Dongfu Jiang commited on
Commit
07a7519
1 Parent(s): 9bd32d0

Upload folder using huggingface_hub

Browse files
config.json ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "adept/fuyu-8b",
3
+ "architectures": [
4
+ "MFuyuForCausalLM"
5
+ ],
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 1,
8
+ "eos_token_id": 71013,
9
+ "hidden_act": "relu2",
10
+ "hidden_dropout": 0.0,
11
+ "hidden_size": 4096,
12
+ "image_size": 300,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 16384,
15
+ "layer_norm_eps": 1e-05,
16
+ "max_position_embeddings": 16384,
17
+ "model_type": "fuyu",
18
+ "num_attention_heads": 64,
19
+ "num_channels": 3,
20
+ "num_hidden_layers": 36,
21
+ "partial_rotary_factor": 0.5,
22
+ "patch_size": 30,
23
+ "qk_layernorm": true,
24
+ "rope_scaling": null,
25
+ "rope_theta": 25000.0,
26
+ "text_config": {
27
+ "model_type": "persimmon"
28
+ },
29
+ "tie_word_embeddings": false,
30
+ "torch_dtype": "bfloat16",
31
+ "transformers_version": "4.37.0",
32
+ "use_cache": true,
33
+ "vocab_size": 262144
34
+ }
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 71013,
5
+ "transformers_version": "4.37.0"
6
+ }
model-00001-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8464739c2fcfa2e7e12e1241d78247d323b16cd9c2c865fcb8aca9e4cfe5f466
3
+ size 4966836144
model-00002-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a29b8ac20e23df823abdfd821cab1a6143a30a53404c250c8fc58972b2afa257
3
+ size 4967399488
model-00003-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea275013536806c2d2bd40c585a754e4fc8bb83b0d1f12b89bcb37a65076ff67
3
+ size 4967398232
model-00004-of-00004.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:96312d46d187ef3cc66aa2a31167585fdb723a445f893e9f495a2785d8b9cb76
3
+ size 3914952240
model.safetensors.index.json ADDED
@@ -0,0 +1,589 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 18816509952
4
+ },
5
+ "weight_map": {
6
+ "language_model.lm_head.weight": "model-00004-of-00004.safetensors",
7
+ "language_model.model.embed_tokens.weight": "model-00001-of-00004.safetensors",
8
+ "language_model.model.final_layernorm.bias": "model-00004-of-00004.safetensors",
9
+ "language_model.model.final_layernorm.weight": "model-00004-of-00004.safetensors",
10
+ "language_model.model.layers.0.input_layernorm.bias": "model-00001-of-00004.safetensors",
11
+ "language_model.model.layers.0.input_layernorm.weight": "model-00001-of-00004.safetensors",
12
+ "language_model.model.layers.0.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
13
+ "language_model.model.layers.0.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
14
+ "language_model.model.layers.0.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
15
+ "language_model.model.layers.0.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
16
+ "language_model.model.layers.0.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
17
+ "language_model.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
18
+ "language_model.model.layers.0.self_attn.dense.bias": "model-00001-of-00004.safetensors",
19
+ "language_model.model.layers.0.self_attn.dense.weight": "model-00001-of-00004.safetensors",
20
+ "language_model.model.layers.0.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
21
+ "language_model.model.layers.0.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
22
+ "language_model.model.layers.0.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
23
+ "language_model.model.layers.0.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
24
+ "language_model.model.layers.0.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
25
+ "language_model.model.layers.0.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
26
+ "language_model.model.layers.1.input_layernorm.bias": "model-00001-of-00004.safetensors",
27
+ "language_model.model.layers.1.input_layernorm.weight": "model-00001-of-00004.safetensors",
28
+ "language_model.model.layers.1.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
29
+ "language_model.model.layers.1.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
30
+ "language_model.model.layers.1.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
31
+ "language_model.model.layers.1.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
32
+ "language_model.model.layers.1.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
33
+ "language_model.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
34
+ "language_model.model.layers.1.self_attn.dense.bias": "model-00001-of-00004.safetensors",
35
+ "language_model.model.layers.1.self_attn.dense.weight": "model-00001-of-00004.safetensors",
36
+ "language_model.model.layers.1.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
37
+ "language_model.model.layers.1.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
38
+ "language_model.model.layers.1.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
39
+ "language_model.model.layers.1.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
40
+ "language_model.model.layers.1.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
41
+ "language_model.model.layers.1.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
42
+ "language_model.model.layers.10.input_layernorm.bias": "model-00002-of-00004.safetensors",
43
+ "language_model.model.layers.10.input_layernorm.weight": "model-00002-of-00004.safetensors",
44
+ "language_model.model.layers.10.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
45
+ "language_model.model.layers.10.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
46
+ "language_model.model.layers.10.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
47
+ "language_model.model.layers.10.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
48
+ "language_model.model.layers.10.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
49
+ "language_model.model.layers.10.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
50
+ "language_model.model.layers.10.self_attn.dense.bias": "model-00002-of-00004.safetensors",
51
+ "language_model.model.layers.10.self_attn.dense.weight": "model-00002-of-00004.safetensors",
52
+ "language_model.model.layers.10.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
53
+ "language_model.model.layers.10.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
54
+ "language_model.model.layers.10.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
55
+ "language_model.model.layers.10.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
56
+ "language_model.model.layers.10.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
57
+ "language_model.model.layers.10.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
58
+ "language_model.model.layers.11.input_layernorm.bias": "model-00002-of-00004.safetensors",
59
+ "language_model.model.layers.11.input_layernorm.weight": "model-00002-of-00004.safetensors",
60
+ "language_model.model.layers.11.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
61
+ "language_model.model.layers.11.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
62
+ "language_model.model.layers.11.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
63
+ "language_model.model.layers.11.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
64
+ "language_model.model.layers.11.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
65
+ "language_model.model.layers.11.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
66
+ "language_model.model.layers.11.self_attn.dense.bias": "model-00002-of-00004.safetensors",
67
+ "language_model.model.layers.11.self_attn.dense.weight": "model-00002-of-00004.safetensors",
68
+ "language_model.model.layers.11.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
69
+ "language_model.model.layers.11.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
70
+ "language_model.model.layers.11.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
71
+ "language_model.model.layers.11.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
72
+ "language_model.model.layers.11.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
73
+ "language_model.model.layers.11.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
74
+ "language_model.model.layers.12.input_layernorm.bias": "model-00002-of-00004.safetensors",
75
+ "language_model.model.layers.12.input_layernorm.weight": "model-00002-of-00004.safetensors",
76
+ "language_model.model.layers.12.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
77
+ "language_model.model.layers.12.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
78
+ "language_model.model.layers.12.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
79
+ "language_model.model.layers.12.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
80
+ "language_model.model.layers.12.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
81
+ "language_model.model.layers.12.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
82
+ "language_model.model.layers.12.self_attn.dense.bias": "model-00002-of-00004.safetensors",
83
+ "language_model.model.layers.12.self_attn.dense.weight": "model-00002-of-00004.safetensors",
84
+ "language_model.model.layers.12.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
85
+ "language_model.model.layers.12.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
86
+ "language_model.model.layers.12.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
87
+ "language_model.model.layers.12.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
88
+ "language_model.model.layers.12.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
89
+ "language_model.model.layers.12.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
90
+ "language_model.model.layers.13.input_layernorm.bias": "model-00002-of-00004.safetensors",
91
