ehartford commited on
Commit
13c50e3
1 Parent(s): a9ae1d0

Upload folder using huggingface_hub

Browse files
added_tokens.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "</s>": 2,
3
+ "<s>": 1,
4
+ "<unk>": 0,
5
+ "<|im_end|>": 32000,
6
+ "<|im_start|>": 32001
7
+ }
config.json ADDED
@@ -0,0 +1,25 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "mistralai/Mistral-7B-v0.1",
3
+ "architectures": [
4
+ "MistralForCausalLM"
5
+ ],
6
+ "bos_token_id": 1,
7
+ "eos_token_id": 32000,
8
+ "hidden_act": "silu",
9
+ "hidden_size": 4096,
10
+ "initializer_range": 0.02,
11
+ "intermediate_size": 14336,
12
+ "max_position_embeddings": 32768,
13
+ "model_type": "mistral",
14
+ "num_attention_heads": 32,
15
+ "num_hidden_layers": 32,
16
+ "num_key_value_heads": 8,
17
+ "rms_norm_eps": 1e-05,
18
+ "rope_theta": 10000.0,
19
+ "sliding_window": 4096,
20
+ "tie_word_embeddings": false,
21
+ "torch_dtype": "bfloat16",
22
+ "transformers_version": "4.34.0.dev0",
23
+ "use_cache": true,
24
+ "vocab_size": 32002
25
+ }
configs/samantha-mistral-7b.yml ADDED
@@ -0,0 +1,70 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ base_model: mistralai/Mistral-7B-v0.1
2
+ base_model_config: mistralai/Mistral-7B-v0.1
3
+ model_type: MistralForCausalLM
4
+ tokenizer_type: LlamaTokenizer
5
+ is_mistral_derived_model: true
6
+
7
+ load_in_8bit: false
8
+ load_in_4bit: false
9
+ strict: false
10
+
11
+ datasets:
12
+ - path: ehartford/samantha-data
13
+ data_files: samantha-1.1.json
14
+ type: sharegpt
15
+ conversation: chatml
16
+
17
+ dataset_prepared_path: last_run_prepared
18
+ val_set_size: 0.01
19
+ output_dir: /workspace/samantha-1.2-mistral-7b
20
+
21
+ sequence_len: 8192
22
+ sample_packing: true
23
+ pad_to_sequence_len: true
24
+
25
+ wandb_project: samantha
26
+ wandb_entity:
27
+ wandb_watch:
28
+ wandb_run_id:
29
+ wandb_log_model:
30
+
31
+ gradient_accumulation_steps: 4
32
+ micro_batch_size: 6
33
+ num_epochs: 40
34
+ adam_beta2: 0.95
35
+ adam_epsilon: 0.00001
36
+ max_grad_norm: 1.0
37
+ lr_scheduler: cosine
38
+ learning_rate: 0.000006
39
+
40
+ train_on_inputs: false
41
+ group_by_length: false
42
+ bf16: true
43
+ fp16: false
44
+ tf32: false
45
+
46
+ gradient_checkpointing: true
47
+ early_stopping_patience:
48
+ resume_from_checkpoint:
49
+ local_rank:
50
+ logging_steps: 1
51
+ xformers_attention:
52
+ flash_attention: true
53
+
54
+ warmup_steps: 100
55
+ eval_steps: 0.05
56
+ eval_table_size:
57
+ eval_table_max_new_tokens:
58
+ save_steps:
59
+ debug:
60
+ deepspeed: deepspeed/zero2.json
61
+ weight_decay: 0.1
62
+ fsdp:
63
+ fsdp_config:
64
+ special_tokens:
65
+ bos_token: "<s>"
66
+ eos_token: "<|im_end|>"
67
+ unk_token: "<unk>"
68
+ tokens:
69
+ - "<|im_start|>"
70
+ - "<|im_end|>"
generation_config.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "transformers_version": "4.35.0.dev0"
6
+ }
latest ADDED
@@ -0,0 +1 @@
 
 
1
+ global_step187
pytorch_model-00001-of-00002.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68273d733eb811024f8dc9b3b480f91a42a5ca44f6f09bbd7843870144ac91fe
3
+ size 9943044428
pytorch_model-00002-of-00002.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:43cae66c14fad9e246e477d92ab15c9d9b76c2d48dd9c36b2f01c5c5eacd4df7
3
+ size 4540552031
pytorch_model.bin.index.json ADDED
@@ -0,0 +1,298 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "metadata": {
3
+ "total_size": 14483496960
4
+ },
5
+ "weight_map": {
6
+ "lm_head.weight": "pytorch_model-00002-of-00002.bin",
7
+ "model.embed_tokens.weight": "pytorch_model-00001-of-00002.bin",
8
+ "model.layers.0.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
9
+ "model.layers.0.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
10
+ "model.layers.0.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
11
+ "model.layers.0.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
12
+ "model.layers.0.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
13
+ "model.layers.0.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
14
+ "model.layers.0.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
15
+ "model.layers.0.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
16
+ "model.layers.0.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
17
+ "model.layers.1.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
18
+ "model.layers.1.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
19
+ "model.layers.1.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
20
+ "model.layers.1.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
21
+ "model.layers.1.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
22
+ "model.layers.1.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
23
+ "model.layers.1.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
24
+ "model.layers.1.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
25
+ "model.layers.1.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
26
+ "model.layers.10.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
27
+ "model.layers.10.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
28
+ "model.layers.10.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
29
+ "model.layers.10.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
30
+ "model.layers.10.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
31
+ "model.layers.10.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
32
+ "model.layers.10.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
33
+ "model.layers.10.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
34
+ "model.layers.10.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
35
+ "model.layers.11.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
36
+ "model.layers.11.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
37
+ "model.layers.11.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
38
+ "model.layers.11.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
39
+ "model.layers.11.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
40
+ "model.layers.11.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
41
+ "model.layers.11.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
42
+ "model.layers.11.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
43
+ "model.layers.11.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
44
+ "model.layers.12.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
45
+ "model.layers.12.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
46
+ "model.layers.12.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
47
+ "model.layers.12.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
48
+ "model.layers.12.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
49
+ "model.layers.12.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
50
