-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathrun_pipeline.sh
210 lines (162 loc) · 8.2 KB
/
run_pipeline.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
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
#!/bin/bash
export CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7
NUM_GPUS=8
SEED=42
RAW_DATASET_LIST=('tulu_300k') # data source
rating_model="meta-llama/Meta-Llama-3.1-8B-Instruct" #"gpt-4o-mini" 'mistralai/Mistral-7B-Instruct-v0.3'
declare -A base_models
# base_models["meta-llama/Meta-Llama-3.1-8B"]="128 1 2048" # TOTAL_BATCH_SIZE BATCH_SIZE_PER_GPU max_seq_length
base_models["meta-llama/Llama-3.2-3B"]="32 1 128" # TOTAL_BATCH_SIZE BATCH_SIZE_PER_GPU max_seq_length
# data types represent the generated subsets by baselines
data_types=('ds2_10k')
#############################################################
######## model finetuning on selected training data #########
#############################################################
cluster_root_path="../model_output"
mkdir -p $cluster_root_path
# for base_model in "${!base_models[@]}"
# do
# IFS=' ' read -r -a params <<< "${base_models[$base_model]}"
# TOTAL_BATCH_SIZE=${params[0]}
# BATCH_SIZE_PER_GPU=${params[1]}
# max_seq_length=${params[2]}
# for raw_dataset_name in "${RAW_DATASET_LIST[@]}"
# do
# for data_type in "${data_types[@]}"
# do
# if [[ $data_type == "base" ]]; then
# echo "Skipping base model finetune"
# continue
# fi
# mkdir -p $cluster_root_path/models/
# train_data="../selected_data/${rating_model}/${raw_dataset_name}/${data_type}_dataset.json"
# GRADIENT_ACC_STEPS=$(($TOTAL_BATCH_SIZE/$NUM_GPUS/$BATCH_SIZE_PER_GPU))
# echo "Training ${base_model} using $NUM_GPUS GPUs, $BATCH_SIZE_PER_GPU batch size per GPU, $GRADIENT_ACC_STEPS gradient accumulation steps"
# echo "Training data path: ${train_data}"
# ### Lora training
# accelerate launch \
# --mixed_precision bf16 \
# --num_machines 1 \
# --num_processes $NUM_GPUS \
# finetune.py \
# --model_name_or_path $base_model \
# --use_lora \
# --lora_rank 64 \
# --lora_alpha 16 \
# --seed $SEED \
# --lora_dropout 0.1 \
# --tokenizer_name $base_model \
# --use_slow_tokenizer \
# --train_file $train_data \
# --max_seq_length $max_seq_length \
# --preprocessing_num_workers 16 \
# --checkpointing_steps epoch \
# --per_device_train_batch_size $BATCH_SIZE_PER_GPU \
# --gradient_accumulation_steps $GRADIENT_ACC_STEPS \
# --learning_rate 1e-4 \
# --lr_scheduler_type linear \
# --warmup_ratio 0.03 \
# --weight_decay 0. \
# --num_train_epochs 5 \
# --output_dir $cluster_root_path/models/${rating_model}/${raw_dataset_name}/${base_model}/lora_${data_type}/ \
# --with_tracking \
# --report_to tensorboard \
# --logging_steps 1
# python merge_lora.py \
# --base_model_name_or_path $base_model \
# --lora_model_name_or_path $cluster_root_path/models/${rating_model}/${raw_dataset_name}/${base_model}/lora_${data_type}/ \
# --output_dir $cluster_root_path/models/${rating_model}/${raw_dataset_name}/${base_model}/lora_merged_${data_type}/ \
# --save_tokenizer
# sleep 10s
# rm -rf $cluster_root_path/models/${rating_model}/${raw_dataset_name}/${base_model}/lora_${data_type}
# done
# done
# done
wait
############################################################
############### finetuned model evaluation ################
############################################################
echo "starting evaluating finetuned models..."