+ "language_model.model.layers.13.input_layernorm.weight": "model-00002-of-00004.safetensors",
92
+ "language_model.model.layers.13.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
93
+ "language_model.model.layers.13.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
94
+ "language_model.model.layers.13.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
95
+ "language_model.model.layers.13.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
96
+ "language_model.model.layers.13.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
97
+ "language_model.model.layers.13.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
98
+ "language_model.model.layers.13.self_attn.dense.bias": "model-00002-of-00004.safetensors",
99
+ "language_model.model.layers.13.self_attn.dense.weight": "model-00002-of-00004.safetensors",
100
+ "language_model.model.layers.13.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
101
+ "language_model.model.layers.13.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
102
+ "language_model.model.layers.13.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
103
+ "language_model.model.layers.13.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
104
+ "language_model.model.layers.13.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
105
+ "language_model.model.layers.13.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
106
+ "language_model.model.layers.14.input_layernorm.bias": "model-00002-of-00004.safetensors",
107
+ "language_model.model.layers.14.input_layernorm.weight": "model-00002-of-00004.safetensors",
108
+ "language_model.model.layers.14.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
109
+ "language_model.model.layers.14.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
110
+ "language_model.model.layers.14.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
111
+ "language_model.model.layers.14.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
112
+ "language_model.model.layers.14.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
113
+ "language_model.model.layers.14.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
114
+ "language_model.model.layers.14.self_attn.dense.bias": "model-00002-of-00004.safetensors",
115
+ "language_model.model.layers.14.self_attn.dense.weight": "model-00002-of-00004.safetensors",
116
+ "language_model.model.layers.14.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
117
+ "language_model.model.layers.14.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
118
+ "language_model.model.layers.14.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
119
+ "language_model.model.layers.14.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
120
+ "language_model.model.layers.14.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
121
+ "language_model.model.layers.14.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
122
+ "language_model.model.layers.15.input_layernorm.bias": "model-00002-of-00004.safetensors",
123
+ "language_model.model.layers.15.input_layernorm.weight": "model-00002-of-00004.safetensors",
124
+ "language_model.model.layers.15.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
125
+ "language_model.model.layers.15.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
126
+ "language_model.model.layers.15.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
127
+ "language_model.model.layers.15.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
128
+ "language_model.model.layers.15.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
129
+ "language_model.model.layers.15.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
130
+ "language_model.model.layers.15.self_attn.dense.bias": "model-00002-of-00004.safetensors",
131
+ "language_model.model.layers.15.self_attn.dense.weight": "model-00002-of-00004.safetensors",
132
+ "language_model.model.layers.15.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
133
+ "language_model.model.layers.15.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
134
+ "language_model.model.layers.15.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
135
+ "language_model.model.layers.15.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
136
+ "language_model.model.layers.15.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
137
+ "language_model.model.layers.15.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
138
+ "language_model.model.layers.16.input_layernorm.bias": "model-00002-of-00004.safetensors",
139
+ "language_model.model.layers.16.input_layernorm.weight": "model-00002-of-00004.safetensors",
140
+ "language_model.model.layers.16.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
141
+ "language_model.model.layers.16.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
142
+ "language_model.model.layers.16.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
143
+ "language_model.model.layers.16.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
144
+ "language_model.model.layers.16.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
145
+ "language_model.model.layers.16.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
146
+ "language_model.model.layers.16.self_attn.dense.bias": "model-00002-of-00004.safetensors",
147
+ "language_model.model.layers.16.self_attn.dense.weight": "model-00002-of-00004.safetensors",
148
+ "language_model.model.layers.16.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
149
+ "language_model.model.layers.16.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
150
+ "language_model.model.layers.16.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
151
+ "language_model.model.layers.16.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
152
+ "language_model.model.layers.16.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
153
+ "language_model.model.layers.16.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
154
+ "language_model.model.layers.17.input_layernorm.bias": "model-00002-of-00004.safetensors",
155
+ "language_model.model.layers.17.input_layernorm.weight": "model-00002-of-00004.safetensors",
156
+ "language_model.model.layers.17.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
157
+ "language_model.model.layers.17.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
158
+ "language_model.model.layers.17.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
159
+ "language_model.model.layers.17.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
160
+ "language_model.model.layers.17.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
161
+ "language_model.model.layers.17.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
162
+ "language_model.model.layers.17.self_attn.dense.bias": "model-00002-of-00004.safetensors",
163
+ "language_model.model.layers.17.self_attn.dense.weight": "model-00002-of-00004.safetensors",
164
+ "language_model.model.layers.17.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
165
+ "language_model.model.layers.17.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
166
+ "language_model.model.layers.17.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
167
+ "language_model.model.layers.17.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
168
+ "language_model.model.layers.17.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
169
+ "language_model.model.layers.17.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
170
+ "language_model.model.layers.18.input_layernorm.bias": "model-00002-of-00004.safetensors",
171
+ "language_model.model.layers.18.input_layernorm.weight": "model-00002-of-00004.safetensors",
172
+ "language_model.model.layers.18.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
173
+ "language_model.model.layers.18.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
174
+ "language_model.model.layers.18.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
175
+ "language_model.model.layers.18.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
176
+ "language_model.model.layers.18.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
177
+ "language_model.model.layers.18.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
178
+ "language_model.model.layers.18.self_attn.dense.bias": "model-00002-of-00004.safetensors",
179
+ "language_model.model.layers.18.self_attn.dense.weight": "model-00002-of-00004.safetensors",
180
+ "language_model.model.layers.18.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
181
+ "language_model.model.layers.18.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
182
+ "language_model.model.layers.18.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
183
+ "language_model.model.layers.18.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
184
+ "language_model.model.layers.18.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
185
+ "language_model.model.layers.18.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
186
+ "language_model.model.layers.19.input_layernorm.bias": "model-00003-of-00004.safetensors",
187
+ "language_model.model.layers.19.input_layernorm.weight": "model-00003-of-00004.safetensors",
188
+ "language_model.model.layers.19.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
189
+ "language_model.model.layers.19.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
190
+ "language_model.model.layers.19.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
191
+ "language_model.model.layers.19.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
192
+ "language_model.model.layers.19.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
193
+ "language_model.model.layers.19.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
194
+ "language_model.model.layers.19.self_attn.dense.bias": "model-00002-of-00004.safetensors",