+ "model.layers.12.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
51
+ "model.layers.12.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
52
+ "model.layers.12.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
53
+ "model.layers.13.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
54
+ "model.layers.13.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
55
+ "model.layers.13.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
56
+ "model.layers.13.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
57
+ "model.layers.13.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
58
+ "model.layers.13.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
59
+ "model.layers.13.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
60
+ "model.layers.13.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
61
+ "model.layers.13.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
62
+ "model.layers.14.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
63
+ "model.layers.14.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
64
+ "model.layers.14.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
65
+ "model.layers.14.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
66
+ "model.layers.14.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
67
+ "model.layers.14.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
68
+ "model.layers.14.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
69
+ "model.layers.14.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
70
+ "model.layers.14.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
71
+ "model.layers.15.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
72
+ "model.layers.15.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
73
+ "model.layers.15.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
74
+ "model.layers.15.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
75
+ "model.layers.15.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
76
+ "model.layers.15.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
77
+ "model.layers.15.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
78
+ "model.layers.15.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
79
+ "model.layers.15.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
80
+ "model.layers.16.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
81
+ "model.layers.16.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
82
+ "model.layers.16.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
83
+ "model.layers.16.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
84
+ "model.layers.16.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
85
+ "model.layers.16.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
86
+ "model.layers.16.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
87
+ "model.layers.16.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
88
+ "model.layers.16.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
89
+ "model.layers.17.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
90
+ "model.layers.17.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
91
+ "model.layers.17.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
92
+ "model.layers.17.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
93
+ "model.layers.17.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
94
+ "model.layers.17.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
95
+ "model.layers.17.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
96
+ "model.layers.17.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
97
+ "model.layers.17.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
98
+ "model.layers.18.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
99
+ "model.layers.18.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
100
+ "model.layers.18.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
101
+ "model.layers.18.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
102
+ "model.layers.18.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
103
+ "model.layers.18.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
104
+ "model.layers.18.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
105
+ "model.layers.18.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
106
+ "model.layers.18.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
107
+ "model.layers.19.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
108
+ "model.layers.19.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
109
+ "model.layers.19.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
110
+ "model.layers.19.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
111
+ "model.layers.19.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
112
+ "model.layers.19.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
113
+ "model.layers.19.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
114
+ "model.layers.19.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
115
+ "model.layers.19.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
116
+ "model.layers.2.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
117
+ "model.layers.2.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
118
+ "model.layers.2.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
119
+ "model.layers.2.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
120
+ "model.layers.2.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
121
+ "model.layers.2.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
122
+ "model.layers.2.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
123
+ "model.layers.2.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
124
+ "model.layers.2.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
125
+ "model.layers.20.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
126
+ "model.layers.20.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
127
+ "model.layers.20.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
128
+ "model.layers.20.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
129
+ "model.layers.20.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
130
+ "model.layers.20.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
131
+ "model.layers.20.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
132
+ "model.layers.20.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
133
+ "model.layers.20.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
134
+ "model.layers.21.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
135
+ "model.layers.21.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
136
+ "model.layers.21.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
137
+ "model.layers.21.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
138
+ "model.layers.21.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
