SIFT/eval_3b.sh at main · SJTU-DENG-Lab/SIFT · GitHub

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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
#!/bin/bash
set -euo pipefail
IFS=$'\n\t'

########################################
# Environment Variables & Constants
########################################

export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/faketime/libfaketime.so.1
export FAKETIME="@2025-01-20 15:32:02"
export CUDA_VISIBLE_DEVICES=0,1,2,3
export WANDB_MODE=offline
export PYTHONHASHSEED=42
export RANDOM_SEED=42

# Common Variables
VERSION='0208_llama3_2_3b_chat_AIME'
MODEL_PATH="/models/llama3_3b_instruct"
NUM_GPUS=4
OUTPUT_BASE="outputs/llama3_2_3b_AIME/gsm8k/${VERSION}"

########################################
# Stage 0: GSM8K Q2P (Query => Prediction)
########################################
sed -i "s|path='[^']*'|path='/data/gsm8k/AIME.jsonl'|" "configs/datasets/gsm8k/gsm8k_q2p.py"

python run_no_random.py \
    --datasets gsm8k_q2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs0/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs0/predictions/llama3_3b_instruct_hf/gsm8k.json"

########################################
# Stage 1 & 2
########################################

sed -i "s|path='[^']*'|path='/data/gsm8k/AIME.jsonl'|" "configs/datasets/gsm8k/gsm8k_q2a.py"

# Stage 1: Query => Abstract
python run_no_random.py \
    --datasets gsm8k_q2a \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs1/q2a/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs1/q2a/predictions/llama3_3b_instruct_hf/gsm8k.json"

python abs_postprocessing.py \
    --input_file "${OUTPUT_BASE}/abs1/q2a/final.jsonl" \
    --output_file "${OUTPUT_BASE}/abs1/q2a/q_a.jsonl"

# Stage 1: Query + Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs1/q2a/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_q_a2p.py"

python run_no_random.py \
    --datasets gsm8k_q_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs1/q_a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs1/q_a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Stage 1: Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs1/q2a/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_a2p.py"

python run_no_random.py \
    --datasets gsm8k_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs1/a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs1/a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Accumulate Stage 2 Results
python acc_stage2.py \
    --q2p "${OUTPUT_BASE}/abs0/final.jsonl" \
    --q2a "${OUTPUT_BASE}/abs1/q2a/final.jsonl" \
    --q_a2p "${OUTPUT_BASE}/abs1/q_a2p/final.jsonl" \
    --a2p "${OUTPUT_BASE}/abs1/a2p/final.jsonl" \
    --abs2improve "${OUTPUT_BASE}/abs1/abs2improve.jsonl" \
    --stage2res "${OUTPUT_BASE}/abs1/stage2res.jsonl"

########################################
# Stage 3
########################################

# Stage 3: Query + Abstract => Abstract (Forward)
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs1/abs2improve.jsonl'|" "configs/datasets/gsm8k/gsm8k_q_a2a.py"

python run_no_random.py \
    --datasets gsm8k_q_a2a \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs2/q_a2a/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs2/q_a2a/predictions/llama3_3b_instruct_hf/gsm8k.json"

python abs_postprocessing_v2.py \
    --input_file "${OUTPUT_BASE}/abs2/q_a2a/final.jsonl" \
    --output_file "${OUTPUT_BASE}/abs2/q_a.jsonl"