195
+ "language_model.model.layers.19.self_attn.dense.weight": "model-00002-of-00004.safetensors",
196
+ "language_model.model.layers.19.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
197
+ "language_model.model.layers.19.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
198
+ "language_model.model.layers.19.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
199
+ "language_model.model.layers.19.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
200
+ "language_model.model.layers.19.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
201
+ "language_model.model.layers.19.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
202
+ "language_model.model.layers.2.input_layernorm.bias": "model-00001-of-00004.safetensors",
203
+ "language_model.model.layers.2.input_layernorm.weight": "model-00001-of-00004.safetensors",
204
+ "language_model.model.layers.2.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
205
+ "language_model.model.layers.2.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
206
+ "language_model.model.layers.2.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
207
+ "language_model.model.layers.2.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
208
+ "language_model.model.layers.2.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
209
+ "language_model.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
210
+ "language_model.model.layers.2.self_attn.dense.bias": "model-00001-of-00004.safetensors",
211
+ "language_model.model.layers.2.self_attn.dense.weight": "model-00001-of-00004.safetensors",
212
+ "language_model.model.layers.2.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
213
+ "language_model.model.layers.2.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
214
+ "language_model.model.layers.2.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
215
+ "language_model.model.layers.2.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
216
+ "language_model.model.layers.2.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
217
+ "language_model.model.layers.2.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
218
+ "language_model.model.layers.20.input_layernorm.bias": "model-00003-of-00004.safetensors",
219
+ "language_model.model.layers.20.input_layernorm.weight": "model-00003-of-00004.safetensors",
220
+ "language_model.model.layers.20.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
221
+ "language_model.model.layers.20.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
222
+ "language_model.model.layers.20.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
223
+ "language_model.model.layers.20.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
224
+ "language_model.model.layers.20.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
225
+ "language_model.model.layers.20.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
226
+ "language_model.model.layers.20.self_attn.dense.bias": "model-00003-of-00004.safetensors",
227
+ "language_model.model.layers.20.self_attn.dense.weight": "model-00003-of-00004.safetensors",
228
+ "language_model.model.layers.20.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
229
+ "language_model.model.layers.20.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
230
+ "language_model.model.layers.20.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
231
+ "language_model.model.layers.20.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
232
+ "language_model.model.layers.20.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
233
+ "language_model.model.layers.20.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
234
+ "language_model.model.layers.21.input_layernorm.bias": "model-00003-of-00004.safetensors",
235
+ "language_model.model.layers.21.input_layernorm.weight": "model-00003-of-00004.safetensors",
236
+ "language_model.model.layers.21.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
237
+ "language_model.model.layers.21.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
238
+ "language_model.model.layers.21.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
239
+ "language_model.model.layers.21.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
240
+ "language_model.model.layers.21.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
241
+ "language_model.model.layers.21.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
242
+ "language_model.model.layers.21.self_attn.dense.bias": "model-00003-of-00004.safetensors",
243
+ "language_model.model.layers.21.self_attn.dense.weight": "model-00003-of-00004.safetensors",
244
+ "language_model.model.layers.21.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
245
+ "language_model.model.layers.21.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
246
+ "language_model.model.layers.21.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
247
+ "language_model.model.layers.21.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
248
+ "language_model.model.layers.21.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
249
+ "language_model.model.layers.21.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
250
+ "language_model.model.layers.22.input_layernorm.bias": "model-00003-of-00004.safetensors",
251
+ "language_model.model.layers.22.input_layernorm.weight": "model-00003-of-00004.safetensors",
252
+ "language_model.model.layers.22.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
253
+ "language_model.model.layers.22.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
254
+ "language_model.model.layers.22.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
255
+ "language_model.model.layers.22.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
256
+ "language_model.model.layers.22.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
257
+ "language_model.model.layers.22.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
258
+ "language_model.model.layers.22.self_attn.dense.bias": "model-00003-of-00004.safetensors",
259
+ "language_model.model.layers.22.self_attn.dense.weight": "model-00003-of-00004.safetensors",
260
+ "language_model.model.layers.22.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
261
+ "language_model.model.layers.22.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
262
+ "language_model.model.layers.22.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
263
+ "language_model.model.layers.22.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
264
+ "language_model.model.layers.22.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
265
+ "language_model.model.layers.22.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
266
+ "language_model.model.layers.23.input_layernorm.bias": "model-00003-of-00004.safetensors",
267
+ "language_model.model.layers.23.input_layernorm.weight": "model-00003-of-00004.safetensors",
268
+ "language_model.model.layers.23.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
269
+ "language_model.model.layers.23.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
270
+ "language_model.model.layers.23.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
271
+ "language_model.model.layers.23.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
272
+ "language_model.model.layers.23.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
273
+ "language_model.model.layers.23.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
274
+ "language_model.model.layers.23.self_attn.dense.bias": "model-00003-of-00004.safetensors",
275
+ "language_model.model.layers.23.self_attn.dense.weight": "model-00003-of-00004.safetensors",
276
+ "language_model.model.layers.23.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
277
+ "language_model.model.layers.23.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
278
+ "language_model.model.layers.23.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
279
+ "language_model.model.layers.23.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
280
+ "language_model.model.layers.23.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
281
+ "language_model.model.layers.23.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
282
+ "language_model.model.layers.24.input_layernorm.bias": "model-00003-of-00004.safetensors",
283
+ "language_model.model.layers.24.input_layernorm.weight": "model-00003-of-00004.safetensors",
284
+ "language_model.model.layers.24.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
285
+ "language_model.model.layers.24.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
286
+ "language_model.model.layers.24.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
287
+ "language_model.model.layers.24.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
288
+ "language_model.model.layers.24.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
289
+ "language_model.model.layers.24.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
290
+ "language_model.model.layers.24.self_attn.dense.bias": "model-00003-of-00004.safetensors",
291
+ "language_model.model.layers.24.self_attn.dense.weight": "model-00003-of-00004.safetensors",
292
+ "language_model.model.layers.24.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
293
+ "language_model.model.layers.24.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
294
+ "language_model.model.layers.24.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
295
+ "language_model.model.layers.24.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
296
+ "language_model.model.layers.24.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
297
+ "language_model.model.layers.24.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
298
+ "language_model.model.layers.25.input_layernorm.bias": "model-00003-of-00004.safetensors",
