Hugging Face's logo

amd
/
DeepSeek-R1-0528-MXFP4-ASQ

Safetensors
deepseek_v3
custom_code
8-bit precision
quark
Model card Files
xet
Community
2

upload model

#1
by linzhao-amd - opened
base: refs/heads/main
←
from: refs/pr/1
Discussion Files changed
+744156
-0
This view is limited to 50 files because it contains too many changes.  See the raw diff here.
Files changed (50) hide show
  1. chat_template.jinja +101 -0
  2. config.json +451 -0
  3. configuration_deepseek.py +199 -0
  4. generation_config.json +9 -0
  5. model-00001-of-00076.safetensors +3 -0
  6. model-00002-of-00076.safetensors +3 -0
  7. model-00003-of-00076.safetensors +3 -0
  8. model-00004-of-00076.safetensors +3 -0
  9. model-00005-of-00076.safetensors +3 -0
  10. model-00006-of-00076.safetensors +3 -0
  11. model-00007-of-00076.safetensors +3 -0
  12. model-00008-of-00076.safetensors +3 -0
  13. model-00009-of-00076.safetensors +3 -0
  14. model-00010-of-00076.safetensors +3 -0
  15. model-00011-of-00076.safetensors +3 -0
  16. model-00012-of-00076.safetensors +3 -0
  17. model-00013-of-00076.safetensors +3 -0
  18. model-00014-of-00076.safetensors +3 -0
  19. model-00015-of-00076.safetensors +3 -0
  20. model-00016-of-00076.safetensors +3 -0
  21. model-00017-of-00076.safetensors +3 -0
  22. model-00018-of-00076.safetensors +3 -0
  23. model-00019-of-00076.safetensors +3 -0
  24. model-00020-of-00076.safetensors +3 -0
  25. model-00021-of-00076.safetensors +3 -0
  26. model-00022-of-00076.safetensors +3 -0
  27. model-00023-of-00076.safetensors +3 -0
  28. model-00024-of-00076.safetensors +3 -0
  29. model-00025-of-00076.safetensors +3 -0
  30. model-00026-of-00076.safetensors +3 -0
  31. model-00027-of-00076.safetensors +3 -0
  32. model-00028-of-00076.safetensors +3 -0
  33. model-00029-of-00076.safetensors +3 -0
  34. model-00030-of-00076.safetensors +3 -0
  35. model-00031-of-00076.safetensors +3 -0
  36. model-00032-of-00076.safetensors +3 -0
  37. model-00033-of-00076.safetensors +3 -0
  38. model-00034-of-00076.safetensors +3 -0
  39. model-00035-of-00076.safetensors +3 -0
  40. model-00036-of-00076.safetensors +3 -0
  41. model-00037-of-00076.safetensors +3 -0
  42. model-00038-of-00076.safetensors +3 -0
  43. model-00039-of-00076.safetensors +3 -0
  44. model-00040-of-00076.safetensors +3 -0
  45. model-00041-of-00076.safetensors +3 -0
  46. model-00042-of-00076.safetensors +3 -0
  47. model-00043-of-00076.safetensors +3 -0
  48. model-00044-of-00076.safetensors +3 -0
  49. model-00045-of-00076.safetensors +3 -0
  50. model-00046-of-00076.safetensors +3 -0
chat_template.jinja ADDED
@@ -0,0 +1,101 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- if not add_generation_prompt is defined %}
2
+ {%- set add_generation_prompt = false %}
3
+ {%- endif %}
4
+ {%- set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='', is_first_sp=true, is_last_user=false) %}
5
+ {%- for message in messages %}
6
+ {%- if message['role'] == 'system' %}
7
+ {%- if ns.is_first_sp %}
8
+ {%- set ns.system_prompt = ns.system_prompt + message['content'] %}
9
+ {%- set ns.is_first_sp = false %}
10
+ {%- else %}
11
+ {%- set ns.system_prompt = ns.system_prompt + '\n\n' + message['content'] %}
12
+ {%- endif %}
13
+ {%- endif %}
14
+ {%- endfor %}
15
+
16
+ {#- Adapted from https://github.com/sgl-project/sglang/blob/main/examples/chat_template/tool_chat_template_deepseekr1.jinja #}
17
+ {%- if tools is defined and tools is not none %}
18
+ {%- set tool_ns = namespace(text='You are a helpful assistant with tool calling capabilities. ' + 'When a tool call is needed, you MUST use the following format to issue the call:\n' + '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>function<｜tool▁sep｜>FUNCTION_NAME\n' + '```json\n{"param1": "value1", "param2": "value2"}\n```<｜tool▁call▁end｜><｜tool▁calls▁end｜>\n\n' + 'Make sure the JSON is valid.' + '## Tools\n\n### Function\n\nYou have the following functions available:\n\n') %}
19
+ {%- for tool in tools %}
20
+ {%- set tool_ns.text = tool_ns.text + '\n```json\n' + (tool | tojson) + '\n```\n' %}
21
+ {%- endfor %}
22
+ {%- if ns.system_prompt|length != 0 %}
23
+ {%- set ns.system_prompt = ns.system_prompt + '\n\n' + tool_ns.text %}
24
+ {%- else %}
25
+ {%- set ns.system_prompt = tool_ns.text %}
26
+ {%- endif %}
27
+ {%- endif %}
28
+ {{- bos_token }}
29
+ {{- ns.system_prompt }}
30
+ {%- set last_index = (messages|length - 1) %}
31
+ {%- for message in messages %}
32
+ {%- set content = message['content'] %}
33
+ {%- if message['role'] == 'user' %}
34
+ {%- set ns.is_tool = false -%}
35
+ {%- set ns.is_first = false -%}
36
+ {%- set ns.is_last_user = true -%}
37
+ {%- if loop.index0 == last_index %}
38
+ {{- '<｜User｜>' + content }}
39
+ {%- else %}
40
+ {{- '<｜User｜>' + content + '<｜Assistant｜>'}}
41
+ {%- endif %}
42
+ {%- endif %}
43
+ {%- if message['role'] == 'assistant' %}
44