139
+ "model.layers.21.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
140
+ "model.layers.21.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
141
+ "model.layers.21.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
142
+ "model.layers.21.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
143
+ "model.layers.22.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
144
+ "model.layers.22.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
145
+ "model.layers.22.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
146
+ "model.layers.22.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
147
+ "model.layers.22.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
148
+ "model.layers.22.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
149
+ "model.layers.22.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
150
+ "model.layers.22.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
151
+ "model.layers.22.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
152
+ "model.layers.23.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
153
+ "model.layers.23.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
154
+ "model.layers.23.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
155
+ "model.layers.23.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
156
+ "model.layers.23.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
157
+ "model.layers.23.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
158
+ "model.layers.23.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
159
+ "model.layers.23.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
160
+ "model.layers.23.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
161
+ "model.layers.24.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
162
+ "model.layers.24.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
163
+ "model.layers.24.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
164
+ "model.layers.24.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
165
+ "model.layers.24.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
166
+ "model.layers.24.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
167
+ "model.layers.24.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
168
+ "model.layers.24.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
169
+ "model.layers.24.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
170
+ "model.layers.25.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
171
+ "model.layers.25.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
172
+ "model.layers.25.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
173
+ "model.layers.25.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
174
+ "model.layers.25.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
175
+ "model.layers.25.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
176
+ "model.layers.25.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
177
+ "model.layers.25.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
178
+ "model.layers.25.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
179
+ "model.layers.26.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
180
+ "model.layers.26.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
181
+ "model.layers.26.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
182
+ "model.layers.26.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
183
+ "model.layers.26.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
184
+ "model.layers.26.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
185
+ "model.layers.26.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
186
+ "model.layers.26.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
187
+ "model.layers.26.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
188
+ "model.layers.27.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
189
+ "model.layers.27.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
190
+ "model.layers.27.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
191
+ "model.layers.27.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
192
+ "model.layers.27.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
193
+ "model.layers.27.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
194
+ "model.layers.27.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
195
+ "model.layers.27.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
196
+ "model.layers.27.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
197
+ "model.layers.28.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
198
+ "model.layers.28.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
199
+ "model.layers.28.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
200
+ "model.layers.28.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
201
+ "model.layers.28.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
202
+ "model.layers.28.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
203
+ "model.layers.28.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
204
+ "model.layers.28.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
205
+ "model.layers.28.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
206
+ "model.layers.29.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
207
+ "model.layers.29.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
208
+ "model.layers.29.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
209
+ "model.layers.29.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
210
+ "model.layers.29.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
211
+ "model.layers.29.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
212
+ "model.layers.29.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
213
+ "model.layers.29.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
214
+ "model.layers.29.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
215
+ "model.layers.3.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
216
+ "model.layers.3.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
217
+ "model.layers.3.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
218
+ "model.layers.3.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
219
+ "model.layers.3.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
220
+ "model.layers.3.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
221
+ "model.layers.3.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
222
+ "model.layers.3.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
223
+ "model.layers.3.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
224
+ "model.layers.30.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
225
+ "model.layers.30.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
226
+ "model.layers.30.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