299
+ "language_model.model.layers.25.input_layernorm.weight": "model-00003-of-00004.safetensors",
300
+ "language_model.model.layers.25.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
301
+ "language_model.model.layers.25.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
302
+ "language_model.model.layers.25.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
303
+ "language_model.model.layers.25.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
304
+ "language_model.model.layers.25.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
305
+ "language_model.model.layers.25.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
306
+ "language_model.model.layers.25.self_attn.dense.bias": "model-00003-of-00004.safetensors",
307
+ "language_model.model.layers.25.self_attn.dense.weight": "model-00003-of-00004.safetensors",
308
+ "language_model.model.layers.25.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
309
+ "language_model.model.layers.25.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
310
+ "language_model.model.layers.25.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
311
+ "language_model.model.layers.25.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
312
+ "language_model.model.layers.25.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
313
+ "language_model.model.layers.25.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
314
+ "language_model.model.layers.26.input_layernorm.bias": "model-00003-of-00004.safetensors",
315
+ "language_model.model.layers.26.input_layernorm.weight": "model-00003-of-00004.safetensors",
316
+ "language_model.model.layers.26.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
317
+ "language_model.model.layers.26.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
318
+ "language_model.model.layers.26.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
319
+ "language_model.model.layers.26.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
320
+ "language_model.model.layers.26.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
321
+ "language_model.model.layers.26.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
322
+ "language_model.model.layers.26.self_attn.dense.bias": "model-00003-of-00004.safetensors",
323
+ "language_model.model.layers.26.self_attn.dense.weight": "model-00003-of-00004.safetensors",
324
+ "language_model.model.layers.26.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
325
+ "language_model.model.layers.26.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
326
+ "language_model.model.layers.26.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
327
+ "language_model.model.layers.26.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
328
+ "language_model.model.layers.26.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
329
+ "language_model.model.layers.26.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
330
+ "language_model.model.layers.27.input_layernorm.bias": "model-00003-of-00004.safetensors",
331
+ "language_model.model.layers.27.input_layernorm.weight": "model-00003-of-00004.safetensors",
332
+ "language_model.model.layers.27.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
333
+ "language_model.model.layers.27.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
334
+ "language_model.model.layers.27.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
335
+ "language_model.model.layers.27.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
336
+ "language_model.model.layers.27.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
337
+ "language_model.model.layers.27.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
338
+ "language_model.model.layers.27.self_attn.dense.bias": "model-00003-of-00004.safetensors",
339
+ "language_model.model.layers.27.self_attn.dense.weight": "model-00003-of-00004.safetensors",
340
+ "language_model.model.layers.27.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
341
+ "language_model.model.layers.27.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
342
+ "language_model.model.layers.27.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
343
+ "language_model.model.layers.27.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
344
+ "language_model.model.layers.27.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
345
+ "language_model.model.layers.27.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
346
+ "language_model.model.layers.28.input_layernorm.bias": "model-00003-of-00004.safetensors",
347
+ "language_model.model.layers.28.input_layernorm.weight": "model-00003-of-00004.safetensors",
348
+ "language_model.model.layers.28.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
349
+ "language_model.model.layers.28.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
350
+ "language_model.model.layers.28.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
351
+ "language_model.model.layers.28.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
352
+ "language_model.model.layers.28.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
353
+ "language_model.model.layers.28.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
354
+ "language_model.model.layers.28.self_attn.dense.bias": "model-00003-of-00004.safetensors",
355
+ "language_model.model.layers.28.self_attn.dense.weight": "model-00003-of-00004.safetensors",
356
+ "language_model.model.layers.28.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
357
+ "language_model.model.layers.28.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
358
+ "language_model.model.layers.28.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
359
+ "language_model.model.layers.28.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
360
+ "language_model.model.layers.28.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
361
+ "language_model.model.layers.28.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
362
+ "language_model.model.layers.29.input_layernorm.bias": "model-00003-of-00004.safetensors",
363
+ "language_model.model.layers.29.input_layernorm.weight": "model-00003-of-00004.safetensors",
364
+ "language_model.model.layers.29.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
365
+ "language_model.model.layers.29.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
366
+ "language_model.model.layers.29.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
367
+ "language_model.model.layers.29.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
368
+ "language_model.model.layers.29.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
369
+ "language_model.model.layers.29.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
370
+ "language_model.model.layers.29.self_attn.dense.bias": "model-00003-of-00004.safetensors",
371
+ "language_model.model.layers.29.self_attn.dense.weight": "model-00003-of-00004.safetensors",
372
+ "language_model.model.layers.29.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
373
+ "language_model.model.layers.29.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
374
+ "language_model.model.layers.29.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
375
+ "language_model.model.layers.29.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
376
+ "language_model.model.layers.29.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
377
+ "language_model.model.layers.29.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
378
+ "language_model.model.layers.3.input_layernorm.bias": "model-00001-of-00004.safetensors",
379
+ "language_model.model.layers.3.input_layernorm.weight": "model-00001-of-00004.safetensors",
380
+ "language_model.model.layers.3.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
381
+ "language_model.model.layers.3.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
382
+ "language_model.model.layers.3.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
383
+ "language_model.model.layers.3.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
384
+ "language_model.model.layers.3.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
385
+ "language_model.model.layers.3.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
386
+ "language_model.model.layers.3.self_attn.dense.bias": "model-00001-of-00004.safetensors",
387
+ "language_model.model.layers.3.self_attn.dense.weight": "model-00001-of-00004.safetensors",
388
+ "language_model.model.layers.3.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
389
+ "language_model.model.layers.3.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
390
+ "language_model.model.layers.3.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
391
+ "language_model.model.layers.3.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
392
+ "language_model.model.layers.3.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
393
+ "language_model.model.layers.3.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
394
+ "language_model.model.layers.30.input_layernorm.bias": "model-00003-of-00004.safetensors",
395
+ "language_model.model.layers.30.input_layernorm.weight": "model-00003-of-00004.safetensors",
396
+ "language_model.model.layers.30.mlp.dense_4h_to_h.bias": "model-00003-of-00004.safetensors",
397
+ "language_model.model.layers.30.mlp.dense_4h_to_h.weight": "model-00003-of-00004.safetensors",
398
+ "language_model.model.layers.30.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
399
+ "language_model.model.layers.30.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
400
+ "language_model.model.layers.30.post_attention_layernorm.bias": "model-00003-of-00004.safetensors",
401
+ "language_model.model.layers.30.post_attention_layernorm.weight": "model-00003-of-00004.safetensors",
