meenchen commited on
Commit
d12ff8d
·
0 Parent(s):

Initial Commit

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +35 -0
  2. LICENSE +9 -0
  3. README.md +189 -0
  4. config.json +67 -0
  5. configuration_deepseek.py +199 -0
  6. generation_config.json +9 -0
  7. hf_quant_config.json +259 -0
  8. model-00001-of-000163.safetensors +3 -0
  9. model-00002-of-000163.safetensors +3 -0
  10. model-00003-of-000163.safetensors +3 -0
  11. model-00004-of-000163.safetensors +3 -0
  12. model-00005-of-000163.safetensors +3 -0
  13. model-00006-of-000163.safetensors +3 -0
  14. model-00007-of-000163.safetensors +3 -0
  15. model-00008-of-000163.safetensors +3 -0
  16. model-00009-of-000163.safetensors +3 -0
  17. model-00010-of-000163.safetensors +3 -0
  18. model-00011-of-000163.safetensors +3 -0
  19. model-00012-of-000163.safetensors +3 -0
  20. model-00013-of-000163.safetensors +3 -0
  21. model-00014-of-000163.safetensors +3 -0
  22. model-00015-of-000163.safetensors +3 -0
  23. model-00016-of-000163.safetensors +3 -0
  24. model-00017-of-000163.safetensors +3 -0
  25. model-00018-of-000163.safetensors +3 -0
  26. model-00019-of-000163.safetensors +3 -0
  27. model-00020-of-000163.safetensors +3 -0
  28. model-00021-of-000163.safetensors +3 -0
  29. model-00022-of-000163.safetensors +3 -0
  30. model-00023-of-000163.safetensors +3 -0
  31. model-00024-of-000163.safetensors +3 -0
  32. model-00025-of-000163.safetensors +3 -0
  33. model-00026-of-000163.safetensors +3 -0
  34. model-00027-of-000163.safetensors +3 -0
  35. model-00028-of-000163.safetensors +3 -0
  36. model-00029-of-000163.safetensors +3 -0
  37. model-00030-of-000163.safetensors +3 -0
  38. model-00031-of-000163.safetensors +3 -0
  39. model-00032-of-000163.safetensors +3 -0
  40. model-00033-of-000163.safetensors +3 -0
  41. model-00034-of-000163.safetensors +3 -0
  42. model-00035-of-000163.safetensors +3 -0
  43. model-00036-of-000163.safetensors +3 -0
  44. model-00037-of-000163.safetensors +3 -0
  45. model-00038-of-000163.safetensors +3 -0
  46. model-00039-of-000163.safetensors +3 -0
  47. model-00040-of-000163.safetensors +3 -0
  48. model-00041-of-000163.safetensors +3 -0
  49. model-00042-of-000163.safetensors +3 -0
  50. model-00043-of-000163.safetensors +3 -0
.gitattributes ADDED
@@ -0,0 +1,35 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ *.7z filter=lfs diff=lfs merge=lfs -text
2
+ *.arrow filter=lfs diff=lfs merge=lfs -text
3
+ *.bin filter=lfs diff=lfs merge=lfs -text
4
+ *.bz2 filter=lfs diff=lfs merge=lfs -text
5
+ *.ckpt filter=lfs diff=lfs merge=lfs -text
6
+ *.ftz filter=lfs diff=lfs merge=lfs -text
7
+ *.gz filter=lfs diff=lfs merge=lfs -text
8
+ *.h5 filter=lfs diff=lfs merge=lfs -text
9
+ *.joblib filter=lfs diff=lfs merge=lfs -text
10
+ *.lfs.* filter=lfs diff=lfs merge=lfs -text
11
+ *.mlmodel filter=lfs diff=lfs merge=lfs -text
12
+ *.model filter=lfs diff=lfs merge=lfs -text
13
+ *.msgpack filter=lfs diff=lfs merge=lfs -text
14
+ *.npy filter=lfs diff=lfs merge=lfs -text
15
+ *.npz filter=lfs diff=lfs merge=lfs -text
16
+ *.onnx filter=lfs diff=lfs merge=lfs -text
17
+ *.ot filter=lfs diff=lfs merge=lfs -text
18
+ *.parquet filter=lfs diff=lfs merge=lfs -text
19
+ *.pb filter=lfs diff=lfs merge=lfs -text
20
+ *.pickle filter=lfs diff=lfs merge=lfs -text
21
+ *.pkl filter=lfs diff=lfs merge=lfs -text
22
+ *.pt filter=lfs diff=lfs merge=lfs -text
23
+ *.pth filter=lfs diff=lfs merge=lfs -text
24
+ *.rar filter=lfs diff=lfs merge=lfs -text
25
+ *.safetensors filter=lfs diff=lfs merge=lfs -text
26
+ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
27
+ *.tar.* filter=lfs diff=lfs merge=lfs -text
28
+ *.tar filter=lfs diff=lfs merge=lfs -text
29
+ *.tflite filter=lfs diff=lfs merge=lfs -text
30
+ *.tgz filter=lfs diff=lfs merge=lfs -text
31
+ *.wasm filter=lfs diff=lfs merge=lfs -text
32
+ *.xz 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
LICENSE ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ MIT License
2
+ Copyright (c) 2025 Nvidia
3
+ Copyright (c) 2023 DeepSeek
4
+
5
+ Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
6
+
7
+ The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
8
+
9
+ THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
