_token + '\n' }}
183
+ {%- endfor %}
184
+ {{- toolcalls_end_token + eos_token }}
185
+ {%- else %}
186
+ {%- if loop.last and raw_last_assistant %}
187
+ {{- visible_text(message['content']) }}
188
+ {%- elif not loop.last or is_training %}
189
+ {{- content + eos_token }}
190
+ {%- else %}
191
+ {{- content }}
192
+ {%- endif %}
193
+ {%- endif %}
194
+ {%- set prev_ns.is_tool = false %}
195
+ {%- endif %}
196
+ {%- if message['role'] == 'tool' %}
197
+ {%- set prev_ns.is_tool = true %}
198
+ {%- if prev_ns.is_tool_first %}
199
+ {{- toolresponses_begin_token + '\n' }}
200
+ {%- set prev_ns.is_tool_first = false %}
201
+ {%- endif %}
202
+ {{- toolresponse_begin_token + '\n' + visible_text(message['content']) + '\n' + toolresponse_end_token + '\n' }}
203
+ {%- endif %}
204
+ {%- if loop.last and message['role'] == 'assistant' %}
205
+ {%- set last_ns.last_is_assistant = true %}
206
+ {%- endif %}
207
+
208
+ {%- endfor %}
209
+ {%- if prev_ns.is_tool %}
210
+ {{- toolresponses_end_token }}
211
+ {%- endif %}
212
+ {%- if add_generation_prompt %}
213
+ {%- if not last_ns.last_is_assistant %}
214
+ {%- if reasoning_effort is defined and reasoning_effort in ['low', 'high'] %}
215
+ {{- assistant_token + think_begin_token }}
216
+ {%- elif reasoning_effort is defined and reasoning_effort == 'no_think' %}
217
+ {{- assistant_token + think_begin_token + think_end_token }}
218
+ {%- else %}
219
+ {{- assistant_token }}
220
+ {%- endif %}
221
+ {%- endif %}
222
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,692 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "HYV3ForCausalLM"
4
+ ],
5
+ "bos_token_id": 120000,
6
+ "enable_attention_fp32_softmax": false,
7
+ "enable_lm_head_fp32": true,
8
+ "enable_moe_fp32_combine": false,
9
+ "eod_token_id": 120026,
10
+ "eos_token_id": 120025,
11
+ "expert_hidden_dim": 1536,
12
+ "moe_intermediate_size": 1536,
13
+ "first_k_dense_replace": 1,
14
+ "head_dim": 128,
15
+ "hidden_act": "silu",
16
+ "hidden_size": 4096,
17
+ "initializer_range": 0.006,
18
+ "intermediate_size": 13312,
19
+ "max_position_embeddings": 262144,
20
+ "model_type": "hy_v3",
21
+ "moe_router_enable_expert_bias": true,
22
+ "moe_router_use_sigmoid": true,
23
+ "num_attention_heads": 64,
24
+ "num_experts": 48,
25
+ "num_experts_per_tok": 8,
26
+ "num_hidden_layers": 80,
27
+ "num_key_value_heads": 8,
28
+ "num_shared_experts": 1,
29
+ "output_router_logits": true,
30
+ "pad_token_id": 120002,
31
+ "qk_norm": true,
32
+ "rms_norm_eps": 1e-05,
33
+ "rope_parameters": {
34
+ "rope_theta": 11158840.0,
35
+ "rope_type": "default"
36
+ },
37
+ "route_norm": true,
38
+ "router_scaling_factor": 2.826,
39
+ "sep_token_id": 120007,
40
+ "tie_word_embeddings": false,
41
+ "transformers_version": "5.6.0",
42
+ "use_cache": true,
43
+ "use_grouped_mm": false,
44
+ "vocab_size": 120832,
45
+ "num_nextn_predict_layers": 1,
46
+ "quantization": {
47
+ "group_size": 64,
48
+ "bits": 4,
49
+ "mode": "affine",
50
+ "model.layers.1.mlp.router.gate": {
51
+ "group_size": 64,
52
+ "bits": 8
53
+ },
54
+ "model.layers.2.mlp.router.gate": {
55
+ "group_size": 64,
56
+ "bits": 8
57
+ },
58
+ "model.layers.3.mlp.router.gate": {
59
+ "group_size": 64,
60
+ "bits": 8
61
+ },
62