402
+ "language_model.model.layers.30.self_attn.dense.bias": "model-00003-of-00004.safetensors",
403
+ "language_model.model.layers.30.self_attn.dense.weight": "model-00003-of-00004.safetensors",
404
+ "language_model.model.layers.30.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
405
+ "language_model.model.layers.30.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
406
+ "language_model.model.layers.30.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
407
+ "language_model.model.layers.30.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
408
+ "language_model.model.layers.30.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
409
+ "language_model.model.layers.30.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
410
+ "language_model.model.layers.31.input_layernorm.bias": "model-00004-of-00004.safetensors",
411
+ "language_model.model.layers.31.input_layernorm.weight": "model-00004-of-00004.safetensors",
412
+ "language_model.model.layers.31.mlp.dense_4h_to_h.bias": "model-00004-of-00004.safetensors",
413
+ "language_model.model.layers.31.mlp.dense_4h_to_h.weight": "model-00004-of-00004.safetensors",
414
+ "language_model.model.layers.31.mlp.dense_h_to_4h.bias": "model-00003-of-00004.safetensors",
415
+ "language_model.model.layers.31.mlp.dense_h_to_4h.weight": "model-00003-of-00004.safetensors",
416
+ "language_model.model.layers.31.post_attention_layernorm.bias": "model-00004-of-00004.safetensors",
417
+ "language_model.model.layers.31.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
418
+ "language_model.model.layers.31.self_attn.dense.bias": "model-00003-of-00004.safetensors",
419
+ "language_model.model.layers.31.self_attn.dense.weight": "model-00003-of-00004.safetensors",
420
+ "language_model.model.layers.31.self_attn.k_layernorm.bias": "model-00003-of-00004.safetensors",
421
+ "language_model.model.layers.31.self_attn.k_layernorm.weight": "model-00003-of-00004.safetensors",
422
+ "language_model.model.layers.31.self_attn.q_layernorm.bias": "model-00003-of-00004.safetensors",
423
+ "language_model.model.layers.31.self_attn.q_layernorm.weight": "model-00003-of-00004.safetensors",
424
+ "language_model.model.layers.31.self_attn.query_key_value.bias": "model-00003-of-00004.safetensors",
425
+ "language_model.model.layers.31.self_attn.query_key_value.weight": "model-00003-of-00004.safetensors",
426
+ "language_model.model.layers.32.input_layernorm.bias": "model-00004-of-00004.safetensors",
427
+ "language_model.model.layers.32.input_layernorm.weight": "model-00004-of-00004.safetensors",
428
+ "language_model.model.layers.32.mlp.dense_4h_to_h.bias": "model-00004-of-00004.safetensors",
429
+ "language_model.model.layers.32.mlp.dense_4h_to_h.weight": "model-00004-of-00004.safetensors",
430
+ "language_model.model.layers.32.mlp.dense_h_to_4h.bias": "model-00004-of-00004.safetensors",
431
+ "language_model.model.layers.32.mlp.dense_h_to_4h.weight": "model-00004-of-00004.safetensors",
432
+ "language_model.model.layers.32.post_attention_layernorm.bias": "model-00004-of-00004.safetensors",
433
+ "language_model.model.layers.32.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
434
+ "language_model.model.layers.32.self_attn.dense.bias": "model-00004-of-00004.safetensors",
435
+ "language_model.model.layers.32.self_attn.dense.weight": "model-00004-of-00004.safetensors",
436
+ "language_model.model.layers.32.self_attn.k_layernorm.bias": "model-00004-of-00004.safetensors",
437
+ "language_model.model.layers.32.self_attn.k_layernorm.weight": "model-00004-of-00004.safetensors",
438
+ "language_model.model.layers.32.self_attn.q_layernorm.bias": "model-00004-of-00004.safetensors",
439
+ "language_model.model.layers.32.self_attn.q_layernorm.weight": "model-00004-of-00004.safetensors",
440
+ "language_model.model.layers.32.self_attn.query_key_value.bias": "model-00004-of-00004.safetensors",
441
+ "language_model.model.layers.32.self_attn.query_key_value.weight": "model-00004-of-00004.safetensors",
442
+ "language_model.model.layers.33.input_layernorm.bias": "model-00004-of-00004.safetensors",
443
+ "language_model.model.layers.33.input_layernorm.weight": "model-00004-of-00004.safetensors",
444
+ "language_model.model.layers.33.mlp.dense_4h_to_h.bias": "model-00004-of-00004.safetensors",
445
+ "language_model.model.layers.33.mlp.dense_4h_to_h.weight": "model-00004-of-00004.safetensors",
446
+ "language_model.model.layers.33.mlp.dense_h_to_4h.bias": "model-00004-of-00004.safetensors",
447
+ "language_model.model.layers.33.mlp.dense_h_to_4h.weight": "model-00004-of-00004.safetensors",
448
+ "language_model.model.layers.33.post_attention_layernorm.bias": "model-00004-of-00004.safetensors",
449
+ "language_model.model.layers.33.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
450
+ "language_model.model.layers.33.self_attn.dense.bias": "model-00004-of-00004.safetensors",
451
+ "language_model.model.layers.33.self_attn.dense.weight": "model-00004-of-00004.safetensors",
452
+ "language_model.model.layers.33.self_attn.k_layernorm.bias": "model-00004-of-00004.safetensors",
453
+ "language_model.model.layers.33.self_attn.k_layernorm.weight": "model-00004-of-00004.safetensors",
454
+ "language_model.model.layers.33.self_attn.q_layernorm.bias": "model-00004-of-00004.safetensors",
455
+ "language_model.model.layers.33.self_attn.q_layernorm.weight": "model-00004-of-00004.safetensors",
456
+ "language_model.model.layers.33.self_attn.query_key_value.bias": "model-00004-of-00004.safetensors",
457
+ "language_model.model.layers.33.self_attn.query_key_value.weight": "model-00004-of-00004.safetensors",
458
+ "language_model.model.layers.34.input_layernorm.bias": "model-00004-of-00004.safetensors",
459
+ "language_model.model.layers.34.input_layernorm.weight": "model-00004-of-00004.safetensors",
460
+ "language_model.model.layers.34.mlp.dense_4h_to_h.bias": "model-00004-of-00004.safetensors",
461
+ "language_model.model.layers.34.mlp.dense_4h_to_h.weight": "model-00004-of-00004.safetensors",
462
+ "language_model.model.layers.34.mlp.dense_h_to_4h.bias": "model-00004-of-00004.safetensors",
463
+ "language_model.model.layers.34.mlp.dense_h_to_4h.weight": "model-00004-of-00004.safetensors",
464
+ "language_model.model.layers.34.post_attention_layernorm.bias": "model-00004-of-00004.safetensors",
465
+ "language_model.model.layers.34.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
466
+ "language_model.model.layers.34.self_attn.dense.bias": "model-00004-of-00004.safetensors",
467
+ "language_model.model.layers.34.self_attn.dense.weight": "model-00004-of-00004.safetensors",
468
+ "language_model.model.layers.34.self_attn.k_layernorm.bias": "model-00004-of-00004.safetensors",
469
+ "language_model.model.layers.34.self_attn.k_layernorm.weight": "model-00004-of-00004.safetensors",
470
+ "language_model.model.layers.34.self_attn.q_layernorm.bias": "model-00004-of-00004.safetensors",
471
+ "language_model.model.layers.34.self_attn.q_layernorm.weight": "model-00004-of-00004.safetensors",
472
+ "language_model.model.layers.34.self_attn.query_key_value.bias": "model-00004-of-00004.safetensors",
473
+ "language_model.model.layers.34.self_attn.query_key_value.weight": "model-00004-of-00004.safetensors",
474
+ "language_model.model.layers.35.input_layernorm.bias": "model-00004-of-00004.safetensors",
475
+ "language_model.model.layers.35.input_layernorm.weight": "model-00004-of-00004.safetensors",
476
+ "language_model.model.layers.35.mlp.dense_4h_to_h.bias": "model-00004-of-00004.safetensors",
477
+ "language_model.model.layers.35.mlp.dense_4h_to_h.weight": "model-00004-of-00004.safetensors",
478
+ "language_model.model.layers.35.mlp.dense_h_to_4h.bias": "model-00004-of-00004.safetensors",
479
+ "language_model.model.layers.35.mlp.dense_h_to_4h.weight": "model-00004-of-00004.safetensors",
480
+ "language_model.model.layers.35.post_attention_layernorm.bias": "model-00004-of-00004.safetensors",
481
+ "language_model.model.layers.35.post_attention_layernorm.weight": "model-00004-of-00004.safetensors",
482
+ "language_model.model.layers.35.self_attn.dense.bias": "model-00004-of-00004.safetensors",
483
+ "language_model.model.layers.35.self_attn.dense.weight": "model-00004-of-00004.safetensors",
484
+ "language_model.model.layers.35.self_attn.k_layernorm.bias": "model-00004-of-00004.safetensors",
485
+ "language_model.model.layers.35.self_attn.k_layernorm.weight": "model-00004-of-00004.safetensors",
486
+ "language_model.model.layers.35.self_attn.q_layernorm.bias": "model-00004-of-00004.safetensors",
487
+ "language_model.model.layers.35.self_attn.q_layernorm.weight": "model-00004-of-00004.safetensors",
488
+ "language_model.model.layers.35.self_attn.query_key_value.bias": "model-00004-of-00004.safetensors",
489
+ "language_model.model.layers.35.self_attn.query_key_value.weight": "model-00004-of-00004.safetensors",
490
+ "language_model.model.layers.4.input_layernorm.bias": "model-00001-of-00004.safetensors",
491
+ "language_model.model.layers.4.input_layernorm.weight": "model-00001-of-00004.safetensors",
492
+ "language_model.model.layers.4.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
493
+ "language_model.model.layers.4.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
494
+ "language_model.model.layers.4.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
495
+ "language_model.model.layers.4.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
496
+ "language_model.model.layers.4.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
497
+ "language_model.model.layers.4.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
498
+ "language_model.model.layers.4.self_attn.dense.bias": "model-00001-of-00004.safetensors",