README.md ADDED
@@ -0,0 +1,189 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ pipeline_tag: text-generation
3
+ base_model:
4
+ - deepseek-ai/DeepSeek-R1-0528
5
+ license: mit
6
+ library_name: Model Optimizer
7
+ tags:
8
+ - nvidia
9
+ - ModelOpt
10
+ - DeepSeekR1
11
+ - quantized
12
+ - FP4
13
+ ---
14
+ # Model Overview
15
+
16
+ ## Description:
17
+ The NVIDIA DeepSeek-R1-0528-FP4 v2 model is the quantized version of the DeepSeek AI's DeepSeek R1 0528 model, which is an auto-regressive language model that uses an optimized transformer architecture. For more information, please check [here](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528). The NVIDIA DeepSeek R1 FP4 model is quantized with [TensorRT Model Optimizer](https://github.com/NVIDIA/TensorRT-Model-Optimizer).
18
+
19
+ Compared to [nvidia/DeepSeek-R1-0528-FP4](https://huggingface.co/nvidia/DeepSeek-R1-0528-FP4), this checkpoint additionally quantizes the wo module in attention layers.
20
+
21
+ This model is ready for commercial/non-commercial use. <br>
22
+
23
+ ## Third-Party Community Consideration
24
+ This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA [(DeepSeek R1) Model Card](https://huggingface.co/deepseek-ai/DeepSeek-R1-0528).
25
+
26
+ ### License/Terms of Use:
27
+ [MIT](https://huggingface.co/datasets/choosealicense/licenses/blob/main/markdown/mit.md)
28
+
29
+
30
+ ## Model Architecture:
31
+ **Architecture Type:** Transformers <br>
32
+ **Network Architecture:** DeepSeek R1 <br>
33
+
34
+ ## Input:
35
+ **Input Type(s):** Text <br>
36
+ **Input Format(s):** String <br>
37
+ **Input Parameters:** 1D (One Dimensional): Sequences <br>
38
+ **Other Properties Related to Input:** DeepSeek recommends adhering to the following configurations when utilizing the DeepSeek-R1 series models, including benchmarking, to achieve the expected performance: \
39
+
40
+ - Set the temperature within the range of 0.5-0.7 (0.6 is recommended) to prevent endless repetitions or incoherent outputs.
41
+ - Avoid adding a system prompt; all instructions should be contained within the user prompt.
42
+ - For mathematical problems, it is advisable to include a directive in your prompt such as: "Please reason step by step, and put your final answer within \boxed{}."
43
+ - When evaluating model performance, it is recommended to conduct multiple tests and average the results. <br>
44
+
45
+ ## Output:
46
+ **Output Type(s):** Text <br>
47
+ **Output Format:** String <br>
48
+ **Output Parameters:** 1D (One Dimensional): Sequences <br>
49
+
50
+ ## Software Integration:
51
+ **Supported Runtime Engine(s):** <br>
52
+ * TensorRT-LLM <br>
53
+
54
+ **Supported Hardware Microarchitecture Compatibility:** <br>
55
+ * NVIDIA Blackwell <br>
56
+
57
+ **Preferred Operating System(s):** <br>
58
+ * Linux <br>
59
+
60
+ ## Model Version(s):
61
+ ** The model is quantized with nvidia-modelopt **v0.33.0** <br>
62
+
63
+ ## Training Dataset: <br>
64
+ ** Data Collection Method by dataset: Hybrid: Human, Automated <br>
65
+ ** Labeling Method by dataset: Hybrid: Human, Automated <br>
66
+
67
+ ## Testing Dataset: <br>
68
+ ** Data Collection Method by dataset: Hybrid: Human, Automated <br>
69
+ ** Labeling Method by dataset: Hybrid: Human, Automated <br>
70
+
71
+ ## Evaluation Dataset: <br>
72
+ ** Data Collection Method by dataset: Hybrid: Human, Automated <br>
73
+ ** Labeling Method by dataset: Hybrid: Human, Automated <br>
74
+
75
+ ## Calibration Datasets:
76
+ * Calibration Dataset: [cnn_dailymail](https://huggingface.co/datasets/abisee/cnn_dailymail) <br>
77
+ ** Data collection method: Automated. <br>
78
+ ** Labeling method: Automated. <br>
79
+
80
+
81
+ ## Inference:
82
+ **Engine:** TensorRT-LLM <br>
83
+ **Test Hardware:** B200 <br>
84
+
85
+ ## Post Training Quantization
86
+ This model was obtained by quantizing the weights and activations of DeepSeek R1-0528 to FP4 data type, ready for inference with TensorRT-LLM. Only the weights and activations of the linear operators within transformer blocks are quantized. This optimization reduces the number of bits per parameter from 8 to 4, reducing the disk size and GPU memory requirements by approximately 1.6x.