227
+ "model.layers.30.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
228
+ "model.layers.30.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
229
+ "model.layers.30.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
230
+ "model.layers.30.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
231
+ "model.layers.30.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
232
+ "model.layers.30.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
233
+ "model.layers.31.input_layernorm.weight": "pytorch_model-00002-of-00002.bin",
234
+ "model.layers.31.mlp.down_proj.weight": "pytorch_model-00002-of-00002.bin",
235
+ "model.layers.31.mlp.gate_proj.weight": "pytorch_model-00002-of-00002.bin",
236
+ "model.layers.31.mlp.up_proj.weight": "pytorch_model-00002-of-00002.bin",
237
+ "model.layers.31.post_attention_layernorm.weight": "pytorch_model-00002-of-00002.bin",
238
+ "model.layers.31.self_attn.k_proj.weight": "pytorch_model-00002-of-00002.bin",
239
+ "model.layers.31.self_attn.o_proj.weight": "pytorch_model-00002-of-00002.bin",
240
+ "model.layers.31.self_attn.q_proj.weight": "pytorch_model-00002-of-00002.bin",
241
+ "model.layers.31.self_attn.v_proj.weight": "pytorch_model-00002-of-00002.bin",
242
+ "model.layers.4.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
243
+ "model.layers.4.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
244
+ "model.layers.4.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
245
+ "model.layers.4.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
246
+ "model.layers.4.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
247
+ "model.layers.4.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
248
+ "model.layers.4.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
249
+ "model.layers.4.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
250
+ "model.layers.4.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
251
+ "model.layers.5.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
252
+ "model.layers.5.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
253
+ "model.layers.5.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
254
+ "model.layers.5.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
255
+ "model.layers.5.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
256
+ "model.layers.5.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
257
+ "model.layers.5.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
258
+ "model.layers.5.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
259
+ "model.layers.5.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
260
+ "model.layers.6.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
261
+ "model.layers.6.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
262
+ "model.layers.6.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
263
+ "model.layers.6.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
264
+ "model.layers.6.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
265
+ "model.layers.6.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
266
+ "model.layers.6.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
267
+ "model.layers.6.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
268
+ "model.layers.6.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
269
+ "model.layers.7.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
270
+ "model.layers.7.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
271
+ "model.layers.7.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
272
+ "model.layers.7.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
273
+ "model.layers.7.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
274
+ "model.layers.7.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
275
+ "model.layers.7.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
276
+ "model.layers.7.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
277
+ "model.layers.7.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
278
+ "model.layers.8.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
279
+ "model.layers.8.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
280
+ "model.layers.8.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
281
+ "model.layers.8.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
282
+ "model.layers.8.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
283
+ "model.layers.8.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
284
+ "model.layers.8.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
285
+ "model.layers.8.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
286
+ "model.layers.8.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
287
+ "model.layers.9.input_layernorm.weight": "pytorch_model-00001-of-00002.bin",
288
+ "model.layers.9.mlp.down_proj.weight": "pytorch_model-00001-of-00002.bin",
289
+ "model.layers.9.mlp.gate_proj.weight": "pytorch_model-00001-of-00002.bin",
290
+ "model.layers.9.mlp.up_proj.weight": "pytorch_model-00001-of-00002.bin",
291
+ "model.layers.9.post_attention_layernorm.weight": "pytorch_model-00001-of-00002.bin",
292
+ "model.layers.9.self_attn.k_proj.weight": "pytorch_model-00001-of-00002.bin",
293
+ "model.layers.9.self_attn.o_proj.weight": "pytorch_model-00001-of-00002.bin",
294
+ "model.layers.9.self_attn.q_proj.weight": "pytorch_model-00001-of-00002.bin",
295
+ "model.layers.9.self_attn.v_proj.weight": "pytorch_model-00001-of-00002.bin",
296
+ "model.norm.weight": "pytorch_model-00002-of-00002.bin"
297
+ }
298
+ }
special_tokens_map.json ADDED
@@ -0,0 +1,6 @@
 
 
 
 
 
 
 
1
+ {
2
+ "bos_token": "<s>",
3
+ "eos_token": "<|im_end|>",
4
+ "pad_token": "</s>",
5
+ "unk_token": "<unk>"
6
+ }
tokenizer.model ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
3
+ size 493443
tokenizer_config.json ADDED
@@ -0,0 +1,61 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_bos_token": true,
3
+ "add_eos_token": false,
4
+ "added_tokens_decoder": {
5
+ "0": {
6
+ "content": "<unk>",
7
+ "lstrip": true,
8
+ "normalized": false,
9
+ "rstrip": true,
10
+ "single_word": false,
11
+ "special": true
12
+ },
13
+ "1": {
14
+ "content": "<s>",
15
+ "lstrip": true,
16
+ "normalized": false,
17
+ "rstrip": true,
18
+ "single_word": false,
19
+ "special": true
20
+ },
21
+ "2": {
22
+ "content": "</s>",
23
+ "lstrip": false,
24
+ "normalized": false,
25
+ "rstrip": false,
26
+ "single_word": false,
27
+ "special": true
28
+ },
29
+ "32000": {
30
+ "content": "<|im_end|>",
31
+ "lstrip": false,
32
+ "normalized": false,
33
+ "rstrip": false,
34
+ "single_word": false,
35
+ "special": true
36
+ },
37
+ "32001": {
38
+ "content": "<|im_start|>",
39
+ "lstrip": true,
40
+ "normalized": false,
41
+ "rstrip": true,
42
+ "single_word": false,
43