# Stage 3: Query + Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs2/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_q_a2p.py"

python run_no_random.py \
    --datasets gsm8k_q_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs2/q_a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs2/q_a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Stage 3: Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs2/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_a2p.py"

python run_no_random.py \
    --datasets gsm8k_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs2/a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs2/a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Accumulate Stage 3 Results
python acc_stage3.py \
    --q2p "${OUTPUT_BASE}/abs0/final.jsonl" \
    --q2a "${OUTPUT_BASE}/abs1/q2a/final.jsonl" \
    --q_a2p "${OUTPUT_BASE}/abs1/q_a2p/final.jsonl" \
    --a2p "${OUTPUT_BASE}/abs1/a2p/final.jsonl" \
    --q_a2a "${OUTPUT_BASE}/abs2/q_a2a/final.jsonl" \
    --q_a2p2 "${OUTPUT_BASE}/abs2/q_a2p/final.jsonl" \
    --a2p2 "${OUTPUT_BASE}/abs2/a2p/final.jsonl" \
    --abs2improve "${OUTPUT_BASE}/abs2/abs2improve.jsonl" \
    --stage3res "${OUTPUT_BASE}/abs2/stage3res.jsonl"

########################################
# Stage 4
########################################

# Stage 4: Prediction => Abstract (Backward)
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs2/abs2improve.jsonl'|" "configs/datasets/gsm8k/gsm8k_p2a.py"

python run_no_random.py \
    --datasets gsm8k_p2a \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs3/p2a/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs3/p2a/predictions/llama3_3b_instruct_hf/gsm8k.json"

python abs_postprocessing_v3.py \
    --input_file "${OUTPUT_BASE}/abs3/p2a/final.jsonl" \
    --input_file2 "${OUTPUT_BASE}/abs2/abs2improve.jsonl" \
    --output_file "${OUTPUT_BASE}/abs3/p2a/q_a.jsonl"

# Stage 4: Query + Abstract => Abstract (Forward)
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs3/p2a/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_q_a2a.py"

python run_no_random.py \
    --datasets gsm8k_q_a2a \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs3/q_a2a/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs3/q_a2a/predictions/llama3_3b_instruct_hf/gsm8k.json"

python abs_postprocessing_v2.py \
    --input_file "${OUTPUT_BASE}/abs3/q_a2a/final.jsonl" \
    --output_file "${OUTPUT_BASE}/abs3/q_a.jsonl"

# Stage 4: Query + Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs3/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_q_a2p.py"

python run_no_random.py \
    --datasets gsm8k_q_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs3/q_a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs3/q_a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Stage 4: Abstract => Prediction
sed -i "s|path='[^']*'|path='${OUTPUT_BASE}/abs3/q_a.jsonl'|" "configs/datasets/gsm8k/gsm8k_a2p.py"

python run_no_random.py \
    --datasets gsm8k_a2p \
    --hf-type chat \
    --hf-path "${MODEL_PATH}" \
    --dump-eval-details \
    --hf-num-gpus "${NUM_GPUS}" \
    -w "${OUTPUT_BASE}/abs3/a2p/"

python gsm8k_compare_token_num_v2.py \
    --base_path "${OUTPUT_BASE}/abs3/a2p/predictions/llama3_3b_instruct_hf/gsm8k.json"

# Accumulate Stage 4 Results
python acc_stage4.py \
    --q2p "${OUTPUT_BASE}/abs0/final.jsonl" \
    --q2a "${OUTPUT_BASE}/abs1/q2a/final.jsonl" \
    --q_a2p "${OUTPUT_BASE}/abs1/q_a2p/final.jsonl" \
    --a2p "${OUTPUT_BASE}/abs1/a2p/final.jsonl" \
    --q_a2a "${OUTPUT_BASE}/abs2/q_a2a/final.jsonl" \
    --q_a2p2 "${OUTPUT_BASE}/abs2/q_a2p/final.jsonl" \
    --a2p2 "${OUTPUT_BASE}/abs2/a2p/final.jsonl" \
    --p2a "${OUTPUT_BASE}/abs3/p2a/final.jsonl" \
    --q_a2a2 "${OUTPUT_BASE}/abs3/q_a2a/final.jsonl" \
    --q_a2p3 "${OUTPUT_BASE}/abs3/q_a2p/final.jsonl" \
    --a2p3 "${OUTPUT_BASE}/abs3/a2p/final.jsonl" \
    --stage4res "${OUTPUT_BASE}/abs3/stage4res.jsonl"