+ "model.layers.4.mlp.router.gate": {
63
+ "group_size": 64,
64
+ "bits": 8
65
+ },
66
+ "model.layers.5.mlp.router.gate": {
67
+ "group_size": 64,
68
+ "bits": 8
69
+ },
70
+ "model.layers.6.mlp.router.gate": {
71
+ "group_size": 64,
72
+ "bits": 8
73
+ },
74
+ "model.layers.7.mlp.router.gate": {
75
+ "group_size": 64,
76
+ "bits": 8
77
+ },
78
+ "model.layers.8.mlp.router.gate": {
79
+ "group_size": 64,
80
+ "bits": 8
81
+ },
82
+ "model.layers.9.mlp.router.gate": {
83
+ "group_size": 64,
84
+ "bits": 8
85
+ },
86
+ "model.layers.10.mlp.router.gate": {
87
+ "group_size": 64,
88
+ "bits": 8
89
+ },
90
+ "model.layers.11.mlp.router.gate": {
91
+ "group_size": 64,
92
+ "bits": 8
93
+ },
94
+ "model.layers.12.mlp.router.gate": {
95
+ "group_size": 64,
96
+ "bits": 8
97
+ },
98
+ "model.layers.13.mlp.router.gate": {
99
+ "group_size": 64,
100
+ "bits": 8
101
+ },
102
+ "model.layers.14.mlp.router.gate": {
103
+ "group_size": 64,
104
+ "bits": 8
105
+ },
106
+ "model.layers.15.mlp.router.gate": {
107
+ "group_size": 64,
108
+ "bits": 8
109
+ },
110
+ "model.layers.16.mlp.router.gate": {
111
+ "group_size": 64,
112
+ "bits": 8
113
+ },
114
+ "model.layers.17.mlp.router.gate": {
115
+ "group_size": 64,
116
+ "bits": 8
117
+ },
118
+ "model.layers.18.mlp.router.gate": {
119
+ "group_size": 64,
120
+ "bits": 8
121
+ },
122
+ "model.layers.19.mlp.router.gate": {
123
+ "group_size": 64,
124
+ "bits": 8
125
+ },
126
+ "model.layers.20.mlp.router.gate": {
127
+ "group_size": 64,
128
+ "bits": 8
129
+ },
130
+ "model.layers.21.mlp.router.gate": {
131
+ "group_size": 64,
132
+ "bits": 8
133
+ },
134
+ "model.layers.22.mlp.router.gate": {
135
+ "group_size": 64,
136
+ "bits": 8
137
+ },
138
+ "model.layers.23.mlp.router.gate": {
139
+ "group_size": 64,
140
+ "bits": 8
141
+ },
142
+ "model.layers.24.mlp.router.gate": {
143
+ "group_size": 64,
144
+ "bits": 8
145
+ },
146
+ "model.layers.25.mlp.router.gate": {
147
+ "group_size": 64,
148
+ "bits": 8
149
+ },
150
+ "model.layers.26.mlp.router.gate": {
151
+ "group_size": 64,
152
+ "bits": 8
153
+ },
154
+ "model.layers.27.mlp.router.gate": {
155
+ "group_size": 64,
156
+ "bits": 8
157
+ },
158
+ "model.layers.28.mlp.router.gate": {
159
+ "group_size": 64,
160
+ "bits": 8
161
+ },
162
+ "model.layers.29.mlp.router.gate": {
163
+ "group_size": 64,
164
+ "bits": 8
165
+ },
166
+ "model.layers.30.mlp.router.gate": {
167
+ "group_size": 64,
168
+ "bits": 8
169
+ },
170
+ "model.layers.31.mlp.router.gate": {
171
+ "group_size": 64,
172
+ "bits": 8
173
+ },
174
+ "model.layers.32.mlp.router.gate": {
175
+ "group_size": 64,
176
+ "bits": 8
177
+ },
178
+ "model.layers.33.mlp.router.gate": {
179
+ "group_size": 64,
180
+ "bits": 8
181
+ },
182
+ "model.layers.34.mlp.router.gate": {
183
+ "group_size": 64,
184
+ "bits": 8
185
+ },
186
+ "model.layers.35.mlp.router.gate": {
187
+ "group_size": 64,
188
+ "bits": 8
189
+ },
190
+ "model.layers.36.mlp.router.gate": {
191
+ "group_size": 64,
192
+ "bits": 8
193
+ },
194
+ "model.layers.37.mlp.router.gate": {
195
+ "group_size": 64,
196
+ "bits": 8
197
+ },
198
+ "model.layers.38.mlp.router.gate": {
199
+ "group_size": 64,
200
+ "bits": 8
201
+ },
202
+ "model.layers.39.mlp.router.gate": {
203
+ "group_size": 64,