87
+
88
+ ## Usage
89
+
90
+ ### Deploy with TensorRT-LLM
91
+
92
+ To deploy the quantized FP4 checkpoint with [TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM) LLM API, follow the sample codes below (you need 8xB200 GPU and TensorRT LLM built from source with the latest main branch):
93
+
94
+ * LLM API sample usage:
95
+ ```
96
+ from tensorrt_llm import SamplingParams
97
+ from tensorrt_llm._torch import LLM
98
+
99
+ def main():
100
+
101
+ prompts = [
102
+ "Hello, my name is",
103
+ "The president of the United States is",
104
+ "The capital of France is",
105
+ "The future of AI is",
106
+ ]
107
+ sampling_params = SamplingParams(max_tokens=32)
108
+
109
+ llm = LLM(model="nvidia/DeepSeek-R1-0528-FP4-v2", tensor_parallel_size=8, enable_attention_dp=True)
110
+
111
+ outputs = llm.generate(prompts, sampling_params)
112
+
113
+ # Print the outputs.
114
+ for output in outputs:
115
+ prompt = output.prompt
116
+ generated_text = output.outputs[0].text
117
+ print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
118
+
119
+
120
+ # The entry point of the program needs to be protected for spawning processes.
121
+ if __name__ == '__main__':
122
+ main()
123
+
124
+ ```
125
+
126
+ ### Evaluation
127
+ The accuracy benchmark results are presented in the table below:
128
+ <table>
129
+ <tr>
130
+ <td><strong>Precision</strong>
131
+ </td>
132
+ <td><strong>MMLU-Pro</strong>
133
+ </td>
134
+ <td><strong>GPQA Diamond</strong>
135
+ </td>
136
+ <td><strong>LiveCodeBench</strong>
137
+ </td>
138
+ <td><strong>SCICODE</strong>
139
+ </td>
140
+ <td><strong>MATH-500</strong>
141
+ </td>
142
+ <td><strong>AIME 2024</strong>
143
+ </td>
144
+ </tr>
145
+ <tr>
146
+ <td>FP8 (AA Ref)
147
+ </td>
148
+ <td>85
149
+ </td>
150
+ <td>81
151
+ </td>
152
+ <td>77
153
+ </td>
154
+ <td>40
155
+ </td>
156
+ <td>98
157
+ </td>
158
+ <td>89
159
+ </td>
160
+ </tr>
161
+ <tr>
162
+ <td>FP4
163
+ </td>
164
+ <td>84
165
+ </td>
166
+ <td>80
167
+ </td>
168
+ <td>77
169
+ </td>
170
+ <td>44
171
+ </td>
172
+ <td>98
173
+ </td>
174
+ <td>88
175
+ </td>
176
+ </tr>
177
+ <tr>
178
+ </table>
179
+
180
+ *Max OSL for LiveCodeBench eval can be as high as 128K.
181
+
182
+ ## Model Limitations:
183
+ The base model was trained on data that contains toxic language and societal biases originally crawled from the internet. Therefore, the model may amplify those biases and return toxic responses especially when prompted with toxic prompts. The model may generate answers that may be inaccurate, omit key information, or include irrelevant or redundant text producing socially unacceptable or undesirable text, even if the prompt itself does not include anything explicitly offensive.
184
+
185
+ ## Ethical Considerations
186
+
187
+ NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.
188
+
189
+ Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/).