+ "special": true
44
+ }
45
+ },
46
+ "additional_special_tokens": [],
47
+ "bos_token": "<s>",
48
+ "chat_template": "{% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% for message in messages %}{{'<|im_start|>' + message['role'] + '\n' + message['content'] + '<|im_end|>' + '\n'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant\n' }}{% endif %}",
49
+ "clean_up_tokenization_spaces": false,
50
+ "eos_token": "<|im_end|>",
51
+ "legacy": true,
52
+ "model_max_length": 1000000000000000019884624838656,
53
+ "pad_token": null,
54
+ "sp_model_kwargs": {},
55
+ "spaces_between_special_tokens": false,
56
+ "tokenizer_class": "LlamaTokenizer",
57
+ "trust_remote_code": false,
58
+ "unk_token": "<unk>",
59
+ "use_default_system_prompt": true,
60
+ "use_fast": true
61
+ }
trainer_state.json ADDED
@@ -0,0 +1,1293 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 34.0,
5
+ "eval_steps": 10,
6
+ "global_step": 187,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.18,
13
+ "learning_rate": 0.0,
14
+ "loss": 2.0388,
15
+ "step": 1
16
+ },
17
+ {
18
+ "epoch": 0.18,
19
+ "eval_loss": 14.199334144592285,
20
+ "eval_runtime": 1.4513,
21
+ "eval_samples_per_second": 45.477,
22
+ "eval_steps_per_second": 2.067,
23
+ "step": 1
24
+ },
25
+ {
26
+ "epoch": 0.36,
27
+ "learning_rate": 6.000000000000001e-08,
28
+ "loss": 11.2905,
29
+ "step": 2
30
+ },
31
+ {
32
+ "epoch": 0.55,
33
+ "learning_rate": 1.2000000000000002e-07,
34
+ "loss": 11.3574,
35
+ "step": 3
36
+ },
37
+ {
38
+ "epoch": 0.73,
39
+ "learning_rate": 1.8e-07,
40
+ "loss": 10.3132,
41
+ "step": 4
42
+ },
43
+ {
44
+ "epoch": 0.91,
45
+ "learning_rate": 2.4000000000000003e-07,
46
+ "loss": 8.2949,
47
+ "step": 5
48
+ },
49
+ {
50
+ "epoch": 1.09,
51
+ "learning_rate": 3.0000000000000004e-07,
52
+ "loss": 5.3189,
53
+ "step": 6
54
+ },
55
+ {
56
+ "epoch": 1.27,
57
+ "learning_rate": 3.6e-07,
58
+ "loss": 2.2121,
59
+ "step": 7
60
+ },
61
+ {
62
+ "epoch": 1.45,
63
+ "learning_rate": 4.2000000000000006e-07,
64
+ "loss": 2.0155,
65
+ "step": 8
66
+ },
67
+ {
68
+ "epoch": 1.64,
69
+ "learning_rate": 4.800000000000001e-07,
70
+ "loss": 1.9936,
71
+ "step": 9
72
+ },
73
+ {
74
+ "epoch": 1.82,
75
+ "learning_rate": 5.4e-07,
76
+ "loss": 1.8803,
77
+ "step": 10
78
+ },
79
+ {
80
+ "epoch": 1.82,
81
+ "eval_loss": 2.1710054874420166,
82
+ "eval_runtime": 1.463,
83
+ "eval_samples_per_second": 45.113,
84
+ "eval_steps_per_second": 2.051,
85
+ "step": 10
86
+ },
87
+ {
88
+ "epoch": 2.0,
89
+ "learning_rate": 6.000000000000001e-07,
90
+ "loss": 1.8794,
91
+ "step": 11
92
+ },
93
+ {
94
+ "epoch": 2.18,
95
+ "learning_rate": 6.6e-07,
96
+ "loss": 1.884,
97
+ "step": 12
98
+ },
99
+ {
100
+ "epoch": 2.36,
101
+ "learning_rate": 7.2e-07,
102
+ "loss": 1.8201,
103
+ "step": 13
104
+ },
105
+ {
106
+ "epoch": 2.55,
107
+ "learning_rate": 7.8e-07,
108
+ "loss": 1.8356,
109
+ "step": 14
110
+ },
111
+ {
112
+ "epoch": 2.73,
113
+ "learning_rate": 8.400000000000001e-07,
114
+ "loss": 1.7696,
115
+ "step": 15
116
+ },
117
+ {
118
+ "epoch": 2.91,
119
+ "learning_rate": 9e-07,
120
+ "loss": 1.6433,
121
+ "step": 16
122
+ },
123
+ {
124
+ "epoch": 3.09,
125
+ "learning_rate": 9.600000000000001e-07,
126
+ "loss": 1.5564,
127
+ "step": 17
128
+ },
129
+ {
130
+ "epoch": 3.27,
131
+ "learning_rate": 1.0200000000000002e-06,
132
+ "loss": 1.4976,
133
+ "step": 18
134
+ },
135
+ {
136
+ "epoch": 3.45,
137
+ "learning_rate": 1.08e-06,
138
+ "loss": 1.4484,
139
+ "step": 19
140
+ },
141
+ {
142
+ "epoch": 3.64,
143
+ "learning_rate": 1.14e-06,
144
+ "loss": 1.4277,
145
+ "step": 20
146
+ },
147
+ {
148
+ "epoch": 3.64,
149
+ "eval_loss": 1.5980714559555054,
150
+ "eval_runtime": 1.4588,
151
+ "eval_samples_per_second": 45.243,
152
+ "eval_steps_per_second": 2.056,
153
+ "step": 20
154
+ },
155
+ {
156
+ "epoch": 3.82,
157
+ "learning_rate": 1.2000000000000002e-06,
158
+ "loss": 1.4004,
159
+ "step": 21
160
+ },
161
+ {
162
+ "epoch": 4.0,
163
+ "learning_rate": 1.26e-06,
164
+ "loss": 1.3963,
165
+ "step": 22
166
+ },
167
+ {
168
+ "epoch": 4.18,
169
+ "learning_rate": 1.32e-06,
170
+ "loss": 1.3915,
171
+ "step": 23
172
+ },
173
+ {
174
+ "epoch": 4.36,
175
+ "learning_rate": 1.3800000000000001e-06,
176
+ "loss": 1.3803,
177
+ "step": 24
178
+ },
179
+ {
180
+ "epoch": 4.55,
181
+ "learning_rate": 1.44e-06,
182
+ "loss": 1.3711,
183
+ "step": 25
184
+ },
185
+ {
186
+ "epoch": 4.73,
187
+ "learning_rate": 1.5e-06,
188
+ "loss": 1.3596,
189
+ "step": 26
190
+ },
191
+ {
192
+ "epoch": 4.91,
193
+ "learning_rate": 1.56e-06,
194
+ "loss": 1.3376,
195
+ "step": 27
196
+ },
197
+ {
198
+ "epoch": 5.09,
199
+ "learning_rate": 1.6200000000000002e-06,
200
+ "loss": 1.3241,
201
+ "step": 28
202
+ },
203
+ {
204
+ "epoch": 5.27,
205
+ "learning_rate": 1.6800000000000002e-06,
206
+ "loss": 1.2975,
207
+ "step": 29
208
+ },
209
+ {
210
+ "epoch": 5.45,
211
+ "learning_rate": 1.7399999999999999e-06,
212
+ "loss": 1.2847,
213
+ "step": 30
214
+ },
215
+ {
216
+ "epoch": 5.45,
217
+ "eval_loss": 1.4864555597305298,
218
+ "eval_runtime": 1.4609,
219
+ "eval_samples_per_second": 45.176,
220
+ "eval_steps_per_second": 2.053,
221
+ "step": 30
222
+ },
223
+ {
224
+ "epoch": 5.64,
225
+ "learning_rate": 1.8e-06,
226
+ "loss": 1.2798,
227
+ "step": 31
228
+ },
229
+ {
230
+ "epoch": 5.82,
231
+ "learning_rate": 1.86e-06,
232
+ "loss": 1.2616,
233
+ "step": 32
234
+ },
235
+ {
236
+ "epoch": 6.0,
237
+ "learning_rate": 1.9200000000000003e-06,
238
+ "loss": 1.252,
239
+ "step": 33
240
+ },
241
+ {
242
+ "epoch": 6.18,
243
+ "learning_rate": 1.98e-06,
244
+ "loss": 1.2398,
245
+ "step": 34
246
+ },
247
+ {
248
+ "epoch": 6.36,
249
+ "learning_rate": 2.0400000000000004e-06,
250
+ "loss": 1.2229,
251
+ "step": 35
252
+ },
253
+ {
254
+ "epoch": 6.55,
255
+ "learning_rate": 2.1e-06,
256
+ "loss": 1.2221,
257
+ "step": 36
258
+ },
259
+ {
260
+ "epoch": 6.73,
261
+ "learning_rate": 2.16e-06,
262
+ "loss": 1.2046,
263
+ "step": 37
264
+ },
265
+ {
266
+ "epoch": 6.91,
267
+ "learning_rate": 2.22e-06,
268
+ "loss": 1.1908,
269
+ "step": 38
270
+ },
271
+ {
272
+ "epoch": 7.09,
273
+ "learning_rate": 2.28e-06,
274
+ "loss": 1.1916,
275
+ "step": 39
276
+ },
277
+ {
278
+ "epoch": 7.27,
279
+ "learning_rate": 2.34e-06,
280
+ "loss": 1.1733,
281
+ "step": 40
282
+ },
283
+ {
284
+ "epoch": 7.27,
285
+ "eval_loss": 1.37030827999115,
286
+ "eval_runtime": 1.4628,
287
+ "eval_samples_per_second": 45.118,
288
+ "eval_steps_per_second": 2.051,
289
+ "step": 40
290
+ },
291
+ {
292
+ "epoch": 7.45,
293
+ "learning_rate": 2.4000000000000003e-06,
294
+ "loss": 1.1708,
295
+ "step": 41
296
+ },
297
+ {
298
+ "epoch": 7.64,
299
+ "learning_rate": 2.4599999999999997e-06,
300
+ "loss": 1.1727,
301
+ "step": 42
302
+ },
303
+ {
304