+ {%- if '</think>' in content %}
45
+ {%- set content = (content.split('</think>')|last) %}
46
+ {%- endif %}
47
+ {%- endif %}
48
+ {%- if message['role'] == 'assistant' and message['tool_calls'] is defined and message['tool_calls'] is not none %}
49
+ {%- set ns.is_last_user = false -%}
50
+ {%- if ns.is_tool %}
51
+ {{- '<｜tool▁outputs▁end｜>'}}
52
+ {%- endif %}
53
+ {%- set ns.is_first = false %}
54
+ {%- set ns.is_tool = false -%}
55
+ {%- set ns.is_output_first = true %}
56
+ {%- for tool in message['tool_calls'] %}
57
+ {%- set arguments = tool['function']['arguments'] %}
58
+ {%- if arguments is not string %}
59
+ {%- set arguments = arguments|tojson %}
60
+ {%- endif %}
61
+ {%- if not ns.is_first %}
62
+ {%- if content is none %}
63
+ {{- '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + arguments + '\n' + '```' + '<｜tool▁call▁end｜>'}}
64
+ }
65
+ {%- else %}
66
+ {{- content + '<｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + arguments + '\n' + '```' + '<｜tool▁call▁end｜>'}}
67
+ {%- endif %}
68
+ {%- set ns.is_first = true -%}
69
+ {%- else %}
70
+ {{- '\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '```json' + '\n' + arguments + '\n' + '```' + '<｜tool▁call▁end｜>'}}
71
+ {%- endif %}
72
+ {%- endfor %}
73
+ {{- '<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}
74
+ {%- endif %}
75
+ {%- if message['role'] == 'assistant' and (message['tool_calls'] is not defined or message['tool_calls'] is none) %}
76
+ {%- set ns.is_last_user = false -%}
77
+ {%- if ns.is_tool %}
78
+ {{- '<｜tool▁outputs▁end｜>' + content + '<｜end▁of▁sentence｜>'}}
79
+ {%- set ns.is_tool = false -%}
80
+ {%- else %}
81
+ {{- content + '<｜end▁of▁sentence｜>'}}
82
+ {%- endif %}
83
+ {%- endif %}
84
+ {%- if message['role'] == 'tool' %}
85
+ {%- set ns.is_last_user = false -%}
86
+ {%- set ns.is_tool = true -%}
87
+ {%- if ns.is_output_first %}
88
+ {{- '<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + content + '<｜tool▁output▁end｜>'}}
89
+ {%- set ns.is_output_first = false %}
90
+ {%- else %}
91
+ {{- '\n<｜tool▁output▁begin｜>' + content + '<｜tool▁output▁end｜>'}}
92
+ {%- endif %}
93
+ {%- endif %}
94
+ {%- endfor -%}
95
+ {%- if ns.is_tool %}
96
+ {{- '<｜tool▁outputs▁end｜>'}}
97
+ {%- endif %}
98
+ {#- if add_generation_prompt and not ns.is_last_user and not ns.is_tool #}
99
+ {%- if add_generation_prompt and not ns.is_tool %}
100
+ {{- '<｜Assistant｜>'}}
101
+ {%- endif %}
config.json ADDED
@@ -0,0 +1,451 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "DeepseekV3ForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "auto_map": {
8
+ "AutoConfig": "configuration_deepseek.DeepseekV3Config",
9
+ "AutoModel": "modeling_deepseek.DeepseekV3Model",
10
+ "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
11
+ },
12
+ "bos_token_id": 0,
13
+ "eos_token_id": 1,
14
+ "ep_size": 1,
15
+ "first_k_dense_replace": 3,
16
+ "hidden_act": "silu",
17
+ "hidden_size": 7168,
18
+ "initializer_range": 0.02,
19
+ "intermediate_size": 18432,
20
+ "kv_lora_rank": 512,
21
+ "max_position_embeddings": 163840,
22
+ "model_type": "deepseek_v3",
23
+ "moe_intermediate_size": 2048,
24
+ "moe_layer_freq": 1,
25
+ "n_group": 8,
26
+ "n_routed_experts": 256,
27
+ "n_shared_experts": 1,
28
+ "norm_topk_prob": true,
29
+ "num_attention_heads": 128,
30
+ "num_experts_per_tok": 8,
31
+ "num_hidden_layers": 61,
32
+ "num_key_value_heads": 128,
33
+ "num_nextn_predict_layers": 1,
34
+ "pad_token_id": 2,
35
+ "q_lora_rank": 1536,
36
+ "qk_nope_head_dim": 128,
37
+ "qk_rope_head_dim": 64,
38
+ "quantization_config": {
39
+ "algo_config": [
40
+ {
41
+ "compute_scale_loss": "MAE",
42
+ "model_decoder_layers": "model.layers",
43
+ "name": "autosmoothquant",
44
+ "scaling_layers": [
45
+ {
46
+ "inp": "mlp.gate_proj",
47
+ "layers": [
48
+ "mlp.gate_proj",
49
+ "mlp.up_proj"
50
+ ],
51
+ "module2inspect": "mlp",
52
+ "prev_op": "post_attention_layernorm"
53
+ },
54
+ {
55
+ "inp": "mlp.down_proj",
56
+ "layers": [
57
+ "mlp.down_proj"
58
+ ],
59
+ "prev_op": "mlp.up_proj"
60
+ },
61
+ {
62
+ "inp": "down_proj",
63
+ "layers": [
64
+ "down_proj"
65
+ ],
66
+ "prev_op": "up_proj"
67
+ }
68
+ ]
69
+ }
70
+ ],
71
+ "exclude": [
72
+ "model.layers.0.self_attn.q_a_proj",
73
+ "model.layers.0.self_attn.q_b_proj",
74
+ "model.layers.0.self_attn.kv_a_proj_with_mqa",
75
+ "model.layers.0.self_attn.kv_b_proj",
76
+ "model.layers.0.self_attn.o_proj",
77
+ "model.layers.1.self_attn.q_a_proj",
78
+ "model.layers.1.self_attn.q_b_proj",
79
+ "model.layers.1.self_attn.kv_a_proj_with_mqa",
80
+ "model.layers.1.self_attn.kv_b_proj",
81
+ "model.layers.1.self_attn.o_proj",
82
+ "model.layers.2.self_attn.q_a_proj",
83
+ "model.layers.2.self_attn.q_b_proj",
84
+ "model.layers.2.self_attn.kv_a_proj_with_mqa",
85
+ "model.layers.2.self_attn.kv_b_proj",
86
+ "model.layers.2.self_attn.o_proj",
87
+ "model.layers.3.self_attn.q_a_proj",