499
+ "language_model.model.layers.4.self_attn.dense.weight": "model-00001-of-00004.safetensors",
500
+ "language_model.model.layers.4.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
501
+ "language_model.model.layers.4.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
502
+ "language_model.model.layers.4.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
503
+ "language_model.model.layers.4.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
504
+ "language_model.model.layers.4.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
505
+ "language_model.model.layers.4.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
506
+ "language_model.model.layers.5.input_layernorm.bias": "model-00001-of-00004.safetensors",
507
+ "language_model.model.layers.5.input_layernorm.weight": "model-00001-of-00004.safetensors",
508
+ "language_model.model.layers.5.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
509
+ "language_model.model.layers.5.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
510
+ "language_model.model.layers.5.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
511
+ "language_model.model.layers.5.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
512
+ "language_model.model.layers.5.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
513
+ "language_model.model.layers.5.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
514
+ "language_model.model.layers.5.self_attn.dense.bias": "model-00001-of-00004.safetensors",
515
+ "language_model.model.layers.5.self_attn.dense.weight": "model-00001-of-00004.safetensors",
516
+ "language_model.model.layers.5.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
517
+ "language_model.model.layers.5.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
518
+ "language_model.model.layers.5.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
519
+ "language_model.model.layers.5.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
520
+ "language_model.model.layers.5.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
521
+ "language_model.model.layers.5.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
522
+ "language_model.model.layers.6.input_layernorm.bias": "model-00001-of-00004.safetensors",
523
+ "language_model.model.layers.6.input_layernorm.weight": "model-00001-of-00004.safetensors",
524
+ "language_model.model.layers.6.mlp.dense_4h_to_h.bias": "model-00001-of-00004.safetensors",
525
+ "language_model.model.layers.6.mlp.dense_4h_to_h.weight": "model-00001-of-00004.safetensors",
526
+ "language_model.model.layers.6.mlp.dense_h_to_4h.bias": "model-00001-of-00004.safetensors",
527
+ "language_model.model.layers.6.mlp.dense_h_to_4h.weight": "model-00001-of-00004.safetensors",
528
+ "language_model.model.layers.6.post_attention_layernorm.bias": "model-00001-of-00004.safetensors",
529
+ "language_model.model.layers.6.post_attention_layernorm.weight": "model-00001-of-00004.safetensors",
530
+ "language_model.model.layers.6.self_attn.dense.bias": "model-00001-of-00004.safetensors",
531
+ "language_model.model.layers.6.self_attn.dense.weight": "model-00001-of-00004.safetensors",
532
+ "language_model.model.layers.6.self_attn.k_layernorm.bias": "model-00001-of-00004.safetensors",
533
+ "language_model.model.layers.6.self_attn.k_layernorm.weight": "model-00001-of-00004.safetensors",
534
+ "language_model.model.layers.6.self_attn.q_layernorm.bias": "model-00001-of-00004.safetensors",
535
+ "language_model.model.layers.6.self_attn.q_layernorm.weight": "model-00001-of-00004.safetensors",
536
+ "language_model.model.layers.6.self_attn.query_key_value.bias": "model-00001-of-00004.safetensors",
537
+ "language_model.model.layers.6.self_attn.query_key_value.weight": "model-00001-of-00004.safetensors",
538
+ "language_model.model.layers.7.input_layernorm.bias": "model-00002-of-00004.safetensors",
539
+ "language_model.model.layers.7.input_layernorm.weight": "model-00002-of-00004.safetensors",
540
+ "language_model.model.layers.7.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
541
+ "language_model.model.layers.7.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
542
+ "language_model.model.layers.7.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
543
+ "language_model.model.layers.7.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
544
+ "language_model.model.layers.7.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
545
+ "language_model.model.layers.7.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
546
+ "language_model.model.layers.7.self_attn.dense.bias": "model-00002-of-00004.safetensors",
547
+ "language_model.model.layers.7.self_attn.dense.weight": "model-00002-of-00004.safetensors",
548
+ "language_model.model.layers.7.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
549
+ "language_model.model.layers.7.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
550
+ "language_model.model.layers.7.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
551
+ "language_model.model.layers.7.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
552
+ "language_model.model.layers.7.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
553
+ "language_model.model.layers.7.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
554
+ "language_model.model.layers.8.input_layernorm.bias": "model-00002-of-00004.safetensors",
555
+ "language_model.model.layers.8.input_layernorm.weight": "model-00002-of-00004.safetensors",
556
+ "language_model.model.layers.8.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
557
+ "language_model.model.layers.8.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
558
+ "language_model.model.layers.8.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
559
+ "language_model.model.layers.8.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
560
+ "language_model.model.layers.8.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
561
+ "language_model.model.layers.8.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
562
+ "language_model.model.layers.8.self_attn.dense.bias": "model-00002-of-00004.safetensors",
563
+ "language_model.model.layers.8.self_attn.dense.weight": "model-00002-of-00004.safetensors",
564
+ "language_model.model.layers.8.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
565
+ "language_model.model.layers.8.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
566
+ "language_model.model.layers.8.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
567
+ "language_model.model.layers.8.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
568
+ "language_model.model.layers.8.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
569
+ "language_model.model.layers.8.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
570
+ "language_model.model.layers.9.input_layernorm.bias": "model-00002-of-00004.safetensors",
571
+ "language_model.model.layers.9.input_layernorm.weight": "model-00002-of-00004.safetensors",
572
+ "language_model.model.layers.9.mlp.dense_4h_to_h.bias": "model-00002-of-00004.safetensors",
573
+ "language_model.model.layers.9.mlp.dense_4h_to_h.weight": "model-00002-of-00004.safetensors",
574
+ "language_model.model.layers.9.mlp.dense_h_to_4h.bias": "model-00002-of-00004.safetensors",
575
+ "language_model.model.layers.9.mlp.dense_h_to_4h.weight": "model-00002-of-00004.safetensors",
576
+ "language_model.model.layers.9.post_attention_layernorm.bias": "model-00002-of-00004.safetensors",
577
+ "language_model.model.layers.9.post_attention_layernorm.weight": "model-00002-of-00004.safetensors",
578
+ "language_model.model.layers.9.self_attn.dense.bias": "model-00002-of-00004.safetensors",
579
+ "language_model.model.layers.9.self_attn.dense.weight": "model-00002-of-00004.safetensors",
580
+ "language_model.model.layers.9.self_attn.k_layernorm.bias": "model-00002-of-00004.safetensors",
581
+ "language_model.model.layers.9.self_attn.k_layernorm.weight": "model-00002-of-00004.safetensors",
582
+ "language_model.model.layers.9.self_attn.q_layernorm.bias": "model-00002-of-00004.safetensors",
583
+ "language_model.model.layers.9.self_attn.q_layernorm.weight": "model-00002-of-00004.safetensors",
584
+ "language_model.model.layers.9.self_attn.query_key_value.bias": "model-00002-of-00004.safetensors",
585
+ "language_model.model.layers.9.self_attn.query_key_value.weight": "model-00002-of-00004.safetensors",
586
+ "vision_embed_tokens.bias": "model-00004-of-00004.safetensors",
587
+ "vision_embed_tokens.weight": "model-00004-of-00004.safetensors"
588
+ }
589
+ }
scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9e98d18948e683c087d42711453160b73dd5a6a5458fdc5688b93864e60339c6
3
+ size 1064
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:58d473efcd31027e53845d5686b12624a0940efaeb1ab84778e5c7d435b6a50c
3
+ size 5880
zero_to_fp32.py ADDED
@@ -0,0 +1,587 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ #!/usr/bin/env python
2
+
3
+ # Copyright (c) Microsoft Corporation.
4
+ # SPDX-License-Identifier: Apache-2.0
5
+
6
+ # DeepSpeed Team
7
+
8
+ # This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
9
+ # copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
10
+ # the future. Once extracted, the weights don't require DeepSpeed and can be used in any
11
+ # application.
12
+ #
13
+ # example: python zero_to_fp32.py . pytorch_model.bin
14
+
15
+ import argparse
16
+ import torch
17
+ import glob
18
+ import math
19
+ import os
20
+ import re
21
+ from collections import OrderedDict
22
+ from dataclasses import dataclass
23
+
24
+ # while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
25
+ # DeepSpeed data structures it has to be available in the current python environment.