+ "epoch": 7.82,
305
+ "learning_rate": 2.52e-06,
306
+ "loss": 1.1615,
307
+ "step": 43
308
+ },
309
+ {
310
+ "epoch": 8.0,
311
+ "learning_rate": 2.58e-06,
312
+ "loss": 1.1529,
313
+ "step": 44
314
+ },
315
+ {
316
+ "epoch": 8.18,
317
+ "learning_rate": 2.64e-06,
318
+ "loss": 1.1428,
319
+ "step": 45
320
+ },
321
+ {
322
+ "epoch": 8.36,
323
+ "learning_rate": 2.7e-06,
324
+ "loss": 1.1398,
325
+ "step": 46
326
+ },
327
+ {
328
+ "epoch": 8.55,
329
+ "learning_rate": 2.7600000000000003e-06,
330
+ "loss": 1.1392,
331
+ "step": 47
332
+ },
333
+ {
334
+ "epoch": 8.73,
335
+ "learning_rate": 2.82e-06,
336
+ "loss": 1.121,
337
+ "step": 48
338
+ },
339
+ {
340
+ "epoch": 8.91,
341
+ "learning_rate": 2.88e-06,
342
+ "loss": 1.1239,
343
+ "step": 49
344
+ },
345
+ {
346
+ "epoch": 9.09,
347
+ "learning_rate": 2.9400000000000002e-06,
348
+ "loss": 1.1131,
349
+ "step": 50
350
+ },
351
+ {
352
+ "epoch": 9.09,
353
+ "eval_loss": 1.3062629699707031,
354
+ "eval_runtime": 1.4506,
355
+ "eval_samples_per_second": 45.499,
356
+ "eval_steps_per_second": 2.068,
357
+ "step": 50
358
+ },
359
+ {
360
+ "epoch": 9.27,
361
+ "learning_rate": 3e-06,
362
+ "loss": 1.1047,
363
+ "step": 51
364
+ },
365
+ {
366
+ "epoch": 9.45,
367
+ "learning_rate": 3.06e-06,
368
+ "loss": 1.1012,
369
+ "step": 52
370
+ },
371
+ {
372
+ "epoch": 9.64,
373
+ "learning_rate": 3.12e-06,
374
+ "loss": 1.0987,
375
+ "step": 53
376
+ },
377
+ {
378
+ "epoch": 9.82,
379
+ "learning_rate": 3.18e-06,
380
+ "loss": 1.1012,
381
+ "step": 54
382
+ },
383
+ {
384
+ "epoch": 10.0,
385
+ "learning_rate": 3.2400000000000003e-06,
386
+ "loss": 1.093,
387
+ "step": 55
388
+ },
389
+ {
390
+ "epoch": 10.18,
391
+ "learning_rate": 3.3e-06,
392
+ "loss": 1.0876,
393
+ "step": 56
394
+ },
395
+ {
396
+ "epoch": 10.36,
397
+ "learning_rate": 3.3600000000000004e-06,
398
+ "loss": 1.0813,
399
+ "step": 57
400
+ },
401
+ {
402
+ "epoch": 10.55,
403
+ "learning_rate": 3.42e-06,
404
+ "loss": 1.0798,
405
+ "step": 58
406
+ },
407
+ {
408
+ "epoch": 10.73,
409
+ "learning_rate": 3.4799999999999997e-06,
410
+ "loss": 1.0673,
411
+ "step": 59
412
+ },
413
+ {
414
+ "epoch": 10.91,
415
+ "learning_rate": 3.54e-06,
416
+ "loss": 1.0616,
417
+ "step": 60
418
+ },
419
+ {
420
+ "epoch": 10.91,
421
+ "eval_loss": 1.2495708465576172,
422
+ "eval_runtime": 1.4605,
423
+ "eval_samples_per_second": 45.19,
424
+ "eval_steps_per_second": 2.054,
425
+ "step": 60
426
+ },
427
+ {
428
+ "epoch": 11.09,
429
+ "learning_rate": 3.6e-06,
430
+ "loss": 1.0639,
431
+ "step": 61
432
+ },
433
+ {
434
+ "epoch": 11.27,
435
+ "learning_rate": 3.66e-06,
436
+ "loss": 1.0538,
437
+ "step": 62
438
+ },
439
+ {
440
+ "epoch": 11.45,
441
+ "learning_rate": 3.72e-06,
442
+ "loss": 1.0567,
443
+ "step": 63
444
+ },
445
+ {
446
+ "epoch": 11.64,
447
+ "learning_rate": 3.7800000000000002e-06,
448
+ "loss": 1.0465,
449
+ "step": 64
450
+ },
451
+ {
452
+ "epoch": 11.82,
453
+ "learning_rate": 3.8400000000000005e-06,
454
+ "loss": 1.0485,
455
+ "step": 65
456
+ },
457
+ {
458
+ "epoch": 12.0,
459
+ "learning_rate": 3.9e-06,
460
+ "loss": 1.0494,
461
+ "step": 66
462
+ },
463
+ {
464
+ "epoch": 12.18,
465
+ "learning_rate": 3.96e-06,
466
+ "loss": 1.0359,
467
+ "step": 67
468
+ },
469
+ {
470
+ "epoch": 12.36,
471
+ "learning_rate": 4.0200000000000005e-06,
472
+ "loss": 1.0336,
473
+ "step": 68
474
+ },
475
+ {
476
+ "epoch": 12.55,
477
+ "learning_rate": 4.080000000000001e-06,
478
+ "loss": 1.0317,
479
+ "step": 69
480
+ },
481
+ {
482
+ "epoch": 12.73,
483
+ "learning_rate": 4.14e-06,
484
+ "loss": 1.0302,
485
+ "step": 70
486
+ },
487
+ {
488
+ "epoch": 12.73,
489
+ "eval_loss": 1.2099440097808838,
490
+ "eval_runtime": 1.4611,
491
+ "eval_samples_per_second": 45.172,
492
+ "eval_steps_per_second": 2.053,
493
+ "step": 70
494
+ },
495
+ {
496
+ "epoch": 12.91,
497
+ "learning_rate": 4.2e-06,
498
+ "loss": 1.0249,
499
+ "step": 71
500
+ },
501
+ {
502
+ "epoch": 13.09,
503
+ "learning_rate": 4.26e-06,
504
+ "loss": 1.022,
505
+ "step": 72
506
+ },
507
+ {
508
+ "epoch": 13.27,
509
+ "learning_rate": 4.32e-06,
510
+ "loss": 1.0138,
511
+ "step": 73
512
+ },
513
+ {
514
+ "epoch": 13.45,
515
+ "learning_rate": 4.38e-06,
516
+ "loss": 1.0136,
517
+ "step": 74
518
+ },
519
+ {
520
+ "epoch": 13.64,
521
+ "learning_rate": 4.44e-06,
522
+ "loss": 1.0169,
523
+ "step": 75
524
+ },
525
+ {
526
+ "epoch": 13.82,
527
+ "learning_rate": 4.5e-06,
528
+ "loss": 1.0041,
529
+ "step": 76
530
+ },
531
+ {
532
+ "epoch": 14.0,
533
+ "learning_rate": 4.56e-06,
534
+ "loss": 0.9956,
535
+ "step": 77
536
+ },
537
+ {
538
+ "epoch": 14.18,
539
+ "learning_rate": 4.62e-06,
540
+ "loss": 0.9941,
541
+ "step": 78
542
+ },
543
+ {
544
+ "epoch": 14.36,
545
+ "learning_rate": 4.68e-06,
546
+ "loss": 0.9963,
547
+ "step": 79
548
+ },
549
+ {
550
+ "epoch": 14.55,
551
+ "learning_rate": 4.74e-06,
552
+ "loss": 0.9866,
553
+ "step": 80
554
+ },
555
+ {
556
+ "epoch": 14.55,
557
+ "eval_loss": 1.1811805963516235,
558
+ "eval_runtime": 1.4593,
559
+ "eval_samples_per_second": 45.228,
560
+ "eval_steps_per_second": 2.056,
561
+ "step": 80
562
+ },
563
+ {
564
+ "epoch": 14.73,
565
+ "learning_rate": 4.800000000000001e-06,
566
+ "loss": 0.9904,
567
+ "step": 81
568
+ },
569
+ {
570
+ "epoch": 14.91,
571
+ "learning_rate": 4.86e-06,
572
+ "loss": 0.9849,
573
+ "step": 82
574
+ },
575
+ {
576
+ "epoch": 15.09,
577
+ "learning_rate": 4.9199999999999995e-06,
578
+ "loss": 0.9784,
579
+ "step": 83
580
+ },
581
+ {
582
+ "epoch": 15.27,
583
+ "learning_rate": 4.98e-06,
584
+ "loss": 0.9801,
585
+ "step": 84
586
+ },
587
+ {
588
+ "epoch": 15.45,
589
+ "learning_rate": 5.04e-06,
590
+ "loss": 0.9749,
591
+ "step": 85
592
+ },
593
+ {
594
+ "epoch": 15.64,
595
+ "learning_rate": 5.1e-06,
596
+ "loss": 0.9667,
597
+ "step": 86
598
+ },
599
+ {
600
+ "epoch": 15.82,
601
+ "learning_rate": 5.16e-06,
602
+ "loss": 0.9675,
603
+ "step": 87
604
+ },
605
+ {
606
+ "epoch": 16.0,
607
+ "learning_rate": 5.22e-06,
608
+ "loss": 0.9666,
609
+ "step": 88
610
+ },
611
+ {
612
+ "epoch": 16.18,
613
+ "learning_rate": 5.28e-06,
614
+ "loss": 0.9582,
615
+ "step": 89
616
+ },
617
+ {
618
+ "epoch": 16.36,
619
+ "learning_rate": 5.3400000000000005e-06,
620
+ "loss": 0.9594,
621
+ "step": 90
622
+ },
623
+ {
624
+ "epoch": 16.36,
625
+ "eval_loss": 1.1599087715148926,
626
+ "eval_runtime": 1.4615,
627
+ "eval_samples_per_second": 45.16,
628
+ "eval_steps_per_second": 2.053,
629
+ "step": 90
630
+ },
631
+ {
632
+ "epoch": 16.55,
633
+ "learning_rate": 5.4e-06,
634
+ "loss": 0.9462,
635
+ "step": 91
636
+ },
637
+ {
638
+ "epoch": 16.73,
639
+ "learning_rate": 5.46e-06,
640
+ "loss": 0.952,
641
+ "step": 92
642
+ },
643
+ {
644
+ "epoch": 16.91,
645
+ "learning_rate": 5.5200000000000005e-06,
