Hugging Face's logo

inference-optimization
/
Hy3-NVFP4

Safetensors
hy_v3
8-bit precision
compressed-tensors
Model card Files
xet
 Community
kylesayrs commited on
Commit
c0ae660
·
verified ·
1 Parent(s): de2a3b7

Upload folder using huggingface_hub

Browse files
Files changed (17) hide show
  1. .gitattributes +1 -0
  2. chat_template.jinja +222 -0
  3. config.json +581 -0
  4. generation_config.json +10 -0
  5. model-00001-of-00009.safetensors +3 -0
  6. model-00002-of-00009.safetensors +3 -0
  7. model-00003-of-00009.safetensors +3 -0
  8. model-00004-of-00009.safetensors +3 -0
  9. model-00005-of-00009.safetensors +3 -0
  10. model-00006-of-00009.safetensors +3 -0
  11. model-00007-of-00009.safetensors +3 -0
  12. model-00008-of-00009.safetensors +3 -0
  13. model-00009-of-00009.safetensors +3 -0
  14. model.safetensors.index.json +3 -0
  15. recipe.yaml +38 -0
  16. tokenizer.json +0 -0
  17. tokenizer_config.json +12 -0
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text
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 +1,581 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "HYV3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 120000,
8
+ "dtype": "bfloat16",
9
+ "enable_attention_fp32_softmax": false,
10
+ "enable_lm_head_fp32": true,
11
+ "enable_moe_fp32_combine": false,
12
+ "eod_token_id": 120026,
13
+ "eos_token_id": 120025,
14
+ "expert_hidden_dim": 1536,
15
+ "first_k_dense_replace": 1,
16
+ "head_dim": 128,
17
+ "hidden_act": "silu",
18
+ "hidden_size": 4096,
19
+ "initializer_range": 0.006,
20
+ "intermediate_size": 13312,
21
+ "max_position_embeddings": 262144,
22
+ "mlp_bias": false,
23
+ "mlp_layer_types": [
24
+ "dense",
25
+ "sparse",
26
+ "sparse",
27
+ "sparse",
28
+ "sparse",
29
+ "sparse",
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
+ ],
105
+ "model_type": "hy_v3",
106
+ "moe_intermediate_size": 1536,
107
+ "moe_router_enable_expert_bias": true,
108
+ "moe_router_use_sigmoid": true,
109
+ "num_attention_heads": 64,
110
+ "num_experts": 192,
111
+ "num_experts_per_tok": 8,
112
+ "num_hidden_layers": 80,
113
+ "num_key_value_heads": 8,
114
+ "num_nextn_predict_layers": 1,
115
+ "num_shared_experts": 1,
116
+ "output_router_logits": true,
117
+ "pad_token_id": 120002,
118
+ "qk_norm": true,
119
+ "quantization_config": {
120
+ "config_groups": {
121
+ "group_0": {
122
+ "format": "nvfp4-pack-quantized",
123
+ "input_activations": {
124
+ "actorder": null,
125
+ "block_structure": null,
126
+ "dynamic": "local",
127
+ "group_size": 16,
128
+ "num_bits": 4,
129
+ "observer": "static_minmax",
130
+ "observer_kwargs": {},
131
+ "scale_dtype": "torch.float8_e4m3fn",
132
+ "strategy": "tensor_group",
133
+ "symmetric": true,
134
+ "type": "float",
135
+ "zp_dtype": null
136
+ },
137
+ "output_activations": null,
138
+ "targets": [
139
+ "re:.*mlp.*$"
140
+ ],
141
+ "weights": {
142
+ "actorder": null,