88
+ "model.layers.3.self_attn.q_b_proj",
89
+ "model.layers.3.self_attn.kv_a_proj_with_mqa",
90
+ "model.layers.3.self_attn.kv_b_proj",
91
+ "model.layers.3.self_attn.o_proj",
92
+ "model.layers.4.self_attn.q_a_proj",
93
+ "model.layers.4.self_attn.q_b_proj",
94
+ "model.layers.4.self_attn.kv_a_proj_with_mqa",
95
+ "model.layers.4.self_attn.kv_b_proj",
96
+ "model.layers.4.self_attn.o_proj",
97
+ "model.layers.5.self_attn.q_a_proj",
98
+ "model.layers.5.self_attn.q_b_proj",
99
+ "model.layers.5.self_attn.kv_a_proj_with_mqa",
100
+ "model.layers.5.self_attn.kv_b_proj",
101
+ "model.layers.5.self_attn.o_proj",
102
+ "model.layers.6.self_attn.q_a_proj",
103
+ "model.layers.6.self_attn.q_b_proj",
104
+ "model.layers.6.self_attn.kv_a_proj_with_mqa",
105
+ "model.layers.6.self_attn.kv_b_proj",
106
+ "model.layers.6.self_attn.o_proj",
107
+ "model.layers.7.self_attn.q_a_proj",
108
+ "model.layers.7.self_attn.q_b_proj",
109
+ "model.layers.7.self_attn.kv_a_proj_with_mqa",
110
+ "model.layers.7.self_attn.kv_b_proj",
111
+ "model.layers.7.self_attn.o_proj",
112
+ "model.layers.8.self_attn.q_a_proj",
113
+ "model.layers.8.self_attn.q_b_proj",
114
+ "model.layers.8.self_attn.kv_a_proj_with_mqa",
115
+ "model.layers.8.self_attn.kv_b_proj",
116
+ "model.layers.8.self_attn.o_proj",
117
+ "model.layers.9.self_attn.q_a_proj",
118
+ "model.layers.9.self_attn.q_b_proj",
119
+ "model.layers.9.self_attn.kv_a_proj_with_mqa",
120
+ "model.layers.9.self_attn.kv_b_proj",
121
+ "model.layers.9.self_attn.o_proj",
122
+ "model.layers.10.self_attn.q_a_proj",
123
+ "model.layers.10.self_attn.q_b_proj",
124
+ "model.layers.10.self_attn.kv_a_proj_with_mqa",
125
+ "model.layers.10.self_attn.kv_b_proj",
126
+ "model.layers.10.self_attn.o_proj",
127
+ "model.layers.11.self_attn.q_a_proj",
128
+ "model.layers.11.self_attn.q_b_proj",
129
+ "model.layers.11.self_attn.kv_a_proj_with_mqa",
130
+ "model.layers.11.self_attn.kv_b_proj",
131
+ "model.layers.11.self_attn.o_proj",
132
+ "model.layers.12.self_attn.q_a_proj",
133
+ "model.layers.12.self_attn.q_b_proj",
134
+ "model.layers.12.self_attn.kv_a_proj_with_mqa",
135
+ "model.layers.12.self_attn.kv_b_proj",
136
+ "model.layers.12.self_attn.o_proj",
137
+ "model.layers.13.self_attn.q_a_proj",
138
+ "model.layers.13.self_attn.q_b_proj",
139
+ "model.layers.13.self_attn.kv_a_proj_with_mqa",
140
+ "model.layers.13.self_attn.kv_b_proj",
141
+ "model.layers.13.self_attn.o_proj",
142
+ "model.layers.14.self_attn.q_a_proj",
143
+ "model.layers.14.self_attn.q_b_proj",
144
+ "model.layers.14.self_attn.kv_a_proj_with_mqa",
145
+ "model.layers.14.self_attn.kv_b_proj",
146
+ "model.layers.14.self_attn.o_proj",
147
+ "model.layers.15.self_attn.q_a_proj",
148
+ "model.layers.15.self_attn.q_b_proj",
149
+ "model.layers.15.self_attn.kv_a_proj_with_mqa",
150
+ "model.layers.15.self_attn.kv_b_proj",
151
+ "model.layers.15.self_attn.o_proj",
152
+ "model.layers.16.self_attn.q_a_proj",
153
+ "model.layers.16.self_attn.q_b_proj",
154
+ "model.layers.16.self_attn.kv_a_proj_with_mqa",
155
+ "model.layers.16.self_attn.kv_b_proj",
156
+ "model.layers.16.self_attn.o_proj",
157
+ "model.layers.17.self_attn.q_a_proj",
158
+ "model.layers.17.self_attn.q_b_proj",
159
+ "model.layers.17.self_attn.kv_a_proj_with_mqa",
160
+ "model.layers.17.self_attn.kv_b_proj",
161
+ "model.layers.17.self_attn.o_proj",
162
+ "model.layers.18.self_attn.q_a_proj",
163
+ "model.layers.18.self_attn.q_b_proj",
164
+ "model.layers.18.self_attn.kv_a_proj_with_mqa",
165
+ "model.layers.18.self_attn.kv_b_proj",
166
+ "model.layers.18.self_attn.o_proj",
167
+ "model.layers.19.self_attn.q_a_proj",
168
+ "model.layers.19.self_attn.q_b_proj",
169
+ "model.layers.19.self_attn.kv_a_proj_with_mqa",
170
+ "model.layers.19.self_attn.kv_b_proj",
171
+ "model.layers.19.self_attn.o_proj",
172
+ "model.layers.20.self_attn.q_a_proj",
173
+ "model.layers.20.self_attn.q_b_proj",
174
+ "model.layers.20.self_attn.kv_a_proj_with_mqa",
175
+ "model.layers.20.self_attn.kv_b_proj",
176
+ "model.layers.20.self_attn.o_proj",
177
+ "model.layers.21.self_attn.q_a_proj",
178
+ "model.layers.21.self_attn.q_b_proj",
179
+ "model.layers.21.self_attn.kv_a_proj_with_mqa",
180
+ "model.layers.21.self_attn.kv_b_proj",
181
+ "model.layers.21.self_attn.o_proj",
182
+ "model.layers.22.self_attn.q_a_proj",
183
+ "model.layers.22.self_attn.q_b_proj",
184
+ "model.layers.22.self_attn.kv_a_proj_with_mqa",
185
+ "model.layers.22.self_attn.kv_b_proj",
186
+ "model.layers.22.self_attn.o_proj",
187
+ "model.layers.23.self_attn.q_a_proj",
188
+ "model.layers.23.self_attn.q_b_proj",
189
+ "model.layers.23.self_attn.kv_a_proj_with_mqa",
190
+ "model.layers.23.self_attn.kv_b_proj",
191
+ "model.layers.23.self_attn.o_proj",
192
+ "model.layers.24.self_attn.q_a_proj",
193
+ "model.layers.24.self_attn.q_b_proj",
194
+ "model.layers.24.self_attn.kv_a_proj_with_mqa",
195