config.json ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "q_lora_rank": 1536,
35
+ "qk_nope_head_dim": 128,
36
+ "qk_rope_head_dim": 64,
37
+ "quantization_config": {
38
+ "activation_scheme": "dynamic",
39
+ "fmt": "e4m3",
40
+ "quant_method": "fp8",
41
+ "weight_block_size": [
42
+ 128,
43
+ 128
44
+ ]
45
+ },
46
+ "rms_norm_eps": 1e-06,
47
+ "rope_scaling": {
48
+ "beta_fast": 32,
49
+ "beta_slow": 1,
50
+ "factor": 40,
51
+ "mscale": 1.0,
52
+ "mscale_all_dim": 1.0,
53
+ "original_max_position_embeddings": 4096,
54
+ "type": "yarn"
55
+ },
56
+ "rope_theta": 10000,
57
+ "routed_scaling_factor": 2.5,
58
+ "scoring_func": "sigmoid",
59
+ "tie_word_embeddings": false,
60
+ "topk_group": 4,
61
+ "topk_method": "noaux_tc",
62
+ "torch_dtype": "bfloat16",
63
+ "transformers_version": "4.46.3",
64
+ "use_cache": true,
65
+ "v_head_dim": 128,
66
+ "vocab_size": 129280
67
+ }
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
+ "eos_token_id": 1,
5
+ "do_sample": true,
6
+ "temperature": 0.6,
7
+ "top_p": 0.95,
8
+ "transformers_version": "4.46.3"
9
+ }
hf_quant_config.json ADDED
@@ -0,0 +1,259 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "producer": {
3
+ "name": "modelopt",
4
+ "version": "0.34.1.dev3+ge1e885c21"
5
+ },
6
+ "quantization": {
7
+ "quant_algo": "NVFP4",
8
+ "kv_cache_quant_algo": "FP8",
9
+ "group_size": 16,
10
+ "exclude_modules": [
11
+ "lm_head",
12
+ "model.layers.0.self_attn.kv_a_proj_with_mqa",
13
+ "model.layers.0.self_attn.kv_b_proj",
14
+ "model.layers.0.self_attn.q_a_proj",
15
+ "model.layers.0.self_attn.q_b_proj",
16
+ "model.layers.1.self_attn.kv_a_proj_with_mqa",
17
+ "model.layers.1.self_attn.kv_b_proj",
18
+ "model.layers.1.self_attn.q_a_proj",
19
+ "model.layers.1.self_attn.q_b_proj",
20
+ "model.layers.10.self_attn.kv_a_proj_with_mqa",
21
+ "model.layers.10.self_attn.kv_b_proj",
22
+ "model.layers.10.self_attn.q_a_proj",
23
+ "model.layers.10.self_attn.q_b_proj",
24
+ "model.layers.11.self_attn.kv_a_proj_with_mqa",
25
+ "model.layers.11.self_attn.kv_b_proj",
26
+ "model.layers.11.self_attn.q_a_proj",
27
+ "model.layers.11.self_attn.q_b_proj",
28
+ "model.layers.12.self_attn.kv_a_proj_with_mqa",
29
+ "model.layers.12.self_attn.kv_b_proj",
30
+ "model.layers.12.self_attn.q_a_proj",
31
+ "model.layers.12.self_attn.q_b_proj",
32
+ "model.layers.13.self_attn.kv_a_proj_with_mqa",
33
+ "model.layers.13.self_attn.kv_b_proj",
34
+ "model.layers.13.self_attn.q_a_proj",
35
+ "model.layers.13.self_attn.q_b_proj",
36
+ "model.layers.14.self_attn.kv_a_proj_with_mqa",
37
+ "model.layers.14.self_attn.kv_b_proj",
38
+ "model.layers.14.self_attn.q_a_proj",
39
+ "model.layers.14.self_attn.q_b_proj",
40
+ "model.layers.15.self_attn.kv_a_proj_with_mqa",
41
+ "model.layers.15.self_attn.kv_b_proj",
42
+ "model.layers.15.self_attn.q_a_proj",
43
+ "model.layers.15.self_attn.q_b_proj",
44
+ "model.layers.16.self_attn.kv_a_proj_with_mqa",
45
+ "model.layers.16.self_attn.kv_b_proj",
46
+ "model.layers.16.self_attn.q_a_proj",
47
+ "model.layers.16.self_attn.q_b_proj",
48
+ "model.layers.17.self_attn.kv_a_proj_with_mqa",
49
+ "model.layers.17.self_attn.kv_b_proj",
50
+ "model.layers.17.self_attn.q_a_proj",
51
+ "model.layers.17.self_attn.q_b_proj",
52
+ "model.layers.18.self_attn.kv_a_proj_with_mqa",
53
+ "model.layers.18.self_attn.kv_b_proj",
54
+ "model.layers.18.self_attn.q_a_proj",
55
+ "model.layers.18.self_attn.q_b_proj",
56
+ "model.layers.19.self_attn.kv_a_proj_with_mqa",
57
+ "model.layers.19.self_attn.kv_b_proj",
58
+ "model.layers.19.self_attn.q_a_proj",
59
+ "model.layers.19.self_attn.q_b_proj",
60
+ "model.layers.2.self_attn.kv_a_proj_with_mqa",
61
+ "model.layers.2.self_attn.kv_b_proj",
62
+ "model.layers.2.self_attn.q_a_proj",
63
+ "model.layers.2.self_attn.q_b_proj",
64
+ "model.layers.20.self_attn.kv_a_proj_with_mqa",
65