143
+ "block_structure": null,
144
+ "dynamic": false,
145
+ "group_size": 16,
146
+ "num_bits": 4,
147
+ "observer": "memoryless_minmax",
148
+ "observer_kwargs": {},
149
+ "scale_dtype": "torch.float8_e4m3fn",
150
+ "strategy": "tensor_group",
151
+ "symmetric": true,
152
+ "type": "float",
153
+ "zp_dtype": null
154
+ }
155
+ }
156
+ },
157
+ "format": "nvfp4-pack-quantized",
158
+ "global_compression_ratio": null,
159
+ "ignore": [
160
+ "model.layers.0.self_attn.q_proj",
161
+ "model.layers.0.self_attn.k_proj",
162
+ "model.layers.0.self_attn.v_proj",
163
+ "model.layers.0.self_attn.o_proj",
164
+ "model.layers.1.self_attn.q_proj",
165
+ "model.layers.1.self_attn.k_proj",
166
+ "model.layers.1.self_attn.v_proj",
167
+ "model.layers.1.self_attn.o_proj",
168
+ "model.layers.1.mlp.gate",
169
+ "model.layers.2.self_attn.q_proj",
170
+ "model.layers.2.self_attn.k_proj",
171
+ "model.layers.2.self_attn.v_proj",
172
+ "model.layers.2.self_attn.o_proj",
173
+ "model.layers.2.mlp.gate",
174
+ "model.layers.3.self_attn.q_proj",
175
+ "model.layers.3.self_attn.k_proj",
176
+ "model.layers.3.self_attn.v_proj",
177
+ "model.layers.3.self_attn.o_proj",
178
+ "model.layers.3.mlp.gate",
179
+ "model.layers.4.self_attn.q_proj",
180
+ "model.layers.4.self_attn.k_proj",
181
+ "model.layers.4.self_attn.v_proj",
182
+ "model.layers.4.self_attn.o_proj",
183
+ "model.layers.4.mlp.gate",
184
+ "model.layers.5.self_attn.q_proj",
185
+ "model.layers.5.self_attn.k_proj",
186
+ "model.layers.5.self_attn.v_proj",
187
+ "model.layers.5.self_attn.o_proj",
188
+ "model.layers.5.mlp.gate",
189
+ "model.layers.6.self_attn.q_proj",
190
+ "model.layers.6.self_attn.k_proj",
191
+ "model.layers.6.self_attn.v_proj",
192
+ "model.layers.6.self_attn.o_proj",
193
+ "model.layers.6.mlp.gate",
194
+ "model.layers.7.self_attn.q_proj",
195
+ "model.layers.7.self_attn.k_proj",
196
+ "model.layers.7.self_attn.v_proj",
197
+ "model.layers.7.self_attn.o_proj",
198
+ "model.layers.7.mlp.gate",
199
+ "model.layers.8.self_attn.q_proj",
200
+ "model.layers.8.self_attn.k_proj",
201
+ "model.layers.8.self_attn.v_proj",
202
+ "model.layers.8.self_attn.o_proj",
203
+ "model.layers.8.mlp.gate",
204
+ "model.layers.9.self_attn.q_proj",
205
+ "model.layers.9.self_attn.k_proj",
206
+ "model.layers.9.self_attn.v_proj",
207
+ "model.layers.9.self_attn.o_proj",
208
+ "model.layers.9.mlp.gate",
209
+ "model.layers.10.self_attn.q_proj",
210
+ "model.layers.10.self_attn.k_proj",
211
+ "model.layers.10.self_attn.v_proj",
212
+ "model.layers.10.self_attn.o_proj",
213
+ "model.layers.10.mlp.gate",
214
+ "model.layers.11.self_attn.q_proj",
215
+ "model.layers.11.self_attn.k_proj",
216
+ "model.layers.11.self_attn.v_proj",
217
+ "model.layers.11.self_attn.o_proj",
218
+ "model.layers.11.mlp.gate",
219
+ "model.layers.12.self_attn.q_proj",
220
+ "model.layers.12.self_attn.k_proj",
221
+ "model.layers.12.self_attn.v_proj",