+ "model.layers.24.self_attn.kv_b_proj",
196
+ "model.layers.24.self_attn.o_proj",
197
+ "model.layers.25.self_attn.q_a_proj",
198
+ "model.layers.25.self_attn.q_b_proj",
199
+ "model.layers.25.self_attn.kv_a_proj_with_mqa",
200
+ "model.layers.25.self_attn.kv_b_proj",
201
+ "model.layers.25.self_attn.o_proj",
202
+ "model.layers.26.self_attn.q_a_proj",
203
+ "model.layers.26.self_attn.q_b_proj",
204
+ "model.layers.26.self_attn.kv_a_proj_with_mqa",
205
+ "model.layers.26.self_attn.kv_b_proj",
206
+ "model.layers.26.self_attn.o_proj",
207
+ "model.layers.27.self_attn.q_a_proj",
208
+ "model.layers.27.self_attn.q_b_proj",
209
+ "model.layers.27.self_attn.kv_a_proj_with_mqa",
210
+ "model.layers.27.self_attn.kv_b_proj",
211
+ "model.layers.27.self_attn.o_proj",
212
+ "model.layers.28.self_attn.q_a_proj",
213
+ "model.layers.28.self_attn.q_b_proj",
214
+ "model.layers.28.self_attn.kv_a_proj_with_mqa",
215
+ "model.layers.28.self_attn.kv_b_proj",
216
+ "model.layers.28.self_attn.o_proj",
217
+ "model.layers.29.self_attn.q_a_proj",
218
+ "model.layers.29.self_attn.q_b_proj",
219
+ "model.layers.29.self_attn.kv_a_proj_with_mqa",
220
+ "model.layers.29.self_attn.kv_b_proj",
221
+ "model.layers.29.self_attn.o_proj",
222
+ "model.layers.30.self_attn.q_a_proj",
223
+ "model.layers.30.self_attn.q_b_proj",
224
+ "model.layers.30.self_attn.kv_a_proj_with_mqa",
225
+ "model.layers.30.self_attn.kv_b_proj",
226
+ "model.layers.30.self_attn.o_proj",
227
+ "model.layers.31.self_attn.q_a_proj",
228
+ "model.layers.31.self_attn.q_b_proj",
229
+ "model.layers.31.self_attn.kv_a_proj_with_mqa",
230
+ "model.layers.31.self_attn.kv_b_proj",
231
+ "model.layers.31.self_attn.o_proj",
232
+ "model.layers.32.self_attn.q_a_proj",
233
+ "model.layers.32.self_attn.q_b_proj",
234
+ "model.layers.32.self_attn.kv_a_proj_with_mqa",
235
+ "model.layers.32.self_attn.kv_b_proj",
236
+ "model.layers.32.self_attn.o_proj",
237
+ "model.layers.33.self_attn.q_a_proj",
238
+ "model.layers.33.self_attn.q_b_proj",
239
+ "model.layers.33.self_attn.kv_a_proj_with_mqa",
240
+ "model.layers.33.self_attn.kv_b_proj",
241
+ "model.layers.33.self_attn.o_proj",
242
+ "model.layers.34.self_attn.q_a_proj",
243
+ "model.layers.34.self_attn.q_b_proj",
244
+ "model.layers.34.self_attn.kv_a_proj_with_mqa",
245
+ "model.layers.34.self_attn.kv_b_proj",
246
+ "model.layers.34.self_attn.o_proj",
247
+ "model.layers.35.self_attn.q_a_proj",
248
+ "model.layers.35.self_attn.q_b_proj",
249
+ "model.layers.35.self_attn.kv_a_proj_with_mqa",
250
+ "model.layers.35.self_attn.kv_b_proj",
251
+ "model.layers.35.self_attn.o_proj",
252
+ "model.layers.36.self_attn.q_a_proj",
253
+ "model.layers.36.self_attn.q_b_proj",
254
+ "model.layers.36.self_attn.kv_a_proj_with_mqa",
255
+ "model.layers.36.self_attn.kv_b_proj",
256
+ "model.layers.36.self_attn.o_proj",
257
+ "model.layers.37.self_attn.q_a_proj",
258
+ "model.layers.37.self_attn.q_b_proj",
259
+ "model.layers.37.self_attn.kv_a_proj_with_mqa",
260
+ "model.layers.37.self_attn.kv_b_proj",
261
+ "model.layers.37.self_attn.o_proj",
262
+ "model.layers.38.self_attn.q_a_proj",
263
+ "model.layers.38.self_attn.q_b_proj",
264
+ "model.layers.38.self_attn.kv_a_proj_with_mqa",
265
+ "model.layers.38.self_attn.kv_b_proj",
266
+ "model.layers.38.self_attn.o_proj",
267
+ "model.layers.39.self_attn.q_a_proj",
268
+ "model.layers.39.self_attn.q_b_proj",
269
+ "model.layers.39.self_attn.kv_a_proj_with_mqa",
270
+ "model.layers.39.self_attn.kv_b_proj",
271
+ "model.layers.39.self_attn.o_proj",
272
+ "model.layers.40.self_attn.q_a_proj",
273
+ "model.layers.40.self_attn.q_b_proj",
274
+ "model.layers.40.self_attn.kv_a_proj_with_mqa",
275
+ "model.layers.40.self_attn.kv_b_proj",
276
+ "model.layers.40.self_attn.o_proj",
277
+ "model.layers.41.self_attn.q_a_proj",
278
+ "model.layers.41.self_attn.q_b_proj",
279
+ "model.layers.41.self_attn.kv_a_proj_with_mqa",
280
+ "model.layers.41.self_attn.kv_b_proj",
281
+ "model.layers.41.self_attn.o_proj",
282
+ "model.layers.42.self_attn.q_a_proj",
283
+ "model.layers.42.self_attn.q_b_proj",
284
+ "model.layers.42.self_attn.kv_a_proj_with_mqa",
285
+ "model.layers.42.self_attn.kv_b_proj",
286
+ "model.layers.42.self_attn.o_proj",
287
+ "model.layers.43.self_attn.q_a_proj",
288
+ "model.layers.43.self_attn.q_b_proj",
289
+ "model.layers.43.self_attn.kv_a_proj_with_mqa",
290
+ "model.layers.43.self_attn.kv_b_proj",
291
+ "model.layers.43.self_attn.o_proj",
292
+ "model.layers.44.self_attn.q_a_proj",
293
+ "model.layers.44.self_attn.q_b_proj",
294
+ "model.layers.44.self_attn.kv_a_proj_with_mqa",
295
+ "model.layers.44.self_attn.kv_b_proj",
296
+ "model.layers.44.self_attn.o_proj",
297
+ "model.layers.45.self_attn.q_a_proj",
298
+ "model.layers.45.self_attn.q_b_proj",
299
+ "model.layers.45.self_attn.kv_a_proj_with_mqa",
300
+ "model.layers.45.self_attn.kv_b_proj",
301
+ "model.layers.45.self_attn.o_proj",