+ "model.layers.20.self_attn.kv_b_proj",
66
+ "model.layers.20.self_attn.q_a_proj",
67
+ "model.layers.20.self_attn.q_b_proj",
68
+ "model.layers.21.self_attn.kv_a_proj_with_mqa",
69
+ "model.layers.21.self_attn.kv_b_proj",
70
+ "model.layers.21.self_attn.q_a_proj",
71
+ "model.layers.21.self_attn.q_b_proj",
72
+ "model.layers.22.self_attn.kv_a_proj_with_mqa",
73
+ "model.layers.22.self_attn.kv_b_proj",
74
+ "model.layers.22.self_attn.q_a_proj",
75
+ "model.layers.22.self_attn.q_b_proj",
76
+ "model.layers.23.self_attn.kv_a_proj_with_mqa",
77
+ "model.layers.23.self_attn.kv_b_proj",
78
+ "model.layers.23.self_attn.q_a_proj",
79
+ "model.layers.23.self_attn.q_b_proj",
80
+ "model.layers.24.self_attn.kv_a_proj_with_mqa",
81
+ "model.layers.24.self_attn.kv_b_proj",
82
+ "model.layers.24.self_attn.q_a_proj",
83
+ "model.layers.24.self_attn.q_b_proj",
84
+ "model.layers.25.self_attn.kv_a_proj_with_mqa",
85
+ "model.layers.25.self_attn.kv_b_proj",
86
+ "model.layers.25.self_attn.q_a_proj",
87
+ "model.layers.25.self_attn.q_b_proj",
88
+ "model.layers.26.self_attn.kv_a_proj_with_mqa",
89
+ "model.layers.26.self_attn.kv_b_proj",
90
+ "model.layers.26.self_attn.q_a_proj",
91
+ "model.layers.26.self_attn.q_b_proj",
92
+ "model.layers.27.self_attn.kv_a_proj_with_mqa",
93
+ "model.layers.27.self_attn.kv_b_proj",
94
+ "model.layers.27.self_attn.q_a_proj",
95
+ "model.layers.27.self_attn.q_b_proj",
96
+ "model.layers.28.self_attn.kv_a_proj_with_mqa",
97
+ "model.layers.28.self_attn.kv_b_proj",
98
+ "model.layers.28.self_attn.q_a_proj",
99
+ "model.layers.28.self_attn.q_b_proj",
100
+ "model.layers.29.self_attn.kv_a_proj_with_mqa",
101
+ "model.layers.29.self_attn.kv_b_proj",
102
+ "model.layers.29.self_attn.q_a_proj",
103
+ "model.layers.29.self_attn.q_b_proj",
104
+ "model.layers.3.self_attn.kv_a_proj_with_mqa",
105
+ "model.layers.3.self_attn.kv_b_proj",
106
+ "model.layers.3.self_attn.q_a_proj",
107
+ "model.layers.3.self_attn.q_b_proj",
108
+ "model.layers.30.self_attn.kv_a_proj_with_mqa",
109
+ "model.layers.30.self_attn.kv_b_proj",
110
+ "model.layers.30.self_attn.q_a_proj",
111
+ "model.layers.30.self_attn.q_b_proj",
112
+ "model.layers.31.self_attn.kv_a_proj_with_mqa",
113
+ "model.layers.31.self_attn.kv_b_proj",
114
+ "model.layers.31.self_attn.q_a_proj",
115
+ "model.layers.31.self_attn.q_b_proj",
116
+ "model.layers.32.self_attn.kv_a_proj_with_mqa",
117
+ "model.layers.32.self_attn.kv_b_proj",
118
+ "model.layers.32.self_attn.q_a_proj",
119
+ "model.layers.32.self_attn.q_b_proj",
120
+ "model.layers.33.self_attn.kv_a_proj_with_mqa",
121
+ "model.layers.33.self_attn.kv_b_proj",
122
+ "model.layers.33.self_attn.q_a_proj",
123
+ "model.layers.33.self_attn.q_b_proj",
124
+ "model.layers.34.self_attn.kv_a_proj_with_mqa",
125
+ "model.layers.34.self_attn.kv_b_proj",
126
+ "model.layers.34.self_attn.q_a_proj",
127
+ "model.layers.34.self_attn.q_b_proj",
128
+ "model.layers.35.self_attn.kv_a_proj_with_mqa",
129
+ "model.layers.35.self_attn.kv_b_proj",
130
+ "model.layers.35.self_attn.q_a_proj",
131
+ "model.layers.35.self_attn.q_b_proj",
132
+ "model.layers.36.self_attn.kv_a_proj_with_mqa",
133
+ "model.layers.36.self_attn.kv_b_proj",
134
+ "model.layers.36.self_attn.q_a_proj",
135
+ "model.layers.36.self_attn.q_b_proj",
136
+ "model.layers.37.self_attn.kv_a_proj_with_mqa",
137
+ "model.layers.37.self_attn.kv_b_proj",
138
+ "model.layers.37.self_attn.q_a_proj",
139
+ "model.layers.37.self_attn.q_b_proj",
140
+ "model.layers.38.self_attn.kv_a_proj_with_mqa",
141
+ "model.layers.38.self_attn.kv_b_proj",
142
+ "model.layers.38.self_attn.q_a_proj",
143
+ "model.layers.38.self_attn.q_b_proj",
144
+ "model.layers.39.self_attn.kv_a_proj_with_mqa",
145
+ "model.layers.39.self_attn.kv_b_proj",
146
+ "model.layers.39.self_attn.q_a_proj",
147
+ "model.layers.39.self_attn.q_b_proj",
148