222
+ "model.layers.12.self_attn.o_proj",
223
+ "model.layers.12.mlp.gate",
224
+ "model.layers.13.self_attn.q_proj",
225
+ "model.layers.13.self_attn.k_proj",
226
+ "model.layers.13.self_attn.v_proj",
227
+ "model.layers.13.self_attn.o_proj",
228
+ "model.layers.13.mlp.gate",
229
+ "model.layers.14.self_attn.q_proj",
230
+ "model.layers.14.self_attn.k_proj",
231
+ "model.layers.14.self_attn.v_proj",
232
+ "model.layers.14.self_attn.o_proj",
233
+ "model.layers.14.mlp.gate",
234
+ "model.layers.15.self_attn.q_proj",
235
+ "model.layers.15.self_attn.k_proj",
236
+ "model.layers.15.self_attn.v_proj",
237
+ "model.layers.15.self_attn.o_proj",
238
+ "model.layers.15.mlp.gate",
239
+ "model.layers.16.self_attn.q_proj",
240
+ "model.layers.16.self_attn.k_proj",
241
+ "model.layers.16.self_attn.v_proj",
242
+ "model.layers.16.self_attn.o_proj",
243
+ "model.layers.16.mlp.gate",
244
+ "model.layers.17.self_attn.q_proj",
245
+ "model.layers.17.self_attn.k_proj",
246
+ "model.layers.17.self_attn.v_proj",
247
+ "model.layers.17.self_attn.o_proj",
248
+ "model.layers.17.mlp.gate",
249
+ "model.layers.18.self_attn.q_proj",
250
+ "model.layers.18.self_attn.k_proj",
251
+ "model.layers.18.self_attn.v_proj",
252
+ "model.layers.18.self_attn.o_proj",
253
+ "model.layers.18.mlp.gate",
254
+ "model.layers.19.self_attn.q_proj",
255
+ "model.layers.19.self_attn.k_proj",
256
+ "model.layers.19.self_attn.v_proj",
257
+ "model.layers.19.self_attn.o_proj",
258
+ "model.layers.19.mlp.gate",
259
+ "model.layers.20.self_attn.q_proj",
260
+ "model.layers.20.self_attn.k_proj",
261
+ "model.layers.20.self_attn.v_proj",
262
+ "model.layers.20.self_attn.o_proj",
263
+ "model.layers.20.mlp.gate",
264
+ "model.layers.21.self_attn.q_proj",
265
+ "model.layers.21.self_attn.k_proj",
266
+ "model.layers.21.self_attn.v_proj",
267
+ "model.layers.21.self_attn.o_proj",
268
+ "model.layers.21.mlp.gate",
269
+ "model.layers.22.self_attn.q_proj",
270
+ "model.layers.22.self_attn.k_proj",
271
+ "model.layers.22.self_attn.v_proj",
272
+ "model.layers.22.self_attn.o_proj",
273
+ "model.layers.22.mlp.gate",
274
+ "model.layers.23.self_attn.q_proj",
275
+ "model.layers.23.self_attn.k_proj",
276
+ "model.layers.23.self_attn.v_proj",
277
+ "model.layers.23.self_attn.o_proj",
278
+ "model.layers.23.mlp.gate",
279
+ "model.layers.24.self_attn.q_proj",
280
+ "model.layers.24.self_attn.k_proj",
281
+ "model.layers.24.self_attn.v_proj",
282
+ "model.layers.24.self_attn.o_proj",
283
+ "model.layers.24.mlp.gate",
284
+ "model.layers.25.self_attn.q_proj",
285
+ "model.layers.25.self_attn.k_proj",
286
+ "model.layers.25.self_attn.v_proj",
287
+ "model.layers.25.self_attn.o_proj",
288
+ "model.layers.25.mlp.gate",
289
+ "model.layers.26.self_attn.q_proj",
290
+ "model.layers.26.self_attn.k_proj",
291
+ "model.layers.26.self_attn.v_proj",
292
+ "model.layers.26.self_attn.o_proj",
293
+ "model.layers.26.mlp.gate",