302
+ "model.layers.46.self_attn.q_a_proj",
303
+ "model.layers.46.self_attn.q_b_proj",
304
+ "model.layers.46.self_attn.kv_a_proj_with_mqa",
305
+ "model.layers.46.self_attn.kv_b_proj",
306
+ "model.layers.46.self_attn.o_proj",
307
+ "model.layers.47.self_attn.q_a_proj",
308
+ "model.layers.47.self_attn.q_b_proj",
309
+ "model.layers.47.self_attn.kv_a_proj_with_mqa",
310
+ "model.layers.47.self_attn.kv_b_proj",
311
+ "model.layers.47.self_attn.o_proj",
312
+ "model.layers.48.self_attn.q_a_proj",
313
+ "model.layers.48.self_attn.q_b_proj",
314
+ "model.layers.48.self_attn.kv_a_proj_with_mqa",
315
+ "model.layers.48.self_attn.kv_b_proj",
316
+ "model.layers.48.self_attn.o_proj",
317
+ "model.layers.49.self_attn.q_a_proj",
318
+ "model.layers.49.self_attn.q_b_proj",
319
+ "model.layers.49.self_attn.kv_a_proj_with_mqa",
320
+ "model.layers.49.self_attn.kv_b_proj",
321
+ "model.layers.49.self_attn.o_proj",
322
+ "model.layers.50.self_attn.q_a_proj",
323
+ "model.layers.50.self_attn.q_b_proj",
324
+ "model.layers.50.self_attn.kv_a_proj_with_mqa",
325
+ "model.layers.50.self_attn.kv_b_proj",
326
+ "model.layers.50.self_attn.o_proj",
327
+ "model.layers.51.self_attn.q_a_proj",
328
+ "model.layers.51.self_attn.q_b_proj",
329
+ "model.layers.51.self_attn.kv_a_proj_with_mqa",
330
+ "model.layers.51.self_attn.kv_b_proj",
331
+ "model.layers.51.self_attn.o_proj",
332
+ "model.layers.52.self_attn.q_a_proj",
333
+ "model.layers.52.self_attn.q_b_proj",
334
+ "model.layers.52.self_attn.kv_a_proj_with_mqa",
335
+ "model.layers.52.self_attn.kv_b_proj",
336
+ "model.layers.52.self_attn.o_proj",
337
+ "model.layers.53.self_attn.q_a_proj",
338
+ "model.layers.53.self_attn.q_b_proj",
339
+ "model.layers.53.self_attn.kv_a_proj_with_mqa",
340
+ "model.layers.53.self_attn.kv_b_proj",
341
+ "model.layers.53.self_attn.o_proj",
342
+ "model.layers.54.self_attn.q_a_proj",
343
+ "model.layers.54.self_attn.q_b_proj",
344
+ "model.layers.54.self_attn.kv_a_proj_with_mqa",
345
+ "model.layers.54.self_attn.kv_b_proj",
346
+ "model.layers.54.self_attn.o_proj",
347
+ "model.layers.55.self_attn.q_a_proj",
348
+ "model.layers.55.self_attn.q_b_proj",
349
+ "model.layers.55.self_attn.kv_a_proj_with_mqa",
350
+ "model.layers.55.self_attn.kv_b_proj",
351
+ "model.layers.55.self_attn.o_proj",
352
+ "model.layers.56.self_attn.q_a_proj",
353
+ "model.layers.56.self_attn.q_b_proj",
354
+ "model.layers.56.self_attn.kv_a_proj_with_mqa",
355
+ "model.layers.56.self_attn.kv_b_proj",
356
+ "model.layers.56.self_attn.o_proj",
357
+ "model.layers.57.self_attn.q_a_proj",
358
+ "model.layers.57.self_attn.q_b_proj",
359
+ "model.layers.57.self_attn.kv_a_proj_with_mqa",
360
+ "model.layers.57.self_attn.kv_b_proj",
361
+ "model.layers.57.self_attn.o_proj",
362
+ "model.layers.58.self_attn.q_a_proj",
363
+ "model.layers.58.self_attn.q_b_proj",
364
+ "model.layers.58.self_attn.kv_a_proj_with_mqa",
365
+ "model.layers.58.self_attn.kv_b_proj",
366
+ "model.layers.58.self_attn.o_proj",
367
+ "model.layers.59.self_attn.q_a_proj",
368
+ "model.layers.59.self_attn.q_b_proj",
369
+ "model.layers.59.self_attn.kv_a_proj_with_mqa",
370
+ "model.layers.59.self_attn.kv_b_proj",
371
+ "model.layers.59.self_attn.o_proj",
372
+ "model.layers.60.self_attn.q_a_proj",
373
+ "model.layers.60.self_attn.q_b_proj",
374
+ "model.layers.60.self_attn.kv_a_proj_with_mqa",
375
+ "model.layers.60.self_attn.kv_b_proj",
376
+ "model.layers.60.self_attn.o_proj",
377
+ "lm_head"
378
+ ],
379
+ "export": {
380
+ "kv_cache_group": [],
381
+ "min_kv_scale": 0.0,
382
+ "pack_method": "reorder",
383
+ "weight_format": "real_quantized",
384
+ "weight_merge_groups": null
385
+ },
386
+ "global_quant_config": {
387
+ "bias": null,
388
+ "input_tensors": {
389
+ "ch_axis": -1,
390
+ "dtype": "fp4",
391
+ "group_size": 32,
392
+ "is_dynamic": true,
393
+ "is_scale_quant": false,
394
+ "mx_element_dtype": null,
395
+ "observer_cls": "PerBlockMXObserver",
396
+ "qscheme": "per_group",
397
+ "round_method": "half_even",
398
+ "scale_calculation_mode": "even",
399
+ "scale_format": "e8m0",
400
+ "scale_type": "float",
401
+ "symmetric": null
402
+ },
403
+ "output_tensors": null,
404
+ "target_device": null,
405
+ "weight": {
406
+ "ch_axis": -1,
407
+ "dtype": "fp4",
408
+ "group_size": 32,
409
+ "is_dynamic": false,
410
+ "is_scale_quant": false,
411
+ "mx_element_dtype": null,
412
+ "observer_cls": "PerBlockMXObserver",
413
+ "qscheme": "per_group",
414
+ "round_method": "half_even",
415
+ "scale_calculation_mode": "even",
416
+ "scale_format": "e8m0",
417
+ "scale_type": "float",
418
+ "symmetric": null
419
+ }
420
+ },
421
+ "kv_cache_quant_config": {},
422
+ "layer_quant_config": {},
423
+ "layer_type_quant_config": {},
424
+ "quant_method": "quark",
425
+ "quant_mode": "eager_mode",
426
+ "softmax_quant_spec": null,
427
+ "version": "0.10+ab0c435"
428
+ },
429
+ "rms_norm_eps": 1e-06,
