-
Notifications
You must be signed in to change notification settings - Fork 46
/
EasyContext-256K-danube2.sh
165 lines (146 loc) · 4.88 KB
/
EasyContext-256K-danube2.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
# First, download h2oai/h2o-danube2-1.8b-base to output/h2o-danube2-1.8b-base and change the "architectures" from MistralForCausalLM to LlamaForCausalLM and "model_type" from mistral to llama in the config.json file.
# HF's mistral architecture is off when used together with ring attention due to the way it implements rotary embedding. Llama has no such issue.
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 2 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_16K_rope_30K \
--wandb EasyContext \
--seed 2024 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o-danube2-1.8b-base \
--seq-length 16000 \
--rope-theta 30000 \
--parallel_mode data_parallel
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 1 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_32K_rope_100K \
--wandb EasyContext \
--seed 2025 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_400_lr_2e-5_16K_rope_30K \
--seq-length 32000 \
--rope-theta 100000 \
--parallel_mode data_parallel
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 1 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_32K_rope_300K \
--wandb EasyContext \
--seed 2026 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_400_lr_2e-5_32K_rope_100K \
--seq-length 32000 \
--rope-theta 300000 \
--parallel_mode data_parallel
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 4 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_64K_rope_1M \
--wandb EasyContext \
--seed 2027 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_400_lr_2e-5_32K_rope_300K \
--seq-length 64000 \
--rope-theta 1000000 \
--parallel_mode zigzag_ring_attn
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 4 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_64K_rope_2.5M \
--wandb EasyContext \
--seed 2028 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_400_lr_2e-5_64K_rope_1M \
--seq-length 64000 \
--rope-theta 2500000 \
--parallel_mode zigzag_ring_attn
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 4 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_400_lr_2e-5_64K_rope_5M \
--wandb EasyContext \
--seed 2029 \
--max-train-steps 400 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_400_lr_2e-5_64K_rope_2.5M \
--seq-length 64000 \
--rope-theta 5000000 \
--parallel_mode zigzag_ring_attn
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 2 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_300_lr_2e-5_128K_rope_10M \
--wandb EasyContext \
--seed 2030 \
--max-train-steps 300 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model PY007/EasyContext-64K-h2o \
--seq-length 128000 \
--rope-theta 10000000 \
--parallel_mode zigzag_ring_attn
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 2 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_200_lr_2e-5_128K_rope_30M \
--wandb EasyContext \
--seed 2031 \
--max-train-steps 200 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_128K \
--model output/h2o_bs_1M_step_300_lr_2e-5_128K_rope_10M \
--seq-length 128000 \
--rope-theta 30000000 \
--parallel_mode zigzag_ring_attn
accelerate launch \
--config_file accelerate_configs/single_node.yaml \
--main_process_port 12345 \
train.py \
--batch-size 1 \
--gradient-accumulate-every 4 \
--output-dir ./output/h2o_bs_1M_step_200_lr_2e-5_256K_rope_100M \
--wandb EasyContext \
--seed 2032 \
--max-train-steps 200 \
--learning-rate 2e-5 \
--dataset PY007/slimpajama_mistral_tokenized_upsample_4096_chunk_1M \
--model output/h2o_bs_1M_step_200_lr_2e-5_128K_rope_30M \
--seq-length 256000 \
--rope-theta 100000000 \
--parallel_mode zigzag_ring_attn