26
+ from deepspeed.utils import logger
27
+ from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
28
+ FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
29
+ FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
30
+
31
+
32
+ @dataclass
33
+ class zero_model_state:
34
+ buffers: dict()
35
+ param_shapes: dict()
36
+ shared_params: list
37
+ ds_version: int
38
+ frozen_param_shapes: dict()
39
+ frozen_param_fragments: dict()
40
+
41
+
42
+ debug = 0
43
+
44
+ # load to cpu
45
+ device = torch.device('cpu')
46
+
47
+
48
+ def atoi(text):
49
+ return int(text) if text.isdigit() else text
50
+
51
+
52
+ def natural_keys(text):
53
+ '''
54
+ alist.sort(key=natural_keys) sorts in human order
55
+ http://nedbatchelder.com/blog/200712/human_sorting.html
56
+ (See Toothy's implementation in the comments)
57
+ '''
58
+ return [atoi(c) for c in re.split(r'(\d+)', text)]
59
+
60
+
61
+ def get_model_state_file(checkpoint_dir, zero_stage):
62
+ if not os.path.isdir(checkpoint_dir):
63
+ raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
64
+
65
+ # there should be only one file
66
+ if zero_stage <= 2:
67
+ file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
68
+ elif zero_stage == 3:
69
+ file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
70
+
71
+ if not os.path.exists(file):
72
+ raise FileNotFoundError(f"can't find model states file at '{file}'")
73
+
74
+ return file
75
+
76
+
77
+ def get_checkpoint_files(checkpoint_dir, glob_pattern):
78
+ # XXX: need to test that this simple glob rule works for multi-node setup too
79
+ ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
80
+
81
+ if len(ckpt_files) == 0:
82
+ raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
83
+
84
+ return ckpt_files
85
+
86
+
87
+ def get_optim_files(checkpoint_dir):
88
+ return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
89
+
90
+
91
+ def get_model_state_files(checkpoint_dir):
92
+ return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
93
+
94
+
95
+ def parse_model_states(files):
96
+ zero_model_states = []
97
+ for file in files:
98
+ state_dict = torch.load(file, map_location=device)
99
+
100
+ if BUFFER_NAMES not in state_dict:
101
+ raise ValueError(f"{file} is not a model state checkpoint")
102
+ buffer_names = state_dict[BUFFER_NAMES]
103
+ if debug:
104
+ print("Found buffers:", buffer_names)
105
+
106
+ # recover just the buffers while restoring them to fp32 if they were saved in fp16
107
+ buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
108
+ param_shapes = state_dict[PARAM_SHAPES]
109
+
110
+ # collect parameters that are included in param_shapes
111
+ param_names = []
112
+ for s in param_shapes:
113
+ for name in s.keys():
114
+ param_names.append(name)
115
+
116
+ # update with frozen parameters
117
+ frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
118
+ if frozen_param_shapes is not None:
119
+ if debug:
120
+ print(f"Found frozen_param_shapes: {frozen_param_shapes}")
121
+ param_names += list(frozen_param_shapes.keys())
122
+
123
+ # handle shared params
124
+ shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
125
+
126
+ ds_version = state_dict.get(DS_VERSION, None)
127
+
128
+ frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
129
+
130
+ z_model_state = zero_model_state(buffers=buffers,
131
+ param_shapes=param_shapes,
132
+ shared_params=shared_params,
133
+ ds_version=ds_version,
134
+ frozen_param_shapes=frozen_param_shapes,
135
+ frozen_param_fragments=frozen_param_fragments)
136
+ zero_model_states.append(z_model_state)
137
+
138
+ return zero_model_states
139
+
140
+
141
+ def parse_optim_states(files, ds_checkpoint_dir):
142
+
143
+ total_files = len(files)
144
+ state_dicts = []
145
+ for f in files:
146
+ state_dict = torch.load(f, map_location=device)
147
+ # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
148
+ # and also handle the case where it was already removed by another helper script
149
+ state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
150
+ state_dicts.append(state_dict)
151
+
152
+ if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
153
+ raise ValueError(f"{files[0]} is not a zero checkpoint")
154
+ zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
155
+ world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
156
+
157
+ # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
158
+ # parameters can be different from data parallelism for non-expert parameters. So we can just
159
+ # use the max of the partition_count to get the dp world_size.
160
+
161
+ if type(world_size) is list:
162
+ world_size = max(world_size)
163
+
164
+ if world_size != total_files:
165
+ raise ValueError(
166
+ f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
167
+ "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
168
+ )
169
+
170
+ # the groups are named differently in each stage
171
+ if zero_stage <= 2:
172
+ fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
173
+ elif zero_stage == 3:
174
+ fp32_groups_key = FP32_FLAT_GROUPS
175
+ else:
176
+ raise ValueError(f"unknown zero stage {zero_stage}")
177
+
178
+ if zero_stage <= 2:
179
+ fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
180
+ elif zero_stage == 3:
181
+ # if there is more than one param group, there will be multiple flattened tensors - one
182
+ # flattened tensor per group - for simplicity merge them into a single tensor
183
+ #
184
+ # XXX: could make the script more memory efficient for when there are multiple groups - it
185
+ # will require matching the sub-lists of param_shapes for each param group flattened tensor
186
+
187
+ fp32_flat_groups = [
188
+ torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
189
+ ]
190
+
191
+ return zero_stage, world_size, fp32_flat_groups
192
+
193
+
194
+ def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
195
+ """
196
+ Returns fp32 state_dict reconstructed from ds checkpoint
197
+
198
+ Args:
199
+ - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
200
+
201
+ """
202
+ print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
203
+
204
+ optim_files = get_optim_files(ds_checkpoint_dir)
205
+ zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
206
+ print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
207
+
208
+ model_files = get_model_state_files(ds_checkpoint_dir)
209
+
210
+ zero_model_states = parse_model_states(model_files)
211
+ print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
212
+
213
+ if zero_stage <= 2:
214
+ return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
215
+ elif zero_stage == 3:
216
+ return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
217
+
218
+
219
+ def _zero2_merge_frozen_params(state_dict, zero_model_states):
220
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
221
+ return
222
+
223
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
224
+ frozen_param_fragments = zero_model_states[0].frozen_param_fragments
225
+
226
+ if debug:
227
+ num_elem = sum(s.numel() for s in frozen_param_shapes.values())
228
+ print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
229
+
230
+ wanted_params = len(frozen_param_shapes)
231
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
232
+ avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
233
+ print(f'Frozen params: Have {avail_numel} numels to process.')
234
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
235
+
236
+ total_params = 0
237
+ total_numel = 0
238
+ for name, shape in frozen_param_shapes.items():
239
+ total_params += 1
240
+ unpartitioned_numel = shape.numel()
241
+ total_numel += unpartitioned_numel
242
+
243
+ state_dict[name] = frozen_param_fragments[name]
244
+
245
+ if debug:
246
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
247
+
248
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
249
+
250
+
251
+ def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
252
+ param_shapes = zero_model_states[0].param_shapes
253
+
254
+ # Reconstruction protocol:
255
+ #
256
+ # XXX: document this
257
+
258
+ if debug:
259
+ for i in range(world_size):
260
+ for j in range(len(fp32_flat_groups[0])):
261
+ print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
262
+
263
+ # XXX: memory usage doubles here (zero2)
264
+ num_param_groups = len(fp32_flat_groups[0])
265
+ merged_single_partition_of_fp32_groups = []
266
+ for i in range(num_param_groups):
267
+ merged_partitions = [sd[i] for sd in fp32_flat_groups]
268
+ full_single_fp32_vector = torch.cat(merged_partitions, 0)
269
+ merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
270
+ avail_numel = sum(
271
+ [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
272
+
273
+ if debug:
274
+ wanted_params = sum([len(shapes) for shapes in param_shapes])
275
+ wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
276
+ # not asserting if there is a mismatch due to possible padding
277
+ print(f"Have {avail_numel} numels to process.")
278
+ print(f"Need {wanted_numel} numels in {wanted_params} params.")