430
+ "rope_scaling": {
431
+ "beta_fast": 32,
432
+ "beta_slow": 1,
433
+ "factor": 40,
434
+ "mscale": 1.0,
435
+ "mscale_all_dim": 1.0,
436
+ "original_max_position_embeddings": 4096,
437
+ "type": "yarn"
438
+ },
439
+ "rope_theta": 10000,
440
+ "routed_scaling_factor": 2.5,
441
+ "scoring_func": "sigmoid",
442
+ "tie_word_embeddings": false,
443
+ "topk_group": 4,
444
+ "topk_method": "noaux_tc",
445
+ "torch_dtype": "bfloat16",
446
+ "transformers_version": "4.55.0",
447
+ "unsloth_fixed": true,
448
+ "use_cache": true,
449
+ "v_head_dim": 128,
450
+ "vocab_size": 129280
451
+ }
configuration_deepseek.py ADDED
@@ -0,0 +1,199 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ from transformers.configuration_utils import PretrainedConfig
2
+ from transformers.utils import logging
3
+
4
+ logger = logging.get_logger(__name__)
5
+
6
+ DEEPSEEK_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
7
+ class DeepseekV3Config(PretrainedConfig):
8
+ r"""
9
+ This is the configuration class to store the configuration of a [`DeepseekV3Model`]. It is used to instantiate an DeepSeek
10
+ model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
11
+ defaults will yield a similar configuration to that of the DeepSeek-V3.
12
+
13
+ Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
14
+ documentation from [`PretrainedConfig`] for more information.
15
+
16
+
17
+ Args:
18
+ vocab_size (`int`, *optional*, defaults to 129280):
19
+ Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
20
+ `inputs_ids` passed when calling [`DeepseekV3Model`]
21
+ hidden_size (`int`, *optional*, defaults to 4096):
22
+ Dimension of the hidden representations.
23
+ intermediate_size (`int`, *optional*, defaults to 11008):
24
+ Dimension of the MLP representations.
25
+ moe_intermediate_size (`int`, *optional*, defaults to 1407):
26
+ Dimension of the MoE representations.
27
+ num_hidden_layers (`int`, *optional*, defaults to 32):
28
+ Number of hidden layers in the Transformer decoder.
29
+ num_nextn_predict_layers (`int`, *optional*, defaults to 1):
30
+ Number of nextn predict layers in the DeepSeekV3 Model.
31
+ num_attention_heads (`int`, *optional*, defaults to 32):
32
+ Number of attention heads for each attention layer in the Transformer decoder.
33
+ n_shared_experts (`int`, *optional*, defaults to None):
34
+ Number of shared experts, None means dense model.
35
+ n_routed_experts (`int`, *optional*, defaults to None):
36
+ Number of routed experts, None means dense model.
37
+ routed_scaling_factor (`float`, *optional*, defaults to 1.0):
38
+ Scaling factor or routed experts.
39
+ topk_method (`str`, *optional*, defaults to `gready`):
40
+ Topk method used in routed gate.
41
+ n_group (`int`, *optional*, defaults to None):
42
+ Number of groups for routed experts.
43
+ topk_group (`int`, *optional*, defaults to None):
44
+ Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
45
+ num_experts_per_tok (`int`, *optional*, defaults to None):
46
+ Number of selected experts, None means dense model.
47
+ moe_layer_freq (`int`, *optional*, defaults to 1):
48
+ The frequency of the MoE layer: one expert layer for every `moe_layer_freq - 1` dense layers.
49
+ first_k_dense_replace (`int`, *optional*, defaults to 0):
50
+ Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
51
+ \--k dense layers--/
52
+ norm_topk_prob (`bool`, *optional*, defaults to False):
53
+ Whether to normalize the weights of the routed experts.
54
+ scoring_func (`str`, *optional*, defaults to 'softmax'):
55
+ Method of computing expert weights.
56
+ aux_loss_alpha (`float`, *optional*, defaults to 0.001):
57
+ Auxiliary loss weight coefficient.
58
+ seq_aux = (`bool`, *optional*, defaults to True):
59
+ Whether to compute the auxiliary loss for each individual sample.
60
+ num_key_value_heads (`int`, *optional*):
61
+ This is the number of key_value heads that should be used to implement Grouped Query Attention. If
62
+ `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
63
+ `num_key_value_heads=1 the model will use Multi Query Attention (MQA) otherwise GQA is used. When
64
+ converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
65
+ by meanpooling all the original heads within that group. For more details checkout [this
66
+ paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
67
+ `num_attention_heads`.
68
+ hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
69
+ The non-linear activation function (function or string) in the decoder.
70
+ max_position_embeddings (`int`, *optional*, defaults to 2048):
71
+ The maximum sequence length that this model might ever be used with.
72
+ initializer_range (`float`, *optional*, defaults to 0.02):
73
+ The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
74
+ rms_norm_eps (`float`, *optional*, defaults to 1e-06):
75
+ The epsilon used by the rms normalization layers.