for base_model in "${!base_models[@]}"; do
for raw_dataset_name in "${RAW_DATASET_LIST[@]}"; do
for data_type in "${data_types[@]}"; do
model_name_or_path=$cluster_root_path/models/${rating_model}/${raw_dataset_name}/${base_model}/lora_merged_${data_type}
if [[ $data_type == "base" ]]; then
echo "base model evaluation"
model_name_or_path=$base_model
fi
echo "###### Processing data type:: ${data_type}"
#### MMLU: factual knowledge
eval_dataset_name='mmlu'
local_save_dir=${cluster_root_path}/results/${rating_model}/${raw_dataset_name}/${eval_dataset_name}/${base_model}/$data_type
CUDA_VISIBLE_DEVICES=0 python -m eval.mmlu.run_eval \
--ntrain 0 \
--data_dir raw_data/eval/mmlu \
--save_dir ${local_save_dir} \
--model_name_or_path $model_name_or_path \
--tokenizer_name_or_path $model_name_or_path \
--eval_batch_size 8 &
##### GSM8k: reasoning
eval_dataset_name='gsm'
local_save_dir=${cluster_root_path}/results/${rating_model}/${raw_dataset_name}/${eval_dataset_name}/${base_model}/$data_type
CUDA_VISIBLE_DEVICES=1 python -m eval.gsm.run_eval \
--data_dir raw_data/eval/gsm/ \
--max_num_examples 200 \
--save_dir ${local_save_dir} \
--model_name_or_path $model_name_or_path \
--tokenizer_name_or_path $model_name_or_path \
--n_shot 8 \
--use_vllm &
###### BBH: reasoning
eval_dataset_name='bbh'
local_save_dir=${cluster_root_path}/results/${rating_model}/${raw_dataset_name}/${eval_dataset_name}/${base_model}/$data_type
CUDA_VISIBLE_DEVICES=2 python -m eval.bbh.run_eval \
--data_dir raw_data/eval/bbh \
--save_dir ${local_save_dir} \
--model_name_or_path $model_name_or_path \
--tokenizer_name_or_path $model_name_or_path \
--max_num_examples_per_task 40 \
--use_vllm &
##### truthfulness
eval_dataset_name='truthfulqa'
local_save_dir=${cluster_root_path}/results/${rating_model}/${raw_dataset_name}/${eval_dataset_name}/${base_model}/$data_type
CUDA_VISIBLE_DEVICES=3 python -m eval.truthfulqa.run_eval \
--data_dir raw_data/eval/truthfulqa \
--save_dir ${local_save_dir} \
--model_name_or_path $model_name_or_path \
--tokenizer_name_or_path $model_name_or_path \
--metrics truth info mc \
--preset qa \
--hf_truth_model_name_or_path allenai/truthfulqa-truth-judge-llama2-7B \
--hf_info_model_name_or_path allenai/truthfulqa-info-judge-llama2-7B \
--eval_batch_size 20 \
--load_in_8bit &
###### multilinguality
eval_dataset_name='tydiqa'
local_save_dir=${cluster_root_path}/results/${rating_model}/${raw_dataset_name}/${eval_dataset_name}/${base_model}/$data_type
CUDA_VISIBLE_DEVICES=4 python -m eval.tydiqa.run_eval \
--data_dir raw_data/eval/tydiqa/ \
--n_shot 1 \
--max_num_examples_per_lang 100 \
--max_context_length 512 \
--save_dir ${local_save_dir} \
--model_name_or_path $model_name_or_path \
--tokenizer_name_or_path $model_name_or_path \
--eval_batch_size 20 \
--load_in_8bit &
wait
done
done
done
sleep 10s
for base_model in "${!base_models[@]}"; do
for raw_dataset_name in "${RAW_DATASET_LIST[@]}"; do
for data_type in "${data_types[@]}"; do
echo "*** Processing rating model:: ${rating_model} ***"
echo "*** Processing Base model:: ${base_model} ***"
echo "*** Processing training dataset:: ${raw_dataset_name} ***"
echo "*** Processing data type:: ${data_type} ***"
python3 read_results.py --root_result_path "${cluster_root_path}/results" --raw_dataset $raw_dataset_name --base_model $base_model --rating_model $rating_model --baseline_tag $data_type
done
done
done