204
+ "bits": 8
205
+ },
206
+ "model.layers.40.mlp.router.gate": {
207
+ "group_size": 64,
208
+ "bits": 8
209
+ },
210
+ "model.layers.41.mlp.router.gate": {
211
+ "group_size": 64,
212
+ "bits": 8
213
+ },
214
+ "model.layers.42.mlp.router.gate": {
215
+ "group_size": 64,
216
+ "bits": 8
217
+ },
218
+ "model.layers.43.mlp.router.gate": {
219
+ "group_size": 64,
220
+ "bits": 8
221
+ },
222
+ "model.layers.44.mlp.router.gate": {
223
+ "group_size": 64,
224
+ "bits": 8
225
+ },
226
+ "model.layers.45.mlp.router.gate": {
227
+ "group_size": 64,
228
+ "bits": 8
229
+ },
230
+ "model.layers.46.mlp.router.gate": {
231
+ "group_size": 64,
232
+ "bits": 8
233
+ },
234
+ "model.layers.47.mlp.router.gate": {
235
+ "group_size": 64,
236
+ "bits": 8
237
+ },
238
+ "model.layers.48.mlp.router.gate": {
239
+ "group_size": 64,
240
+ "bits": 8
241
+ },
242
+ "model.layers.49.mlp.router.gate": {
243
+ "group_size": 64,
244
+ "bits": 8
245
+ },
246
+ "model.layers.50.mlp.router.gate": {
247
+ "group_size": 64,
248
+ "bits": 8
249
+ },
250
+ "model.layers.51.mlp.router.gate": {
251
+ "group_size": 64,
252
+ "bits": 8
253
+ },
254
+ "model.layers.52.mlp.router.gate": {
255
+ "group_size": 64,
256
+ "bits": 8
257
+ },
258
+ "model.layers.53.mlp.router.gate": {
259
+ "group_size": 64,
260
+ "bits": 8
261
+ },
262
+ "model.layers.54.mlp.router.gate": {
263
+ "group_size": 64,
264
+ "bits": 8
265
+ },
266
+ "model.layers.55.mlp.router.gate": {
267
+ "group_size": 64,
268
+ "bits": 8
269
+ },
270
+ "model.layers.56.mlp.router.gate": {
271
+ "group_size": 64,
272
+ "bits": 8
273
+ },
274
+ "model.layers.57.mlp.router.gate": {
275
+ "group_size": 64,
276
+ "bits": 8
277
+ },
278
+ "model.layers.58.mlp.router.gate": {
279
+ "group_size": 64,
280
+ "bits": 8
281
+ },
282
+ "model.layers.59.mlp.router.gate": {
283
+ "group_size": 64,
284
+ "bits": 8
285
+ },
286
+ "model.layers.60.mlp.router.gate": {
287
+ "group_size": 64,
288
+ "bits": 8
289
+ },
290
+ "model.layers.61.mlp.router.gate": {
291
+ "group_size": 64,
292
+ "bits": 8
293
+ },
294
+ "model.layers.62.mlp.router.gate": {
295
+ "group_size": 64,
296
+ "bits": 8
297
+ },
298
+ "model.layers.63.mlp.router.gate": {
299
+ "group_size": 64,
300
+ "bits": 8
301
+ },
302
+ "model.layers.64.mlp.router.gate": {
303
+ "group_size": 64,
304
+ "bits": 8
305
+ },
306
+ "model.layers.65.mlp.router.gate": {
307
+ "group_size": 64,
308
+ "bits": 8
309
+ },
310
+ "model.layers.66.mlp.router.gate": {
311
+ "group_size": 64,
312
+ "bits": 8
313
+ },
314
+ "model.layers.67.mlp.router.gate": {
315
+ "group_size": 64,
316
+ "bits": 8
317
+ },
318
+ "model.layers.68.mlp.router.gate": {
319
+ "group_size": 64,
320
+ "bits": 8
321
+ },
322
+ "model.layers.69.mlp.router.gate": {
323
+ "group_size": 64,
324
+ "bits": 8
325
+ },
326
+ "model.layers.70.mlp.router.gate": {
327
+ "group_size": 64,
328
+ "bits": 8
329
+ },
330
+ "model.layers.71.mlp.router.gate": {
331
+ "group_size": 64,
332
+ "bits": 8
333
+ },
334
+ "model.layers.72.mlp.router.gate": {
335
+ "group_size": 64,
336
+ "bits": 8
337
+ },
338
+ "model.layers.73.mlp.router.gate": {
339
+ "group_size": 64,
340
+ "bits": 8
341
+ },
342
+ "model.layers.74.mlp.router.gate": {
343
+ "group_size": 64,
344
+ "bits": 8