76
+ use_cache (`bool`, *optional*, defaults to `True`):
77
+ Whether or not the model should return the last key/values attentions (not used by all models). Only
78
+ relevant if `config.is_decoder=True`.
79
+ pad_token_id (`int`, *optional*):
80
+ Padding token id.
81
+ bos_token_id (`int`, *optional*, defaults to 1):
82
+ Beginning of stream token id.
83
+ eos_token_id (`int`, *optional*, defaults to 2):
84
+ End of stream token id.
85
+ tie_word_embeddings (`bool`, *optional*, defaults to `False`):
86
+ Whether to tie weight embeddings
87
+ rope_theta (`float`, *optional*, defaults to 10000.0):
88
+ The base period of the RoPE embeddings.
89
+ rope_scaling (`Dict`, *optional*):
90
+ Dictionary containing the scaling configuration for the RoPE embeddings. Currently supports two scaling
91
+ strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
92
+ `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
93
+ `max_position_embeddings` to the expected new maximum.
94
+ attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
95
+ Whether to use a bias in the query, key, value and output projection layers during self-attention.
96
+ attention_dropout (`float`, *optional*, defaults to 0.0):
97
+ The dropout ratio for the attention probabilities.
98
+
99
+ ```python
100
+ >>> from transformers import DeepseekV3Model, DeepseekV3Config
101
+
102
+ >>> # Initializing a Deepseek-V3 style configuration
103
+ >>> configuration = DeepseekV3Config()
104
+
105
+ >>> # Accessing the model configuration
106
+ >>> configuration = model.config
107
+ ```"""
108
+
109
+ model_type = "deepseek_v3"
110
+ keys_to_ignore_at_inference = ["past_key_values"]
111
+
112
+ def __init__(
113
+ self,
114
+ vocab_size=129280,
115
+ hidden_size=7168,
116
+ intermediate_size=18432,
117
+ moe_intermediate_size = 2048,
118
+ num_hidden_layers=61,
119
+ num_nextn_predict_layers=1,
120
+ num_attention_heads=128,
121
+ num_key_value_heads=128,
122
+ n_shared_experts = 1,
123
+ n_routed_experts = 256,
124
+ ep_size = 1,
125
+ routed_scaling_factor = 2.5,
126
+ kv_lora_rank = 512,
127
+ q_lora_rank = 1536,
128
+ qk_rope_head_dim = 64,
129
+ v_head_dim = 128,
130
+ qk_nope_head_dim = 128,
131
+ topk_method = 'noaux_tc',
132
+ n_group = 8,
133
+ topk_group = 4,
134
+ num_experts_per_tok = 8,
135
+ moe_layer_freq = 1,
136
+ first_k_dense_replace = 3,
137
+ norm_topk_prob = True,
138
+ scoring_func = 'sigmoid',
139
+ hidden_act="silu",
140
+ max_position_embeddings=4096,
141
+ initializer_range=0.02,
142
+ rms_norm_eps=1e-6,
143
+ use_cache=True,
144
+ pad_token_id=None,
145
+ bos_token_id=0,
146
+ eos_token_id=1,
147
+ tie_word_embeddings=False,
148
+ rope_theta=10000.0,
149
+ rope_scaling=None,
150
+ attention_bias=False,
151
+ attention_dropout=0.0,
152
+ **kwargs,
153
+ ):
154
+ self.vocab_size = vocab_size
155
+ self.max_position_embeddings = max_position_embeddings
156
+ self.hidden_size = hidden_size
157
+ self.intermediate_size = intermediate_size
158
+ self.moe_intermediate_size = moe_intermediate_size
159
+ self.num_hidden_layers = num_hidden_layers
160
+ self.num_nextn_predict_layers = num_nextn_predict_layers
161
+ self.num_attention_heads = num_attention_heads
162
+ self.n_shared_experts = n_shared_experts
163
+ self.n_routed_experts = n_routed_experts
164
+ self.ep_size = ep_size
165
+ self.routed_scaling_factor = routed_scaling_factor
166
+ self.kv_lora_rank = kv_lora_rank
167
+ self.q_lora_rank = q_lora_rank
168
+ self.qk_rope_head_dim = qk_rope_head_dim
169
+ self.v_head_dim = v_head_dim
170
+ self.qk_nope_head_dim = qk_nope_head_dim
171
+ self.topk_method = topk_method
172
+ self.n_group = n_group
173
+ self.topk_group = topk_group
174
+ self.num_experts_per_tok = num_experts_per_tok
175
+ self.moe_layer_freq = moe_layer_freq
176
+ self.first_k_dense_replace = first_k_dense_replace
177
+ self.norm_topk_prob = norm_topk_prob
178
+ self.scoring_func = scoring_func
179
+ # for backward compatibility
180
+ if num_key_value_heads is None:
181
+ num_key_value_heads = num_attention_heads
182
+
183
+ self.num_key_value_heads = num_key_value_heads
184
+ self.hidden_act = hidden_act
185
+ self.initializer_range = initializer_range
186
+ self.rms_norm_eps = rms_norm_eps
187
+ self.use_cache = use_cache
188
+ self.rope_theta = rope_theta
189
+ self.rope_scaling = rope_scaling
190
+ self.attention_bias = attention_bias
191
+ self.attention_dropout = attention_dropout
192
+
193
+ super().__init__(
194
+ pad_token_id=pad_token_id,
195
+ bos_token_id=bos_token_id,
196
+ eos_token_id=eos_token_id,
197
+ tie_word_embeddings=tie_word_embeddings,
198
+ **kwargs,
199
+ )
generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 0,
4
+ "do_sample": true,
5
+ "eos_token_id": 1,
6
+ "temperature": 0.6,
7
+ "top_p": 0.95,
8
+ "transformers_version": "4.55.0"
9
+ }
model-00001-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1f48d59af2576247521933186749f25932df87ad060f5b7b14883551fe439c1
3
+ size 4995884768
model-00002-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:03710f79f63e2a662a2ad1f82b0cc554d747d60aefda1e3a6a956ac8035d7752
3
+ size 4995052496
model-00003-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5a5a0c6819938a9430a1ecfc6e5b6dbccbcecd400b99bdf02e90cb3b257e9a7f
3
+ size 4999071368
model-00004-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b60f4ec8a77d1b69690ba5095fddc1199d5a8767a4f65a6d63b4032872cceb36
3
+ size 4994943096
model-00005-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3007ebb0e5b22bbee20bf053005a52e58f465352fd8e10541ded129f6cc24a24
3
+ size 4994943232
model-00006-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3d825d2545bda0cf6cce5c05501511c3c6cf1b89ff690cc7467ca7979ced42c
3
+ size 4994943456
model-00007-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6bab43ec8e5d10de20f2e6b7a4860b6a18f62bf6957083ee9a53c2fa786cd10
3
+ size 4999180624