646
+ "loss": 0.9507,
647
+ "step": 93
648
+ },
649
+ {
650
+ "epoch": 17.09,
651
+ "learning_rate": 5.580000000000001e-06,
652
+ "loss": 0.9535,
653
+ "step": 94
654
+ },
655
+ {
656
+ "epoch": 17.27,
657
+ "learning_rate": 5.64e-06,
658
+ "loss": 0.9436,
659
+ "step": 95
660
+ },
661
+ {
662
+ "epoch": 17.45,
663
+ "learning_rate": 5.7e-06,
664
+ "loss": 0.9338,
665
+ "step": 96
666
+ },
667
+ {
668
+ "epoch": 17.64,
669
+ "learning_rate": 5.76e-06,
670
+ "loss": 0.9355,
671
+ "step": 97
672
+ },
673
+ {
674
+ "epoch": 17.82,
675
+ "learning_rate": 5.82e-06,
676
+ "loss": 0.9355,
677
+ "step": 98
678
+ },
679
+ {
680
+ "epoch": 18.0,
681
+ "learning_rate": 5.8800000000000005e-06,
682
+ "loss": 0.9343,
683
+ "step": 99
684
+ },
685
+ {
686
+ "epoch": 18.18,
687
+ "learning_rate": 5.94e-06,
688
+ "loss": 0.9161,
689
+ "step": 100
690
+ },
691
+ {
692
+ "epoch": 18.18,
693
+ "eval_loss": 1.1503610610961914,
694
+ "eval_runtime": 1.4537,
695
+ "eval_samples_per_second": 45.4,
696
+ "eval_steps_per_second": 2.064,
697
+ "step": 100
698
+ },
699
+ {
700
+ "epoch": 18.36,
701
+ "learning_rate": 6e-06,
702
+ "loss": 0.9156,
703
+ "step": 101
704
+ },
705
+ {
706
+ "epoch": 18.55,
707
+ "learning_rate": 5.94e-06,
708
+ "loss": 0.9185,
709
+ "step": 102
710
+ },
711
+ {
712
+ "epoch": 18.73,
713
+ "learning_rate": 5.8800000000000005e-06,
714
+ "loss": 0.9155,
715
+ "step": 103
716
+ },
717
+ {
718
+ "epoch": 18.91,
719
+ "learning_rate": 5.82e-06,
720
+ "loss": 0.9114,
721
+ "step": 104
722
+ },
723
+ {
724
+ "epoch": 19.09,
725
+ "learning_rate": 5.76e-06,
726
+ "loss": 0.8976,
727
+ "step": 105
728
+ },
729
+ {
730
+ "epoch": 19.27,
731
+ "learning_rate": 5.7e-06,
732
+ "loss": 0.8985,
733
+ "step": 106
734
+ },
735
+ {
736
+ "epoch": 19.45,
737
+ "learning_rate": 5.64e-06,
738
+ "loss": 0.8972,
739
+ "step": 107
740
+ },
741
+ {
742
+ "epoch": 19.64,
743
+ "learning_rate": 5.580000000000001e-06,
744
+ "loss": 0.8961,
745
+ "step": 108
746
+ },
747
+ {
748
+ "epoch": 19.82,
749
+ "learning_rate": 5.5200000000000005e-06,
750
+ "loss": 0.8948,
751
+ "step": 109
752
+ },
753
+ {
754
+ "epoch": 20.0,
755
+ "learning_rate": 5.46e-06,
756
+ "loss": 0.8823,
757
+ "step": 110
758
+ },
759
+ {
760
+ "epoch": 20.0,
761
+ "eval_loss": 1.13896906375885,
762
+ "eval_runtime": 1.4633,
763
+ "eval_samples_per_second": 45.103,
764
+ "eval_steps_per_second": 2.05,
765
+ "step": 110
766
+ },
767
+ {
768
+ "epoch": 20.18,
769
+ "learning_rate": 5.4e-06,
770
+ "loss": 0.8689,
771
+ "step": 111
772
+ },
773
+ {
774
+ "epoch": 20.36,
775
+ "learning_rate": 5.3400000000000005e-06,
776
+ "loss": 0.8832,
777
+ "step": 112
778
+ },
779
+ {
780
+ "epoch": 20.55,
781
+ "learning_rate": 5.28e-06,
782
+ "loss": 0.875,
783
+ "step": 113
784
+ },
785
+ {
786
+ "epoch": 20.73,
787
+ "learning_rate": 5.22e-06,
788
+ "loss": 0.8655,
789
+ "step": 114
790
+ },
791
+ {
792
+ "epoch": 20.91,
793
+ "learning_rate": 5.16e-06,
794
+ "loss": 0.8695,
795
+ "step": 115
796
+ },
797
+ {
798
+ "epoch": 21.09,
799
+ "learning_rate": 5.1e-06,
800
+ "loss": 0.8636,
801
+ "step": 116
802
+ },
803
+ {
804
+ "epoch": 21.27,
805
+ "learning_rate": 5.04e-06,
806
+ "loss": 0.8513,
807
+ "step": 117
808
+ },
809
+ {
810
+ "epoch": 21.45,
811
+ "learning_rate": 4.98e-06,
812
+ "loss": 0.8523,
813
+ "step": 118
814
+ },
815
+ {
816
+ "epoch": 21.64,
817
+ "learning_rate": 4.9199999999999995e-06,
818
+ "loss": 0.8497,
819
+ "step": 119
820
+ },
821
+ {
822
+ "epoch": 21.82,
823
+ "learning_rate": 4.86e-06,
824
+ "loss": 0.8441,
825
+ "step": 120
826
+ },
827
+ {
828
+ "epoch": 21.82,
829
+ "eval_loss": 1.1353310346603394,
830
+ "eval_runtime": 1.4613,
831
+ "eval_samples_per_second": 45.164,
832
+ "eval_steps_per_second": 2.053,
833
+ "step": 120
834
+ },
835
+ {
836
+ "epoch": 22.0,
837
+ "learning_rate": 4.800000000000001e-06,
838
+ "loss": 0.8543,
839
+ "step": 121
840
+ },
841
+ {
842
+ "epoch": 22.18,
843
+ "learning_rate": 4.74e-06,
844
+ "loss": 0.838,
845
+ "step": 122
846
+ },
847
+ {
848
+ "epoch": 22.36,
849
+ "learning_rate": 4.68e-06,
850
+ "loss": 0.8348,
851
+ "step": 123
852
+ },
853
+ {
854
+ "epoch": 22.55,
855
+ "learning_rate": 4.62e-06,
856
+ "loss": 0.8215,
857
+ "step": 124
858
+ },
859
+ {
860
+ "epoch": 22.73,
861
+ "learning_rate": 4.56e-06,
862
+ "loss": 0.8313,
863
+ "step": 125
864
+ },
865
+ {
866
+ "epoch": 22.91,
867
+ "learning_rate": 4.5e-06,
868
+ "loss": 0.8303,
869
+ "step": 126
870
+ },
871
+ {
872
+ "epoch": 23.09,
873
+ "learning_rate": 4.44e-06,
874
+ "loss": 0.8188,
875
+ "step": 127
876
+ },
877
+ {
878
+ "epoch": 23.27,
879
+ "learning_rate": 4.38e-06,
880
+ "loss": 0.812,
881
+ "step": 128
882
+ },
883
+ {
884
+ "epoch": 23.45,
885
+ "learning_rate": 4.32e-06,
886
+ "loss": 0.8077,
887
+ "step": 129
888
+ },
889
+ {
890
+ "epoch": 23.64,
891
+ "learning_rate": 4.26e-06,
892
+ "loss": 0.7956,
893
+ "step": 130
894
+ },
895
+ {
896
+ "epoch": 23.64,
897
+ "eval_loss": 1.1432788372039795,
898
+ "eval_runtime": 1.461,
899
+ "eval_samples_per_second": 45.176,
900
+ "eval_steps_per_second": 2.053,
901
+ "step": 130
902
+ },
903
+ {
904
+ "epoch": 23.82,
905
+ "learning_rate": 4.2e-06,
906
+ "loss": 0.8099,
907
+ "step": 131
908
+ },
909
+ {
910
+ "epoch": 24.0,
911
+ "learning_rate": 4.14e-06,
912
+ "loss": 0.8223,
913
+ "step": 132
914
+ },
915
+ {
916
+ "epoch": 24.18,
917
+ "learning_rate": 4.080000000000001e-06,
918
+ "loss": 0.7956,
919
+ "step": 133
920
+ },
921
+ {
922
+ "epoch": 24.36,
923
+ "learning_rate": 4.0200000000000005e-06,
924
+ "loss": 0.7889,
925
+ "step": 134
926
+ },
927
+ {
928
+ "epoch": 24.55,
929
+ "learning_rate": 3.96e-06,
930
+ "loss": 0.7844,
931
+ "step": 135
932
+ },
933
+ {
934
+ "epoch": 24.73,
935
+ "learning_rate": 3.9e-06,
936
+ "loss": 0.7867,
937
+ "step": 136
938
+ },
939
+ {
940
+ "epoch": 24.91,
941
+ "learning_rate": 3.8400000000000005e-06,
942
+ "loss": 0.7875,
943
+ "step": 137
944
+ },
945
+ {
946
+ "epoch": 25.09,
947
+ "learning_rate": 3.7800000000000002e-06,
948
+ "loss": 0.7804,
949
+ "step": 138
950
+ },
951
+ {
952
+ "epoch": 25.27,
953
+ "learning_rate": 3.72e-06,
954
+ "loss": 0.7661,
955
+ "step": 139
956
+ },
957
+ {
958
+ "epoch": 25.45,
959
+ "learning_rate": 3.66e-06,
960
+ "loss": 0.7696,
961
+ "step": 140
962
+ },
963
+ {
964
+ "epoch": 25.45,
965
+ "eval_loss": 1.1572602987289429,
966
+ "eval_runtime": 1.4603,
967
+ "eval_samples_per_second": 45.197,
968
+ "eval_steps_per_second": 2.054,
969
+ "step": 140
970
+ },
971
+ {
972
+ "epoch": 25.64,
973
+ "learning_rate": 3.6e-06,
974
+ "loss": 0.7531,
975
+ "step": 141
976
+ },
977
+ {
978
+ "epoch": 25.82,
979
+ "learning_rate": 3.54e-06,
980
+ "loss": 0.7538,
981
+ "step": 142
982
+ },
983
+ {
984
+ "epoch": 26.0,
985