345
+ },
346
+ "model.layers.75.mlp.router.gate": {
347
+ "group_size": 64,
348
+ "bits": 8
349
+ },
350
+ "model.layers.76.mlp.router.gate": {
351
+ "group_size": 64,
352
+ "bits": 8
353
+ },
354
+ "model.layers.77.mlp.router.gate": {
355
+ "group_size": 64,
356
+ "bits": 8
357
+ },
358
+ "model.layers.78.mlp.router.gate": {
359
+ "group_size": 64,
360
+ "bits": 8
361
+ },
362
+ "model.layers.79.mlp.router.gate": {
363
+ "group_size": 64,
364
+ "bits": 8
365
+ }
366
+ },
367
+ "quantization_config": {
368
+ "group_size": 64,
369
+ "bits": 4,
370
+ "mode": "affine",
371
+ "model.layers.1.mlp.router.gate": {
372
+ "group_size": 64,
373
+ "bits": 8
374
+ },
375
+ "model.layers.2.mlp.router.gate": {
376
+ "group_size": 64,
377
+ "bits": 8
378
+ },
379
+ "model.layers.3.mlp.router.gate": {
380
+ "group_size": 64,
381
+ "bits": 8
382
+ },
383
+ "model.layers.4.mlp.router.gate": {
384
+ "group_size": 64,
385
+ "bits": 8
386
+ },
387
+ "model.layers.5.mlp.router.gate": {
388
+ "group_size": 64,
389
+ "bits": 8
390
+ },
391
+ "model.layers.6.mlp.router.gate": {
392
+ "group_size": 64,
393
+ "bits": 8
394
+ },
395
+ "model.layers.7.mlp.router.gate": {
396
+ "group_size": 64,
397
+ "bits": 8
398
+ },
399
+ "model.layers.8.mlp.router.gate": {
400
+ "group_size": 64,
401
+ "bits": 8
402
+ },
403
+ "model.layers.9.mlp.router.gate": {
404
+ "group_size": 64,
405
+ "bits": 8
406
+ },
407
+ "model.layers.10.mlp.router.gate": {
408
+ "group_size": 64,
409
+ "bits": 8
410
+ },
411
+ "model.layers.11.mlp.router.gate": {
412
+ "group_size": 64,
413
+ "bits": 8
414
+ },
415
+ "model.layers.12.mlp.router.gate": {
416
+ "group_size": 64,
417
+ "bits": 8
418
+ },
419
+ "model.layers.13.mlp.router.gate": {
420
+ "group_size": 64,
421
+ "bits": 8
422
+ },
423
+ "model.layers.14.mlp.router.gate": {
424
+ "group_size": 64,
425
+ "bits": 8
426
+ },
427
+ "model.layers.15.mlp.router.gate": {
428
+ "group_size": 64,
429
+ "bits": 8
430
+ },
431
+ "model.layers.16.mlp.router.gate": {
432
+ "group_size": 64,
433
+ "bits": 8
434
+ },
435
+ "model.layers.17.mlp.router.gate": {
436
+ "group_size": 64,
437
+ "bits": 8
438
+ },
439
+ "model.layers.18.mlp.router.gate": {
440
+ "group_size": 64,
441
+ "bits": 8
442
+ },
443
+ "model.layers.19.mlp.router.gate": {
444
+ "group_size": 64,
445
+ "bits": 8
446
+ },
447
+ "model.layers.20.mlp.router.gate": {
448
+ "group_size": 64,
449
+ "bits": 8
450
+ },
451
+ "model.layers.21.mlp.router.gate": {
452
+ "group_size": 64,
453
+ "bits": 8
454
+ },
455
+ "model.layers.22.mlp.router.gate": {
456
+ "group_size": 64,
457
+ "bits": 8
458
+ },
459
+ "model.layers.23.mlp.router.gate": {
460
+ "group_size": 64,
461
+ "bits": 8
462
+ },
463
+ "model.layers.24.mlp.router.gate": {
464
+ "group_size": 64,
465
+ "bits": 8
466
+ },
467
+ "model.layers.25.mlp.router.gate": {
468
+ "group_size": 64,
469
+ "bits": 8
470
+ },
471
+ "model.layers.26.mlp.router.gate": {
472
+ "group_size": 64,
473
+ "bits": 8
474
+ },
475
+ "model.layers.27.mlp.router.gate": {
476
+ "group_size": 64,
477
+ "bits": 8
478
+ },
479
+ "model.layers.28.mlp.router.gate": {
480
+ "group_size": 64,
481
+ "bits": 8
482
+ },
483
+ "model.layers.29.mlp.router.gate": {
484
+ "group_size": 64,
485
+ "bits": 8
486