294
+ "model.layers.27.self_attn.q_proj",
295
+ "model.layers.27.self_attn.k_proj",
296
+ "model.layers.27.self_attn.v_proj",
297
+ "model.layers.27.self_attn.o_proj",
298
+ "model.layers.27.mlp.gate",
299
+ "model.layers.28.self_attn.q_proj",
300
+ "model.layers.28.self_attn.k_proj",
301
+ "model.layers.28.self_attn.v_proj",
302
+ "model.layers.28.self_attn.o_proj",
303
+ "model.layers.28.mlp.gate",
304
+ "model.layers.29.self_attn.q_proj",
305
+ "model.layers.29.self_attn.k_proj",
306
+ "model.layers.29.self_attn.v_proj",
307
+ "model.layers.29.self_attn.o_proj",
308
+ "model.layers.29.mlp.gate",
309
+ "model.layers.30.self_attn.q_proj",
310
+ "model.layers.30.self_attn.k_proj",
311
+ "model.layers.30.self_attn.v_proj",
312
+ "model.layers.30.self_attn.o_proj",
313
+ "model.layers.30.mlp.gate",
314
+ "model.layers.31.self_attn.q_proj",
315
+ "model.layers.31.self_attn.k_proj",
316
+ "model.layers.31.self_attn.v_proj",
317
+ "model.layers.31.self_attn.o_proj",
318
+ "model.layers.31.mlp.gate",
319
+ "model.layers.32.self_attn.q_proj",
320
+ "model.layers.32.self_attn.k_proj",
321
+ "model.layers.32.self_attn.v_proj",
322
+ "model.layers.32.self_attn.o_proj",
323
+ "model.layers.32.mlp.gate",
324
+ "model.layers.33.self_attn.q_proj",
325
+ "model.layers.33.self_attn.k_proj",
326
+ "model.layers.33.self_attn.v_proj",
327
+ "model.layers.33.self_attn.o_proj",
328
+ "model.layers.33.mlp.gate",
329
+ "model.layers.34.self_attn.q_proj",
330
+ "model.layers.34.self_attn.k_proj",
331
+ "model.layers.34.self_attn.v_proj",
332
+ "model.layers.34.self_attn.o_proj",
333
+ "model.layers.34.mlp.gate",
334
+ "model.layers.35.self_attn.q_proj",
335
+ "model.layers.35.self_attn.k_proj",
336
+ "model.layers.35.self_attn.v_proj",
337
+ "model.layers.35.self_attn.o_proj",
338
+ "model.layers.35.mlp.gate",
339
+ "model.layers.36.self_attn.q_proj",
340
+ "model.layers.36.self_attn.k_proj",
341
+ "model.layers.36.self_attn.v_proj",
342
+ "model.layers.36.self_attn.o_proj",
343
+ "model.layers.36.mlp.gate",
344
+ "model.layers.37.self_attn.q_proj",
345
+ "model.layers.37.self_attn.k_proj",
346
+ "model.layers.37.self_attn.v_proj",
347
+ "model.layers.37.self_attn.o_proj",
348
+ "model.layers.37.mlp.gate",
349
+ "model.layers.38.self_attn.q_proj",
350
+ "model.layers.38.self_attn.k_proj",
351
+ "model.layers.38.self_attn.v_proj",
352
+ "model.layers.38.self_attn.o_proj",
353
+ "model.layers.38.mlp.gate",
354
+ "model.layers.39.self_attn.q_proj",
355
+ "model.layers.39.self_attn.k_proj",
356
+ "model.layers.39.self_attn.v_proj",
357
+ "model.layers.39.self_attn.o_proj",
358
+ "model.layers.39.mlp.gate",
359
+ "model.layers.40.self_attn.q_proj",
360
+ "model.layers.40.self_attn.k_proj",
361
+ "model.layers.40.self_attn.v_proj",
362
+ "model.layers.40.self_attn.o_proj",
363
+ "model.layers.40.mlp.gate",
364
+ "model.layers.41.self_attn.q_proj",
365