279
+
280
+ # params
281
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
282
+ # out-of-core computing solution
283
+ total_numel = 0
284
+ total_params = 0
285
+ for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
286
+ offset = 0
287
+ avail_numel = full_single_fp32_vector.numel()
288
+ for name, shape in shapes.items():
289
+
290
+ unpartitioned_numel = shape.numel()
291
+ total_numel += unpartitioned_numel
292
+ total_params += 1
293
+
294
+ if debug:
295
+ print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
296
+ state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
297
+ offset += unpartitioned_numel
298
+
299
+ # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
300
+ # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
301
+ # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
302
+ # live optimizer object, so we are checking that the numbers are within the right range
303
+ align_to = 2 * world_size
304
+
305
+ def zero2_align(x):
306
+ return align_to * math.ceil(x / align_to)
307
+
308
+ if debug:
309
+ print(f"original offset={offset}, avail_numel={avail_numel}")
310
+
311
+ offset = zero2_align(offset)
312
+ avail_numel = zero2_align(avail_numel)
313
+
314
+ if debug:
315
+ print(f"aligned offset={offset}, avail_numel={avail_numel}")
316
+
317
+ # Sanity check
318
+ if offset != avail_numel:
319
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
320
+
321
+ print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
322
+
323
+
324
+ def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
325
+ state_dict = OrderedDict()
326
+
327
+ # buffers
328
+ buffers = zero_model_states[0].buffers
329
+ state_dict.update(buffers)
330
+ if debug:
331
+ print(f"added {len(buffers)} buffers")
332
+
333
+ _zero2_merge_frozen_params(state_dict, zero_model_states)
334
+
335
+ _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
336
+
337
+ # recover shared parameters
338
+ for pair in zero_model_states[0].shared_params:
339
+ if pair[1] in state_dict:
340
+ state_dict[pair[0]] = state_dict[pair[1]]
341
+
342
+ return state_dict
343
+
344
+
345
+ def zero3_partitioned_param_info(unpartitioned_numel, world_size):
346
+ remainder = unpartitioned_numel % world_size
347
+ padding_numel = (world_size - remainder) if remainder else 0
348
+ partitioned_numel = math.ceil(unpartitioned_numel / world_size)
349
+ return partitioned_numel, padding_numel
350
+
351
+
352
+ def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
353
+ if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
354
+ return
355
+
356
+ if debug:
357
+ for i in range(world_size):
358
+ num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
359
+ print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
360
+
361
+ frozen_param_shapes = zero_model_states[0].frozen_param_shapes
362
+ wanted_params = len(frozen_param_shapes)
363
+ wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
364
+ avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
365
+ print(f'Frozen params: Have {avail_numel} numels to process.')
366
+ print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
367
+
368
+ total_params = 0
369
+ total_numel = 0
370
+ for name, shape in zero_model_states[0].frozen_param_shapes.items():
371
+ total_params += 1
372
+ unpartitioned_numel = shape.numel()
373
+ total_numel += unpartitioned_numel
374
+
375
+ param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
376
+ state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
377
+
378
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
379
+
380
+ if debug:
381
+ print(
382
+ f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
383
+ )
384
+
385
+ print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
386
+
387
+
388
+ def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
389
+ param_shapes = zero_model_states[0].param_shapes
390
+ avail_numel = fp32_flat_groups[0].numel() * world_size
391
+ # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
392
+ # param, re-consolidating each param, while dealing with padding if any
393
+
394
+ # merge list of dicts, preserving order
395
+ param_shapes = {k: v for d in param_shapes for k, v in d.items()}
396
+
397
+ if debug:
398
+ for i in range(world_size):
399
+ print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
400
+
401
+ wanted_params = len(param_shapes)
402
+ wanted_numel = sum(shape.numel() for shape in param_shapes.values())
403
+ # not asserting if there is a mismatch due to possible padding
404
+ avail_numel = fp32_flat_groups[0].numel() * world_size
405
+ print(f"Trainable params: Have {avail_numel} numels to process.")
406
+ print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
407
+
408
+ # params
409
+ # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
410
+ # out-of-core computing solution
411
+ offset = 0
412
+ total_numel = 0
413
+ total_params = 0
414
+ for name, shape in param_shapes.items():
415
+
416
+ unpartitioned_numel = shape.numel()
417
+ total_numel += unpartitioned_numel
418
+ total_params += 1
419
+
420
+ partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
421
+
422
+ if debug:
423
+ print(
424
+ f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
425
+ )
426
+
427
+ # XXX: memory usage doubles here
428
+ state_dict[name] = torch.cat(
429
+ tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
430
+ 0).narrow(0, 0, unpartitioned_numel).view(shape)
431
+ offset += partitioned_numel
432
+
433
+ offset *= world_size
434
+
435
+ # Sanity check
436
+ if offset != avail_numel:
437
+ raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
438
+
439
+ print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
440
+
441
+
442
+ def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
443
+ state_dict = OrderedDict()
444
+
445
+ # buffers
446
+ buffers = zero_model_states[0].buffers
447
+ state_dict.update(buffers)
448
+ if debug:
449
+ print(f"added {len(buffers)} buffers")
450
+
451
+ _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
452
+
453
+ _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
454
+
455
+ # recover shared parameters
456
+ for pair in zero_model_states[0].shared_params:
457
+ if pair[1] in state_dict:
458
+ state_dict[pair[0]] = state_dict[pair[1]]
459
+
460
+ return state_dict
461
+
462
+
463
+ def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
464
+ """
465
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
466
+ ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
467
+ via a model hub.
468
+
469
+ Args:
470
+ - ``checkpoint_dir``: path to the desired checkpoint folder
471
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
472
+
473
+ Returns:
474
+ - pytorch ``state_dict``
475
+
476
+ Note: this approach may not work if your application doesn't have sufficient free CPU memory and
477
+ you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
478
+ the checkpoint.
479
+
480
+ A typical usage might be ::
481
+
482
+ from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
483
+ # do the training and checkpoint saving
484
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
485
+ model = model.cpu() # move to cpu
486
+ model.load_state_dict(state_dict)
487
+ # submit to model hub or save the model to share with others
488
+
489
+ In this example the ``model`` will no longer be usable in the deepspeed context of the same
490
+ application. i.e. you will need to re-initialize the deepspeed engine, since
491
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
492
+
493
+ If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
494
+
495
+ """
496
+ if tag is None:
497
+ latest_path = os.path.join(checkpoint_dir, 'latest')
498
+ if os.path.isfile(latest_path):
499
+ with open(latest_path, 'r') as fd:
500
+ tag = fd.read().strip()
501
+ else:
502
+ raise ValueError(f"Unable to find 'latest' file at {latest_path}")
503
+
504
+ ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
505
+
506
+ if not os.path.isdir(ds_checkpoint_dir):
507
+ raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
508
+
509
+ return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
510
+
511
+
512
+ def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
513
+ """
514
+ Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
515
+ loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
516
+
517
+ Args:
518
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
519
+ - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
520
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
521
+ """
522
+
523
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
524
+ print(f"Saving fp32 state dict to {output_file}")
525
+ torch.save(state_dict, output_file)
526
+
527
+
528
+ def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
529
+ """
530
+ 1. Put the provided model to cpu
531
+ 2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
532
+ 3. Load it into the provided model
533
+
534
+ Args:
535
+ - ``model``: the model object to update
536
+ - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
537
+ - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
538
+
539
+ Returns:
540
+ - ``model`: modified model
541
+
542
+ Make sure you have plenty of CPU memory available before you call this function. If you don't
543
+ have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
544
+ conveniently placed for you in the checkpoint folder.
545
+
546
+ A typical usage might be ::
547
+
548
+ from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
549
+ model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
550
+ # submit to model hub or save the model to share with others
551
+
552
+ Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
553
+ of the same application. i.e. you will need to re-initialize the deepspeed engine, since
554
+ ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
555
+
556
+ """
557
+ logger.info(f"Extracting fp32 weights")
558
+ state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
559
+
560
+ logger.info(f"Overwriting model with fp32 weights")
561
+ model = model.cpu()
562
+ model.load_state_dict(state_dict, strict=False)
563
+
564
+ return model
565
+
566
+
567
+ if __name__ == "__main__":
568
+
569
+ parser = argparse.ArgumentParser()
570
+ parser.add_argument("checkpoint_dir",
571
+ type=str,
572
+ help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
573
+ parser.add_argument(
574
+ "output_file",
575
+ type=str,
576
+ help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
577
+ parser.add_argument("-t",
578
+ "--tag",
579
+ type=str,
580
+ default=None,
581
+ help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
582
+ parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
583
+ args = parser.parse_args()
584
+
585
+ debug = args.debug
586
+
587
+ convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)