model-00008-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29879b960c2c70e8324a7babb544ee78019159c2c5008ce7ccae77aaa8706f32
3
+ size 4994943096
model-00009-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:656bd6ffed12873c0a0f01fc1601b026745ab5b8ff1d212bb1e87b263a86fcfa
3
+ size 4994943096
model-00010-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2388459ecfe9052469e42a7fb210aa8bde68dd6c46641143f6dccf22bc14b94d
3
+ size 4994943720
model-00011-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:21baaf791ce1c513a273bd3de3d70d3d982338ce9452f87f1f8e1541e3b0d946
3
+ size 4916044152
model-00012-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a91e1e1c4eb641bb08dd753448b8fcc98c59a6506945d75d8705e1c1310e5677
3
+ size 5000091936
model-00013-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fbc87ad36d700393b32f1b98f7d58b3b2d3f88cfc711467a6ca015309442ae9
3
+ size 4994944296
model-00014-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2adfb73f19f45d3027b82a1fee2b9a89a9201492b0cd068910309316624ceb6a
3
+ size 4994944392
model-00015-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:82c740c9be8b5bdb84defb7d87c5e96fbb7aab29f22f02f34e979a2d2c513b8a
3
+ size 4994944648
model-00016-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:757d9a33d0cf70a50544dbd80b469d18ba401c515a22958392d17f1e6314ddaa
3
+ size 4999181952
model-00017-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:aa3aabe771dac558e56c2ab53d0a57aaf8557d58cd4e35b9dfedd15919d638d9
3
+ size 4994944288
model-00018-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8fe1c4ed065132dead39c8644d4bb53890b8612a75d56c4e31c9fc848ec24492
3
+ size 4994944296
model-00019-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d73ac9341e0eeae85c30e58d8e21c6509c0340ecb276b5aaaac2257ff0610735
3
+ size 4994944536
model-00020-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9a5aa66889a0ebc7a1debdcb1feb2f528b3339188afffa1b62651f28758ddbcf
3
+ size 4767862480
model-00021-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb5503253433ae681123db9e184fb3ff264e2bc7a4e4068bd38a32b961e04cb3
3
+ size 5000091936
model-00022-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c44a1fba5a0a7df39c89d8aa8e818f0cea83db62cfa4342ffa844bf68de4c5da
3
+ size 4994944296
model-00023-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52663b7efa7df1f4052152bba32ac9f5c15d9ebd864aea031f0a2697016baa9d
3
+ size 4994944392
model-00024-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e326d957df3c7df69fa02daacb7683a5d8ee1ff2162e5d089859d226b6509857
3
+ size 4994944648
model-00025-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67cf9215375dfc9679e370c89fab42ec1b44f565365a248fbd87876101cde3db
3
+ size 4999181952
model-00026-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48c0d5f199c5d7f08fed06308f6172ffd17267ab819070686e75e83f4c5418da
3
+ size 4994944288
model-00027-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9f088d97b0ce73507bb7ea1b866558cafc413f68990d1f606646c80f0700260
3
+ size 4994944296
model-00028-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ae7bad0879c95d6c4c2887b4e19bf9cde22c640a806014949f1bf7289a38e72a
3
+ size 4994944536
model-00029-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:242e061d2cf4a8d5d8ab3df3c7548ced8625a81e2fc99aa6c7d33dbe11261b5b
3
+ size 4767862480
model-00030-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a727b627ea46661b88f210bc6be6ad2d66cab38732d37d2431bad13764518772
3
+ size 5000091936
model-00031-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df6d504c25c9b2682fe8088b98741a84c7d08c42f971b48c78bb9ad4e3908711
3
+ size 4994944296
model-00032-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8affb0f56c2673ab8d641dd0c01a9365c7091e3b74a1dd5c1b8efd7156ffcd5b
3
+ size 4994944392
model-00033-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2768423a74869ed22cb2605e3ce43f982b0219497a910cc3d27d212537f2acaf
3
+ size 4994944648
model-00034-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e328a402ca90b20f332556cdcd09721b16b5f756454209988a0ba2167bf2892e
3
+ size 4999181952
model-00035-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9621ebb13e9ed083d164e8ba4788eb0b5cbc9842df2ab32b93ef3fbbe95803b9
3
+ size 4994944288
model-00036-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:807839b40a4771aa8f765acec99cd549bfb477ff3a55fd387b7c5aaea85f4c84
3
+ size 4994944296
model-00037-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b995f5749abeda620873fa463538fb34a17f9a6b9936349478eefd7e1fa8e664
3
+ size 4994944536
model-00038-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:374c85b37befa1d7135c4e4b3aa38760d36f0583db33ae7c43b501c1aab561f7
3
+ size 4767862480
model-00039-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:53dde826b40589b9886287d89c2cdd1b3aeb735615fcc774b12298887d506fdc
3
+ size 5000091936
model-00040-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7277a1c12733e9365b7f13e5ae35e794719501f1d4d6dbf7d10e25eb0bff2b7f
3
+ size 4994944296
model-00041-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:538ac0ade4a2d9eb06906ffca1de161dfa16420c871d30c9f06fc8eceda42e53
3
+ size 4994944392
model-00042-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cdd8663ebffe045e98601a33e1594c93ae0f81dd353dc78cfcb62d98c34650c7
3
+ size 4994944648
model-00043-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:db27547c96e147605e4e609d3b11bd170750ef9b7cabc7c60989cfbd58a5ad62
3
+ size 4999181952
model-00044-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f20f9583daeda20fc0b347a6d37c76d0d52325d5f2538861335c3fe13f4d7c3f
3
+ size 4994944288
model-00045-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8920b892e78aba5b5c864b2dcfd8a6c699be0299566dd774095f3cc380d437ea
3
+ size 4994944296
model-00046-of-00076.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5219ea65edc79ca58a2869ca3c2636b90a0f923f759e0ef277f41ed8122225be
3
+ size 4994944536