+ "model.layers.41.self_attn.k_proj",
366
+ "model.layers.41.self_attn.v_proj",
367
+ "model.layers.41.self_attn.o_proj",
368
+ "model.layers.41.mlp.gate",
369
+ "model.layers.42.self_attn.q_proj",
370
+ "model.layers.42.self_attn.k_proj",
371
+ "model.layers.42.self_attn.v_proj",
372
+ "model.layers.42.self_attn.o_proj",
373
+ "model.layers.42.mlp.gate",
374
+ "model.layers.43.self_attn.q_proj",
375
+ "model.layers.43.self_attn.k_proj",
376
+ "model.layers.43.self_attn.v_proj",
377
+ "model.layers.43.self_attn.o_proj",
378
+ "model.layers.43.mlp.gate",
379
+ "model.layers.44.self_attn.q_proj",
380
+ "model.layers.44.self_attn.k_proj",
381
+ "model.layers.44.self_attn.v_proj",
382
+ "model.layers.44.self_attn.o_proj",
383
+ "model.layers.44.mlp.gate",
384
+ "model.layers.45.self_attn.q_proj",
385
+ "model.layers.45.self_attn.k_proj",
386
+ "model.layers.45.self_attn.v_proj",
387
+ "model.layers.45.self_attn.o_proj",
388
+ "model.layers.45.mlp.gate",
389
+ "model.layers.46.self_attn.q_proj",
390
+ "model.layers.46.self_attn.k_proj",
391
+ "model.layers.46.self_attn.v_proj",
392
+ "model.layers.46.self_attn.o_proj",
393
+ "model.layers.46.mlp.gate",
394
+ "model.layers.47.self_attn.q_proj",
395
+ "model.layers.47.self_attn.k_proj",
396
+ "model.layers.47.self_attn.v_proj",
397
+ "model.layers.47.self_attn.o_proj",
398
+ "model.layers.47.mlp.gate",
399
+ "model.layers.48.self_attn.q_proj",
400
+ "model.layers.48.self_attn.k_proj",
401
+ "model.layers.48.self_attn.v_proj",
402
+ "model.layers.48.self_attn.o_proj",
403
+ "model.layers.48.mlp.gate",
404
+ "model.layers.49.self_attn.q_proj",
405
+ "model.layers.49.self_attn.k_proj",
406
+ "model.layers.49.self_attn.v_proj",
407
+ "model.layers.49.self_attn.o_proj",
408
+ "model.layers.49.mlp.gate",
409
+ "model.layers.50.self_attn.q_proj",
410
+ "model.layers.50.self_attn.k_proj",
411
+ "model.layers.50.self_attn.v_proj",
412
+ "model.layers.50.self_attn.o_proj",
413
+ "model.layers.50.mlp.gate",
414
+ "model.layers.51.self_attn.q_proj",
415
+ "model.layers.51.self_attn.k_proj",
416
+ "model.layers.51.self_attn.v_proj",
417
+ "model.layers.51.self_attn.o_proj",
418
+ "model.layers.51.mlp.gate",
419
+ "model.layers.52.self_attn.q_proj",
420
+ "model.layers.52.self_attn.k_proj",
421
+ "model.layers.52.self_attn.v_proj",
422
+ "model.layers.52.self_attn.o_proj",
423
+ "model.layers.52.mlp.gate",
424
+ "model.layers.53.self_attn.q_proj",
425
+ "model.layers.53.self_attn.k_proj",
426
+ "model.layers.53.self_attn.v_proj",
427
+ "model.layers.53.self_attn.o_proj",
428
+ "model.layers.53.mlp.gate",
429
+ "model.layers.54.self_attn.q_proj",
430
+ "model.layers.54.self_attn.k_proj",
431
+ "model.layers.54.self_attn.v_proj",
432
+ "model.layers.54.self_attn.o_proj",
433
+ "model.layers.54.mlp.gate",
434
+ "model.layers.55.self_attn.q_proj",
435
+ "model.layers.55.self_attn.k_proj",
436
+ "model.layers.55.self_attn.v_proj",
437