+ "model.layers.4.self_attn.kv_a_proj_with_mqa",
149
+ "model.layers.4.self_attn.kv_b_proj",
150
+ "model.layers.4.self_attn.q_a_proj",
151
+ "model.layers.4.self_attn.q_b_proj",
152
+ "model.layers.40.self_attn.kv_a_proj_with_mqa",
153
+ "model.layers.40.self_attn.kv_b_proj",
154
+ "model.layers.40.self_attn.q_a_proj",
155
+ "model.layers.40.self_attn.q_b_proj",
156
+ "model.layers.41.self_attn.kv_a_proj_with_mqa",
157
+ "model.layers.41.self_attn.kv_b_proj",
158
+ "model.layers.41.self_attn.q_a_proj",
159
+ "model.layers.41.self_attn.q_b_proj",
160
+ "model.layers.42.self_attn.kv_a_proj_with_mqa",
161
+ "model.layers.42.self_attn.kv_b_proj",
162
+ "model.layers.42.self_attn.q_a_proj",
163
+ "model.layers.42.self_attn.q_b_proj",
164
+ "model.layers.43.self_attn.kv_a_proj_with_mqa",
165
+ "model.layers.43.self_attn.kv_b_proj",
166
+ "model.layers.43.self_attn.q_a_proj",
167
+ "model.layers.43.self_attn.q_b_proj",
168
+ "model.layers.44.self_attn.kv_a_proj_with_mqa",
169
+ "model.layers.44.self_attn.kv_b_proj",
170
+ "model.layers.44.self_attn.q_a_proj",
171
+ "model.layers.44.self_attn.q_b_proj",
172
+ "model.layers.45.self_attn.kv_a_proj_with_mqa",
173
+ "model.layers.45.self_attn.kv_b_proj",
174
+ "model.layers.45.self_attn.q_a_proj",
175
+ "model.layers.45.self_attn.q_b_proj",
176
+ "model.layers.46.self_attn.kv_a_proj_with_mqa",
177
+ "model.layers.46.self_attn.kv_b_proj",
178
+ "model.layers.46.self_attn.q_a_proj",
179
+ "model.layers.46.self_attn.q_b_proj",
180
+ "model.layers.47.self_attn.kv_a_proj_with_mqa",
181
+ "model.layers.47.self_attn.kv_b_proj",
182
+ "model.layers.47.self_attn.q_a_proj",
183
+ "model.layers.47.self_attn.q_b_proj",
184
+ "model.layers.48.self_attn.kv_a_proj_with_mqa",
185
+ "model.layers.48.self_attn.kv_b_proj",
186
+ "model.layers.48.self_attn.q_a_proj",
187
+ "model.layers.48.self_attn.q_b_proj",
188
+ "model.layers.49.self_attn.kv_a_proj_with_mqa",
189
+ "model.layers.49.self_attn.kv_b_proj",
190
+ "model.layers.49.self_attn.q_a_proj",
191
+ "model.layers.49.self_attn.q_b_proj",
192
+ "model.layers.5.self_attn.kv_a_proj_with_mqa",
193
+ "model.layers.5.self_attn.kv_b_proj",
194
+ "model.layers.5.self_attn.q_a_proj",
195
+ "model.layers.5.self_attn.q_b_proj",
196
+ "model.layers.50.self_attn.kv_a_proj_with_mqa",
197
+ "model.layers.50.self_attn.kv_b_proj",
198
+ "model.layers.50.self_attn.q_a_proj",
199
+ "model.layers.50.self_attn.q_b_proj",
200
+ "model.layers.51.self_attn.kv_a_proj_with_mqa",
201
+ "model.layers.51.self_attn.kv_b_proj",
202
+ "model.layers.51.self_attn.q_a_proj",
203
+ "model.layers.51.self_attn.q_b_proj",
204
+ "model.layers.52.self_attn.kv_a_proj_with_mqa",
205
+ "model.layers.52.self_attn.kv_b_proj",
206
+ "model.layers.52.self_attn.q_a_proj",
207
+ "model.layers.52.self_attn.q_b_proj",
208
+ "model.layers.53.self_attn.kv_a_proj_with_mqa",
209
+ "model.layers.53.self_attn.kv_b_proj",
210
+ "model.layers.53.self_attn.q_a_proj",
211
+ "model.layers.53.self_attn.q_b_proj",
212
+ "model.layers.54.self_attn.kv_a_proj_with_mqa",
213
+ "model.layers.54.self_attn.kv_b_proj",
214
+ "model.layers.54.self_attn.q_a_proj",
215
+ "model.layers.54.self_attn.q_b_proj",
216
+ "model.layers.55.self_attn.kv_a_proj_with_mqa",
217
+ "model.layers.55.self_attn.kv_b_proj",
218
+ "model.layers.55.self_attn.q_a_proj",
219
+ "model.layers.55.self_attn.q_b_proj",
220
+ "model.layers.56.self_attn.kv_a_proj_with_mqa",
221
+ "model.layers.56.self_attn.kv_b_proj",
222
+ "model.layers.56.self_attn.q_a_proj",
223
+ "model.layers.56.self_attn.q_b_proj",
224
+ "model.layers.57.self_attn.kv_a_proj_with_mqa",
225
+ "model.layers.57.self_attn.kv_b_proj",
226
+ "model.layers.57.self_attn.q_a_proj",
227
+ "model.layers.57.self_attn.q_b_proj",
228
+ "model.layers.58.self_attn.kv_a_proj_with_mqa",
229
+ "model.layers.58.self_attn.kv_b_proj",
230
+ "model.layers.58.self_attn.q_a_proj",
