-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathlm_eval_8.sh
120 lines (72 loc) · 3.4 KB
/
lm_eval_8.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
#!/bin/bash
export DEBUG=0
export PYTHONPATH=/home/LeiFeng/xiaolong/moe_quantize/optimum/:$PYTHONPATH:/home/LeiFeng/xiaolong/moe_quantize/auto_gptq/:$PYTHONPATH
export CUDA_VISIBLE_DEVICES=0,1,2
nohup python lm_eval.py \
--model_name deepseek-ai/deepseek-moe-16b-base \
--quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.shared_4.other.2+other_block.8 \
--bits moe.shared_4.other.2+other_block.8 \
--is_quantized > eval_output_moe_shared_4_other_2+other_block_8.log &
echo "moe_shared_4_other_2+other_block_8 done"
wait
nohup python lm_eval.py \
--model_name deepseek-ai/deepseek-moe-16b-base \
--quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.shared_2.other.4+other_block.8 \
--bits moe.shared_2.other.4+other_block.8 \
--is_quantized > eval_output_moe_shared_2_other_4+other_block_8.log &
echo "moe_shared_2_other_4+other_block_8 done"
wait
# ======= below finish =========
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_all_4 \
# --bits all_4 \
# --is_quantized > eval_output_all_4.log &
# echo "all_4 done"
# wait
# export DEBUG=0
# export PYTHONPATH=/home/LeiFeng/xiaolong/moe_quantize/optimum/:$PYTHONPATH:/home/LeiFeng/xiaolong/moe_quantize/auto_gptq/:$PYTHONPATH
# export CUDA_VISIBLE_DEVICES=3,4
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_all_2 \
# --bits all_2 \
# --is_quantized > eval_output_all_2.log &
# echo "all_2 done"
# wait
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_all_8 \
# --bits all_8 \
# --is_quantized > eval_output_all_8.log &
# wait
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.all_mlp.2+other_block.4 \
# --bits moe.all_mlp.2+other_block.4 \
# --is_quantized > eval_output_moe_all_mlp_2+other_block_4.log &
# echo "moe_all_mlp_2+other_block_4 done"
# wait
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.shared_4.other.2+other_block_4 \
# --bits moe.shared_4.other.2+other_block_4 \
# --is_quantized > eval_output_moe_shared_4_other_2+other_block_4.log &
# echo "moe_shared_4_other_2+other_block_4 done"
# wait
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.shared_2.other.4+other_block_4 \
# --bits moe.shared_2.other.4+other_block_4 \
# --is_quantized > eval_output_moe_shared_2_other_4+other_block_4.log &
# echo "moe_shared_2_other_4+other_block_4 done"
# wait
# nohup python lm_eval.py \
# --model_name deepseek-ai/deepseek-moe-16b-base \
# --quant_model_path autogptq_deepseek-ai/deepseek-moe-16b-base-gptq_w_bit_moe.all_mlp.4+other_block.8 \
# --bits moe.all_mlp.4+other_block.8 \
# --is_quantized > eval_output_moe_all_mlp_4+other_block_8.log &
# echo "moe_all_mlp_4+other_block_8 done"
# wait
# model scale up
# parameter scale up