+ "model.layers.55.self_attn.o_proj",
438
+ "model.layers.55.mlp.gate",
439
+ "model.layers.56.self_attn.q_proj",
440
+ "model.layers.56.self_attn.k_proj",
441
+ "model.layers.56.self_attn.v_proj",
442
+ "model.layers.56.self_attn.o_proj",
443
+ "model.layers.56.mlp.gate",
444
+ "model.layers.57.self_attn.q_proj",
445
+ "model.layers.57.self_attn.k_proj",
446
+ "model.layers.57.self_attn.v_proj",
447
+ "model.layers.57.self_attn.o_proj",
448
+ "model.layers.57.mlp.gate",
449
+ "model.layers.58.self_attn.q_proj",
450
+ "model.layers.58.self_attn.k_proj",
451
+ "model.layers.58.self_attn.v_proj",
452
+ "model.layers.58.self_attn.o_proj",
453
+ "model.layers.58.mlp.gate",
454
+ "model.layers.59.self_attn.q_proj",
455
+ "model.layers.59.self_attn.k_proj",
456
+ "model.layers.59.self_attn.v_proj",
457
+ "model.layers.59.self_attn.o_proj",
458
+ "model.layers.59.mlp.gate",
459
+ "model.layers.60.self_attn.q_proj",
460
+ "model.layers.60.self_attn.k_proj",
461
+ "model.layers.60.self_attn.v_proj",
462
+ "model.layers.60.self_attn.o_proj",
463
+ "model.layers.60.mlp.gate",
464
+ "model.layers.61.self_attn.q_proj",
465
+ "model.layers.61.self_attn.k_proj",
466
+ "model.layers.61.self_attn.v_proj",
467
+ "model.layers.61.self_attn.o_proj",
468
+ "model.layers.61.mlp.gate",
469
+ "model.layers.62.self_attn.q_proj",
470
+ "model.layers.62.self_attn.k_proj",
471
+ "model.layers.62.self_attn.v_proj",
472
+ "model.layers.62.self_attn.o_proj",
473
+ "model.layers.62.mlp.gate",
474
+ "model.layers.63.self_attn.q_proj",
475
+ "model.layers.63.self_attn.k_proj",
476
+ "model.layers.63.self_attn.v_proj",
477
+ "model.layers.63.self_attn.o_proj",
478
+ "model.layers.63.mlp.gate",
479
+ "model.layers.64.self_attn.q_proj",
480
+ "model.layers.64.self_attn.k_proj",
481
+ "model.layers.64.self_attn.v_proj",
482
+ "model.layers.64.self_attn.o_proj",
483
+ "model.layers.64.mlp.gate",
484
+ "model.layers.65.self_attn.q_proj",
485
+ "model.layers.65.self_attn.k_proj",
486
+ "model.layers.65.self_attn.v_proj",
487
+ "model.layers.65.self_attn.o_proj",
488
+ "model.layers.65.mlp.gate",
489
+ "model.layers.66.self_attn.q_proj",
490
+ "model.layers.66.self_attn.k_proj",
491
+ "model.layers.66.self_attn.v_proj",
492
+ "model.layers.66.self_attn.o_proj",
493
+ "model.layers.66.mlp.gate",
494
+ "model.layers.67.self_attn.q_proj",
495
+ "model.layers.67.self_attn.k_proj",
496
+ "model.layers.67.self_attn.v_proj",
497
+ "model.layers.67.self_attn.o_proj",
498
+ "model.layers.67.mlp.gate",
499
+ "model.layers.68.self_attn.q_proj",
500
+ "model.layers.68.self_attn.k_proj",
501
+ "model.layers.68.self_attn.v_proj",
502
+ "model.layers.68.self_attn.o_proj",
503
+ "model.layers.68.mlp.gate",
504
+ "model.layers.69.self_attn.q_proj",
505
+ "model.layers.69.self_attn.k_proj",
506
+ "model.layers.69.self_attn.v_proj",
507
+ "model.layers.69.self_attn.o_proj",
508
+ "model.layers.69.mlp.gate",