231
+ "model.layers.58.self_attn.q_b_proj",
232
+ "model.layers.59.self_attn.kv_a_proj_with_mqa",
233
+ "model.layers.59.self_attn.kv_b_proj",
234
+ "model.layers.59.self_attn.q_a_proj",
235
+ "model.layers.59.self_attn.q_b_proj",
236
+ "model.layers.6.self_attn.kv_a_proj_with_mqa",
237
+ "model.layers.6.self_attn.kv_b_proj",
238
+ "model.layers.6.self_attn.q_a_proj",
239
+ "model.layers.6.self_attn.q_b_proj",
240
+ "model.layers.60.self_attn.kv_a_proj_with_mqa",
241
+ "model.layers.60.self_attn.kv_b_proj",
242
+ "model.layers.60.self_attn.q_a_proj",
243
+ "model.layers.60.self_attn.q_b_proj",
244
+ "model.layers.7.self_attn.kv_a_proj_with_mqa",
245
+ "model.layers.7.self_attn.kv_b_proj",
246
+ "model.layers.7.self_attn.q_a_proj",
247
+ "model.layers.7.self_attn.q_b_proj",
248
+ "model.layers.8.self_attn.kv_a_proj_with_mqa",
249
+ "model.layers.8.self_attn.kv_b_proj",
250
+ "model.layers.8.self_attn.q_a_proj",
251
+ "model.layers.8.self_attn.q_b_proj",
252
+ "model.layers.9.self_attn.kv_a_proj_with_mqa",
253
+ "model.layers.9.self_attn.kv_b_proj",
254
+ "model.layers.9.self_attn.q_a_proj",
255
+ "model.layers.9.self_attn.q_b_proj",
256
+ "model.layers.61*"
257
+ ]
258
+ }
259
+ }
model-00001-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cce4f6a407fa858e999495a7d6fb4c4570e0db56b393e8efd131c2e9f72b9bdc
3
+ size 4156845944
model-00002-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f33604bc9da094557b16b6f174d3d9e7eace614b567f17641659a9e99a162fcf
3
+ size 2419597576
model-00003-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:07edd4ecb38a75d315bebe28acf5c41a33d41312f8a30259fd65420678273b47
3
+ size 2419598384
model-00004-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1d55ad29ff79e096b984df54566b326ba33b346b7ffa0fd66658a1485d66daf1
3
+ size 2521338792
model-00005-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b626b4f81a9a302b8ec200e3357d6c8c243e400a95b3df9d096f2e4cbab173d7
3
+ size 2419597944
model-00006-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:84c19ed80f568a9fcb11a9f67be390804647d107ff8f36c41ad1f104c4527b1e
3
+ size 2558958464
model-00007-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6d397414936252a2f868375a631b6b1c9efcfb79b5b5c79bbe651e0cc60d16f3
3
+ size 2423268408
model-00008-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96817c49cdd5a43d5ece12164943f678e305bf36f9017a047a8ac188b57d97f0
3
+ size 2419598352
model-00009-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c2949bc565d439426223ee915b20cccf3836bcc69fe33c77bd60ce3710518b0
3
+ size 2521339176
model-00010-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f553cae040912675443d05fa133311916dcd4f0f42cf1a0773f1e080490330fa
3
+ size 2419597560
model-00011-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1fa5f58cc2633f083e618f454a813d4150d85f6b95d720e35884c3d612760c72
3
+ size 2419598384
model-00012-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a65e8032f6d0157fbb6305be1e1a523d49c78575bd8f4cc221feee72d4caa511
3
+ size 743249736
model-00013-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8bef9273b173d56c472e6e3f8463a9750f9e1cc7a9a89602478a1553c349cd0c
3
+ size 2521309560
model-00014-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d04f4d1a88cbdaa4df2d1e5babb2c84682a2a11d466e8779cc63e94abf03ef1
3
+ size 2419598296
model-00015-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38b826ba8766fa2d3fd85eb666470390d6b31ac10e645d244d75b841feb4fb5a
3
+ size 2521339232
model-00016-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cceb85c0356b705af0b8ac889ea9e5deff3f8c331f810b1237b59e976781f152
3
+ size 2419597504
model-00017-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6a06b3a986d22a6a2a77666523c60b75631601703ecd90765d934c0f361491f5
3
+ size 2419598384
model-00018-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0f8e6d457605832b942254b9a57fe0f16abb36f923b4fcc9bf8f08eb1ea6b897