+ },
487
+ "model.layers.30.mlp.router.gate": {
488
+ "group_size": 64,
489
+ "bits": 8
490
+ },
491
+ "model.layers.31.mlp.router.gate": {
492
+ "group_size": 64,
493
+ "bits": 8
494
+ },
495
+ "model.layers.32.mlp.router.gate": {
496
+ "group_size": 64,
497
+ "bits": 8
498
+ },
499
+ "model.layers.33.mlp.router.gate": {
500
+ "group_size": 64,
501
+ "bits": 8
502
+ },
503
+ "model.layers.34.mlp.router.gate": {
504
+ "group_size": 64,
505
+ "bits": 8
506
+ },
507
+ "model.layers.35.mlp.router.gate": {
508
+ "group_size": 64,
509
+ "bits": 8
510
+ },
511
+ "model.layers.36.mlp.router.gate": {
512
+ "group_size": 64,
513
+ "bits": 8
514
+ },
515
+ "model.layers.37.mlp.router.gate": {
516
+ "group_size": 64,
517
+ "bits": 8
518
+ },
519
+ "model.layers.38.mlp.router.gate": {
520
+ "group_size": 64,
521
+ "bits": 8
522
+ },
523
+ "model.layers.39.mlp.router.gate": {
524
+ "group_size": 64,
525
+ "bits": 8
526
+ },
527
+ "model.layers.40.mlp.router.gate": {
528
+ "group_size": 64,
529
+ "bits": 8
530
+ },
531
+ "model.layers.41.mlp.router.gate": {
532
+ "group_size": 64,
533
+ "bits": 8
534
+ },
535
+ "model.layers.42.mlp.router.gate": {
536
+ "group_size": 64,
537
+ "bits": 8
538
+ },
539
+ "model.layers.43.mlp.router.gate": {
540
+ "group_size": 64,
541
+ "bits": 8
542
+ },
543
+ "model.layers.44.mlp.router.gate": {
544
+ "group_size": 64,
545
+ "bits": 8
546
+ },
547
+ "model.layers.45.mlp.router.gate": {
548
+ "group_size": 64,
549
+ "bits": 8
550
+ },
551
+ "model.layers.46.mlp.router.gate": {
552
+ "group_size": 64,
553
+ "bits": 8
554
+ },
555
+ "model.layers.47.mlp.router.gate": {
556
+ "group_size": 64,
557
+ "bits": 8
558
+ },
559
+ "model.layers.48.mlp.router.gate": {
560
+ "group_size": 64,
561
+ "bits": 8
562
+ },
563
+ "model.layers.49.mlp.router.gate": {
564
+ "group_size": 64,
565
+ "bits": 8
566
+ },
567
+ "model.layers.50.mlp.router.gate": {
568
+ "group_size": 64,
569
+ "bits": 8
570
+ },
571
+ "model.layers.51.mlp.router.gate": {
572
+ "group_size": 64,
573
+ "bits": 8
574
+ },
575
+ "model.layers.52.mlp.router.gate": {
576
+ "group_size": 64,
577
+ "bits": 8
578
+ },
579
+ "model.layers.53.mlp.router.gate": {
580
+ "group_size": 64,
581
+ "bits": 8
582
+ },
583
+ "model.layers.54.mlp.router.gate": {
584
+ "group_size": 64,
585
+ "bits": 8
586
+ },
587
+ "model.layers.55.mlp.router.gate": {
588
+ "group_size": 64,
589
+ "bits": 8
590
+ },
591
+ "model.layers.56.mlp.router.gate": {
592
+ "group_size": 64,
593
+ "bits": 8
594
+ },
595
+ "model.layers.57.mlp.router.gate": {
596
+ "group_size": 64,
597
+ "bits": 8
598
+ },
599
+ "model.layers.58.mlp.router.gate": {
600
+ "group_size": 64,
601
+ "bits": 8
602
+ },
603
+ "model.layers.59.mlp.router.gate": {
604
+ "group_size": 64,
605
+ "bits": 8
606
+ },
607
+ "model.layers.60.mlp.router.gate": {
608
+ "group_size": 64,
609
+ "bits": 8
610
+ },
611
+ "model.layers.61.mlp.router.gate": {
612
+ "group_size": 64,
613
+ "bits": 8
614
+ },
615
+ "model.layers.62.mlp.router.gate": {
616
+ "group_size": 64,
617
+ "bits": 8
618
+ },
619
+ "model.layers.63.mlp.router.gate": {
620
+ "group_size": 64,
621
+ "bits": 8
622
+ },
623
+ "model.layers.64.mlp.router.gate": {
624
+ "group_size": 64,
625
+ "bits": 8
626
+ },
627