+ "learning_rate": 3.4799999999999997e-06,
986
+ "loss": 0.7574,
987
+ "step": 143
988
+ },
989
+ {
990
+ "epoch": 26.18,
991
+ "learning_rate": 3.42e-06,
992
+ "loss": 0.746,
993
+ "step": 144
994
+ },
995
+ {
996
+ "epoch": 26.36,
997
+ "learning_rate": 3.3600000000000004e-06,
998
+ "loss": 0.742,
999
+ "step": 145
1000
+ },
1001
+ {
1002
+ "epoch": 26.55,
1003
+ "learning_rate": 3.3e-06,
1004
+ "loss": 0.7371,
1005
+ "step": 146
1006
+ },
1007
+ {
1008
+ "epoch": 26.73,
1009
+ "learning_rate": 3.2400000000000003e-06,
1010
+ "loss": 0.7355,
1011
+ "step": 147
1012
+ },
1013
+ {
1014
+ "epoch": 26.91,
1015
+ "learning_rate": 3.18e-06,
1016
+ "loss": 0.7352,
1017
+ "step": 148
1018
+ },
1019
+ {
1020
+ "epoch": 27.09,
1021
+ "learning_rate": 3.12e-06,
1022
+ "loss": 0.7263,
1023
+ "step": 149
1024
+ },
1025
+ {
1026
+ "epoch": 27.27,
1027
+ "learning_rate": 3.06e-06,
1028
+ "loss": 0.7113,
1029
+ "step": 150
1030
+ },
1031
+ {
1032
+ "epoch": 27.27,
1033
+ "eval_loss": 1.1966105699539185,
1034
+ "eval_runtime": 1.4607,
1035
+ "eval_samples_per_second": 45.185,
1036
+ "eval_steps_per_second": 2.054,
1037
+ "step": 150
1038
+ },
1039
+ {
1040
+ "epoch": 27.45,
1041
+ "learning_rate": 3e-06,
1042
+ "loss": 0.7142,
1043
+ "step": 151
1044
+ },
1045
+ {
1046
+ "epoch": 27.64,
1047
+ "learning_rate": 2.9400000000000002e-06,
1048
+ "loss": 0.7189,
1049
+ "step": 152
1050
+ },
1051
+ {
1052
+ "epoch": 27.82,
1053
+ "learning_rate": 2.88e-06,
1054
+ "loss": 0.7131,
1055
+ "step": 153
1056
+ },
1057
+ {
1058
+ "epoch": 28.0,
1059
+ "learning_rate": 2.82e-06,
1060
+ "loss": 0.7054,
1061
+ "step": 154
1062
+ },
1063
+ {
1064
+ "epoch": 28.18,
1065
+ "learning_rate": 2.7600000000000003e-06,
1066
+ "loss": 0.6965,
1067
+ "step": 155
1068
+ },
1069
+ {
1070
+ "epoch": 28.36,
1071
+ "learning_rate": 2.7e-06,
1072
+ "loss": 0.6879,
1073
+ "step": 156
1074
+ },
1075
+ {
1076
+ "epoch": 28.55,
1077
+ "learning_rate": 2.64e-06,
1078
+ "loss": 0.6872,
1079
+ "step": 157
1080
+ },
1081
+ {
1082
+ "epoch": 28.73,
1083
+ "learning_rate": 2.58e-06,
1084
+ "loss": 0.6909,
1085
+ "step": 158
1086
+ },
1087
+ {
1088
+ "epoch": 28.91,
1089
+ "learning_rate": 2.52e-06,
1090
+ "loss": 0.6734,
1091
+ "step": 159
1092
+ },
1093
+ {
1094
+ "epoch": 29.09,
1095
+ "learning_rate": 2.4599999999999997e-06,
1096
+ "loss": 0.6646,
1097
+ "step": 160
1098
+ },
1099
+ {
1100
+ "epoch": 29.09,
1101
+ "eval_loss": 1.2236099243164062,
1102
+ "eval_runtime": 1.4503,
1103
+ "eval_samples_per_second": 45.508,
1104
+ "eval_steps_per_second": 2.069,
1105
+ "step": 160
1106
+ },
1107
+ {
1108
+ "epoch": 29.27,
1109
+ "learning_rate": 2.4000000000000003e-06,
1110
+ "loss": 0.6715,
1111
+ "step": 161
1112
+ },
1113
+ {
1114
+ "epoch": 29.45,
1115
+ "learning_rate": 2.34e-06,
1116
+ "loss": 0.6621,
1117
+ "step": 162
1118
+ },
1119
+ {
1120
+ "epoch": 29.64,
1121
+ "learning_rate": 2.28e-06,
1122
+ "loss": 0.6483,
1123
+ "step": 163
1124
+ },
1125
+ {
1126
+ "epoch": 29.82,
1127
+ "learning_rate": 2.22e-06,
1128
+ "loss": 0.6487,
1129
+ "step": 164
1130
+ },
1131
+ {
1132
+ "epoch": 30.0,
1133
+ "learning_rate": 2.16e-06,
1134
+ "loss": 0.6553,
1135
+ "step": 165
1136
+ },
1137
+ {
1138
+ "epoch": 30.18,
1139
+ "learning_rate": 2.1e-06,
1140
+ "loss": 0.6344,
1141
+ "step": 166
1142
+ },
1143
+ {
1144
+ "epoch": 30.36,
1145
+ "learning_rate": 2.0400000000000004e-06,
1146
+ "loss": 0.6334,
1147
+ "step": 167
1148
+ },
1149
+ {
1150
+ "epoch": 30.55,
1151
+ "learning_rate": 1.98e-06,
1152
+ "loss": 0.6311,
1153
+ "step": 168
1154
+ },
1155
+ {
1156
+ "epoch": 30.73,
1157
+ "learning_rate": 1.9200000000000003e-06,
1158
+ "loss": 0.6262,
1159
+ "step": 169
1160
+ },
1161
+ {
1162
+ "epoch": 30.91,
1163
+ "learning_rate": 1.86e-06,
1164
+ "loss": 0.6303,
1165
+ "step": 170
1166
+ },
1167
+ {
1168
+ "epoch": 30.91,
1169
+ "eval_loss": 1.2599583864212036,
1170
+ "eval_runtime": 1.4611,
1171
+ "eval_samples_per_second": 45.171,
1172
+ "eval_steps_per_second": 2.053,
1173
+ "step": 170
1174
+ },
1175
+ {
1176
+ "epoch": 31.09,
1177
+ "learning_rate": 1.8e-06,
1178
+ "loss": 0.6196,
1179
+ "step": 171
1180
+ },
1181
+ {
1182
+ "epoch": 31.27,
1183
+ "learning_rate": 1.7399999999999999e-06,
1184
+ "loss": 0.5957,
1185
+ "step": 172
1186
+ },
1187
+ {
1188
+ "epoch": 31.45,
1189
+ "learning_rate": 1.6800000000000002e-06,
1190
+ "loss": 0.5994,
1191
+ "step": 173
1192
+ },
1193
+ {
1194
+ "epoch": 31.64,
1195
+ "learning_rate": 1.6200000000000002e-06,
1196
+ "loss": 0.6109,
1197
+ "step": 174
1198
+ },
1199
+ {
1200
+ "epoch": 31.82,
1201
+ "learning_rate": 1.56e-06,
1202
+ "loss": 0.61,
1203
+ "step": 175
1204
+ },
1205
+ {
1206
+ "epoch": 32.0,
1207
+ "learning_rate": 1.5e-06,
1208
+ "loss": 0.5957,
1209
+ "step": 176
1210
+ },
1211
+ {
1212
+ "epoch": 32.18,
1213
+ "learning_rate": 1.44e-06,
1214
+ "loss": 0.5862,
1215
+ "step": 177
1216
+ },
1217
+ {
1218
+ "epoch": 32.36,
1219
+ "learning_rate": 1.3800000000000001e-06,
1220
+ "loss": 0.5799,
1221
+ "step": 178
1222
+ },
1223
+ {
1224
+ "epoch": 32.55,
1225
+ "learning_rate": 1.32e-06,
1226
+ "loss": 0.5788,
1227
+ "step": 179
1228
+ },
1229
+ {
1230
+ "epoch": 32.73,
1231
+ "learning_rate": 1.26e-06,
1232
+ "loss": 0.5741,
1233
+ "step": 180
1234
+ },
1235
+ {
1236
+ "epoch": 32.73,
1237
+ "eval_loss": 1.3407286405563354,
1238
+ "eval_runtime": 1.4616,
1239
+ "eval_samples_per_second": 45.157,
1240
+ "eval_steps_per_second": 2.053,
1241
+ "step": 180
1242
+ },
1243
+ {
1244
+ "epoch": 32.91,
1245
+ "learning_rate": 1.2000000000000002e-06,
1246
+ "loss": 0.5745,
1247
+ "step": 181
1248
+ },
1249
+ {
1250
+ "epoch": 33.09,
1251
+ "learning_rate": 1.14e-06,
1252
+ "loss": 0.5654,
1253
+ "step": 182
1254
+ },
1255
+ {
1256
+ "epoch": 33.27,
1257
+ "learning_rate": 1.08e-06,
1258
+ "loss": 0.5587,
1259
+ "step": 183
1260
+ },
1261
+ {
1262
+ "epoch": 33.45,
1263
+ "learning_rate": 1.0200000000000002e-06,
1264
+ "loss": 0.5539,
1265
+ "step": 184
1266
+ },
1267
+ {
1268
+ "epoch": 33.64,
1269
+ "learning_rate": 9.600000000000001e-07,
1270
+ "loss": 0.558,
1271
+ "step": 185
1272
+ },
1273
+ {
1274
+ "epoch": 33.82,
1275
+ "learning_rate": 9e-07,
1276
+ "loss": 0.5571,
1277
+ "step": 186
1278
+ },
1279
+ {
1280
+ "epoch": 34.0,
1281
+ "learning_rate": 8.400000000000001e-07,
1282
+ "loss": 0.5496,
1283
+ "step": 187
1284
+ }
1285
+ ],
1286
+ "logging_steps": 1,
1287
+ "max_steps": 200,
1288
+ "num_train_epochs": 40,
1289
+ "save_steps": 500,
1290
+ "total_flos": 6.274288047202763e+18,
1291
+ "trial_name": null,
1292
+ "trial_params": null
1293
+ }
training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:164da52fddefcb4fa95359f4d25b6fb04b19f1c27f0445d05b814bea4965d4b4
3
+ size 5947
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)