3
+ size 2521338856
model-00019-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f443266729e797b8ad52a541b8ccdfd7881e444fe7f254d1ed2efb4cb9b8f7bf
3
+ size 2419597880
model-00020-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24fa2e531ed616be5087da5f7807af83fe17ba132cf3331b289c9f13cdbc3173
3
+ size 2419598392
model-00021-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:359869a44ed54a1b0f6226538076c3956950a187234f49cd8cd99c9d99e0ef23
3
+ size 2521339584
model-00022-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d3254ad0a06f2299cd785c4f03fd223279a459706151c4a389a5dca16329a38e
3
+ size 2419599440
model-00023-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:20045042c3466bb65eed8465791ff10e1bf1fc6b469101c2dc410c523a090c4e
3
+ size 2521340392
model-00024-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a79cfd3f891f13155ec110b2a87016676efab24d5e339fd12aa5910956ed12e
3
+ size 2419598648
model-00025-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22de7092a770df65646b7a7600c0a2b1fdcfd818b799286ea4f3529b294b719b
3
+ size 2419599560
model-00026-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:519a07efaffd31a663f866223c23183ff79586b725e935e21bbc4d733aff98be
3
+ size 2521340016
model-00027-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2420b8dd8ca1b05c00f3f1af0cf8a1f0d18d4d8968639975f68d919bbf150891
3
+ size 2419599024
model-00028-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c57416f058db0498f715596711b03e201122b6e4d758d6c7b3bd79605a67fd18
3
+ size 2419599560
model-00029-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4cb2f4e589fdbc20a3d25d9542d87ca849a5437bdbe01f26862779411783fd4a
3
+ size 2521339640
model-00030-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7fe24aabb4d964b89b428e71dc52173dd218d4b9c4f0db11c2fd14883e3b5e39
3
+ size 2419599400
model-00031-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ca8c7dbbedbf0b9370b9c6400533660fc68708f4ee25e97447126394c6ac1508
3
+ size 2521340432
model-00032-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:12c04a563a47f20d27bcb9779cd13a1c22c6c0fbdbf0176286a9dbb136645e31
3
+ size 2419598608
model-00033-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8a0207376a970e050d9a51c9b27ee359afd560d139a1c6cfa653ae46d2ed6e3
3
+ size 2419599568
model-00034-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cec4e4434149156d1112b6d34f5af01a5191ed8c34a3e80a03899b134e8b75a
3
+ size 982732544
model-00035-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2183d7058e7d56f4f25e9ba913b59a413d404d78f04c554d42ecce0b0d8c4bff
3
+ size 2521310688
model-00036-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c9474ea2a26bd20ada476dcdf272d2975ff1d1fc0957f1becc6010beb9e40d6f
3
+ size 2419599464
model-00037-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0b2de2720e162fc18349900aa9da63ee2f2963ed94b93773c556d7443cc0fc07
3
+ size 2521340368
model-00038-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:031cad2a2b030425273b3547caeab838326858d3a8714482b99bb3a728b69131
3
+ size 2419598672
model-00039-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:72c4a8185d3805be5d1f5a188d71b81af3307ff8fe649c404b91c9bc35f95c3b
3
+ size 2419599560
model-00040-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5a809e7dc46b540b0c9c689231b5cbdb05d26aeb53885f5152505c7fc1634473
3
+ size 2521339984
model-00041-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5234b69ef48104ce8e87ae3b18daad704fe0964cc2898755d2e275c675e930b
3
+ size 2419599056
model-00042-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b94c96b5ee64f808860a3760bd23d977fa22b1b96c90971836c7f401a37af40d
3
+ size 2419599568
model-00043-of-000163.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ad9b826245c74095222f0b0fd165403618598a81faf242d6506196ec02762800
3
+ size 2521339600