+ "model.layers.65.mlp.router.gate": {
628
+ "group_size": 64,
629
+ "bits": 8
630
+ },
631
+ "model.layers.66.mlp.router.gate": {
632
+ "group_size": 64,
633
+ "bits": 8
634
+ },
635
+ "model.layers.67.mlp.router.gate": {
636
+ "group_size": 64,
637
+ "bits": 8
638
+ },
639
+ "model.layers.68.mlp.router.gate": {
640
+ "group_size": 64,
641
+ "bits": 8
642
+ },
643
+ "model.layers.69.mlp.router.gate": {
644
+ "group_size": 64,
645
+ "bits": 8
646
+ },
647
+ "model.layers.70.mlp.router.gate": {
648
+ "group_size": 64,
649
+ "bits": 8
650
+ },
651
+ "model.layers.71.mlp.router.gate": {
652
+ "group_size": 64,
653
+ "bits": 8
654
+ },
655
+ "model.layers.72.mlp.router.gate": {
656
+ "group_size": 64,
657
+ "bits": 8
658
+ },
659
+ "model.layers.73.mlp.router.gate": {
660
+ "group_size": 64,
661
+ "bits": 8
662
+ },
663
+ "model.layers.74.mlp.router.gate": {
664
+ "group_size": 64,
665
+ "bits": 8
666
+ },
667
+ "model.layers.75.mlp.router.gate": {
668
+ "group_size": 64,
669
+ "bits": 8
670
+ },
671
+ "model.layers.76.mlp.router.gate": {
672
+ "group_size": 64,
673
+ "bits": 8
674
+ },
675
+ "model.layers.77.mlp.router.gate": {
676
+ "group_size": 64,
677
+ "bits": 8
678
+ },
679
+ "model.layers.78.mlp.router.gate": {
680
+ "group_size": 64,
681
+ "bits": 8
682
+ },
683
+ "model.layers.79.mlp.router.gate": {
684
+ "group_size": 64,
685
+ "bits": 8
686
+ }
687
+ },
688
+ "reap": {
689
+ "kept_per_layer": 48,
690
+ "orig_experts": 192
691
+ }
692
+ }
generation_config.json ADDED
@@ -0,0 +1,10 @@
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token_id": 120000,
3
+ "do_sample": true,
4
+ "eos_token_id": 120025,
5
+ "pad_token_id": 120002,
6
+ "temperature": 0.9,
7
+ "top_k": -1,
8
+ "top_p": 1,
9
+ "transformers_version": "5.6.0"
10
+ }
model-00001-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4396e4515ce0b3b276e2776898ea3dd6f8993e05f258990c4707f24831f62cfa
3
+ size 5298486170
model-00002-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d7e485af20da5617986a89ead2dcfe8560052d6eaf66dbb2b24a28c606a95a8
3
+ size 5289486702
model-00003-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8da0449acbe7899aedc1954a52240cabeca1ac331d4bdbe8f2aff1c08265d226
3
+ size 5236173257
model-00004-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30af72f092b0080afec4c2235ef39fc97b773e9c6701ab77c1720bc79d76775a
3
+ size 5236173287
model-00005-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e61506c6d7f32fdcb0e657e894df56fe4308a064b3ac43a6fc98f56bcf0b221
3
+ size 5289486772
model-00006-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ec46430582b6f27d479a5be7f9e9544b4c782684d7e4051f334184df6633fdb
3
+ size 5236173235
model-00007-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1242c9f69c13d003e4db2c2eadccd6a6267bf9eca48ed38c6c71c40cb54fda47
3
+ size 5236173261
model-00008-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea3c1f603f07752dabbd51bced4df26e9f766539dad544417e7342a6211a9d66
3
+ size 5289486764
model-00009-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d04018446f935fae39840e995d5fe0bef92f8b0333816cf28661918c327f3d9d
3
+ size 3050549059
model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
The diff for this file is too large to render. See raw diff