509
+ "model.layers.70.self_attn.q_proj",
510
+ "model.layers.70.self_attn.k_proj",
511
+ "model.layers.70.self_attn.v_proj",
512
+ "model.layers.70.self_attn.o_proj",
513
+ "model.layers.70.mlp.gate",
514
+ "model.layers.71.self_attn.q_proj",
515
+ "model.layers.71.self_attn.k_proj",
516
+ "model.layers.71.self_attn.v_proj",
517
+ "model.layers.71.self_attn.o_proj",
518
+ "model.layers.71.mlp.gate",
519
+ "model.layers.72.self_attn.q_proj",
520
+ "model.layers.72.self_attn.k_proj",
521
+ "model.layers.72.self_attn.v_proj",
522
+ "model.layers.72.self_attn.o_proj",
523
+ "model.layers.72.mlp.gate",
524
+ "model.layers.73.self_attn.q_proj",
525
+ "model.layers.73.self_attn.k_proj",
526
+ "model.layers.73.self_attn.v_proj",
527
+ "model.layers.73.self_attn.o_proj",
528
+ "model.layers.73.mlp.gate",
529
+ "model.layers.74.self_attn.q_proj",
530
+ "model.layers.74.self_attn.k_proj",
531
+ "model.layers.74.self_attn.v_proj",
532
+ "model.layers.74.self_attn.o_proj",
533
+ "model.layers.74.mlp.gate",
534
+ "model.layers.75.self_attn.q_proj",
535
+ "model.layers.75.self_attn.k_proj",
536
+ "model.layers.75.self_attn.v_proj",
537
+ "model.layers.75.self_attn.o_proj",
538
+ "model.layers.75.mlp.gate",
539
+ "model.layers.76.self_attn.q_proj",
540
+ "model.layers.76.self_attn.k_proj",
541
+ "model.layers.76.self_attn.v_proj",
542
+ "model.layers.76.self_attn.o_proj",
543
+ "model.layers.76.mlp.gate",
544
+ "model.layers.77.self_attn.q_proj",
545
+ "model.layers.77.self_attn.k_proj",
546
+ "model.layers.77.self_attn.v_proj",
547
+ "model.layers.77.self_attn.o_proj",
548
+ "model.layers.77.mlp.gate",
549
+ "model.layers.78.self_attn.q_proj",
550
+ "model.layers.78.self_attn.k_proj",
551
+ "model.layers.78.self_attn.v_proj",
552
+ "model.layers.78.self_attn.o_proj",
553
+ "model.layers.78.mlp.gate",
554
+ "model.layers.79.self_attn.q_proj",
555
+ "model.layers.79.self_attn.k_proj",
556
+ "model.layers.79.self_attn.v_proj",
557
+ "model.layers.79.self_attn.o_proj",
558
+ "model.layers.79.mlp.gate",
559
+ "lm_head"
560
+ ],
561
+ "kv_cache_scheme": null,
562
+ "quant_method": "compressed-tensors",
563
+ "quantization_status": "compressed",
564
+ "sparsity_config": {},
565
+ "transform_config": {},
566
+ "version": "0.18.1.dev12+gbc35079"
567
+ },
568
+ "rms_norm_eps": 1e-05,
569
+ "rope_parameters": {
570
+ "rope_theta": 11158840.0,
571
+ "rope_type": "default"
572
+ },
573
+ "route_norm": true,
574
+ "router_scaling_factor": 2.826,
575
+ "sep_token_id": 120007,
576
+ "tie_word_embeddings": false,
577
+ "transformers_version": "5.15.1",
578
+ "use_cache": true,
579
+ "use_grouped_mm": false,
580
+ "vocab_size": 120832
581
+ }
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": "5.15.1"
10
+ }
model-00001-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a47d623ec616f21e9659aedac0fe9e3f55e69b83a01f7056944422ad92ad168e
3
+ size 20002147400
model-00002-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4bbc96dadfbf12fa4c977dee2fcd5b541880ac51505711552294d81e08fd5671
3
+ size 20001817600
model-00003-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66364f94b0d6b4026f376b7b22175a1f01315b561107a040f687453f838c7262
3
+ size 20001425232
model-00004-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:92885dd03d75dad7b78d3a463dbb3c296a64e5cdd5252ddf89616055928a4df0
3
+ size 20001425232
model-00005-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d35e661d67ecc5c5e5e85d5eb6275140b3e8dfaa998dd2e0d404255d5824a9c
3
+ size 20001425240
model-00006-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d1f4678d20880a00f39cde41af4c0042a0581e53ea7a62722624679ab4cde515
3
+ size 20001425232
model-00007-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:698a741438e7667170f5db885da382e9004fe71bb066d3159cead98dc29fe428
3
+ size 19944810256
model-00008-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0af17d57703a09f08abe39a5ce9db1e99c07b1db22f1ac8ed001e045125c402c
3
+ size 20001798408
model-00009-of-00009.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9af1dbbe657a849fb4022264b4be7da78bbb1de1091335a686472f5e2099ac25
3
+ size 16218949848
model.safetensors.index.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1670c9c46dd31a9e60fa860a63a9ff4fd7a89d05a9234309f7f7992d5e1d8e48
3
+ size 18390140
recipe.yaml ADDED
@@ -0,0 +1,38 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ default_stage:
2
+ default_modifiers:
3
+ QuantizationModifier:
4
+ config_groups:
5
+ experts:
6
+ targets: ['re:.*mlp.*$']
7
+ weights:
8
+ num_bits: 4
9
+ type: float
10
+ symmetric: true
11
+ group_size: 16
12
+ strategy: tensor_group
13
+ block_structure: null
14
+ dynamic: false
15
+ actorder: null
16
+ scale_dtype: torch.float8_e4m3fn
17
+ zp_dtype: null
18
+ observer: memoryless_minmax
19
+ observer_kwargs: {}
20
+ input_activations:
21
+ num_bits: 4
22
+ type: float
23
+ symmetric: true
24
+ group_size: 16
25
+ strategy: tensor_group
26
+ block_structure: null
27
+ dynamic: local
28
+ actorder: null
29
+ scale_dtype: torch.float8_e4m3fn
30
+ zp_dtype: null
31
+ observer: static_minmax
32
+ observer_kwargs: {}
33
+ output_activations: null
34
+ format: null
35
+ targets: [Linear]
36
+ ignore: ['re:.*lm_head', 're:.*mlp.gate$']
37
+ bypass_divisibility_checks: false
38
+ requires_calibration_data: true
tokenizer.json ADDED
The diff for this file is too large to render. See raw diff
 
tokenizer_config.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "bos_token": "<｜hy_begin_of_sentence:opensource｜>",
4
+ "clean_up_tokenization_spaces": false,
5
+ "eos_token": "<｜hy_eos:opensource｜>",
6
+ "is_local": false,
7
+ "local_files_only": false,
8
+ "model_max_length": 1000000000000000019884624838656,
9
+ "pad_token": "<｜hy_pad:opensource｜>",
10
+ "token_suffix": ":opensource",
11
+ "tokenizer_class": "TokenizersBackend"
12
+ }