-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathInstruction Guidelines for SAFE-MEME-QA.txt
44 lines (26 loc) · 3.3 KB
/
Instruction Guidelines for SAFE-MEME-QA.txt
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
Download unit_modules from - https://drive.google.com/drive/folders/1PWESNhZIDa1YL6aipVyrl_Pwo4LYo1Zu?usp=sharing
#
# Training the SAFE-MEME:
#
* File: main_trainUnitModels.py is used sequentially to train the model.
* Stage 1: Kindly use the following command to train the query generation module: trainAllQueryGen
CUDA_VISIBLE_DEVICES=0 python3 main_trainUnitModels.py --full_FT YES --img_type vit --execution_mode trainAllQueryGen --user_msg rationale --epoch 50 --lr 5e-5 --output_len 128 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
* Stage 1: Kindly use the following command to train the response generation module: trainAllQueryGen_1Q1A
CUDA_VISIBLE_DEVICES=0 python3 main_trainUnitModels.py --full_FT YES --img_type vit --execution_mode trainAllQueryGen_1Q1A --user_msg rationale --epoch 50 --lr 5e-5 --output_len 85 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
#
# Evaluating SAFE-MEME:
#
* For evaluation of the MGS datset, use pipe_testUnitModelsOutputGen, and utils_dataTestUnitModels.py
* Kindly use VIT_T5Base_FullFT_trainAllQueryGen (checkpoint-9350) and VIT_T5Base_FullFT_trainAllQueryGen_1Q1A (checkpoint-43450)
* Stage 1: Kindly use the following command to generate the set of queries.
CUDA_VISIBLE_DEVICES=0 python3 pipe_testUnitModelsOutputGen.py --execution_mode testAllQueryGen --img_type vit --output_len 128 --full_FT YES --user_msg rationale --epoch 50 --lr 5e-5 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
* Stage 2: Kindly use the following command to generate the set of queries.
CUDA_VISIBLE_DEVICES=0 python3 pipe_testUnitModelsOutputGen.py --execution_mode testAllQueryGen_1Q1A --img_type vit --output_len 85 --full_FT YES --user_msg rationale --epoch 50 --lr 5e-5 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
* Now, run showFinalResult-A.py to generate (print) the formatted outputs.
###
* For evaluation of the DatasetB-Confounder datset, use utils_dataTestUnitModels_confounder.py
* Stage 1: Kindly use the following command to generate the set of queries.
CUDA_VISIBLE_DEVICES=0 python3 pipe_testUnitModelsOutputGen_confounder.py --execution_mode testAllQueryGen --img_type vit --output_len 128 --full_FT YES --user_msg rationale --epoch 50 --lr 5e-5 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
* Stage 2: Kindly use the following command to generate the set of queries.
CUDA_VISIBLE_DEVICES=0 python3 pipe_testUnitModelsOutputGen_confounder.py --execution_mode testAllQueryGen_1Q1A --img_type vit --output_len 85 --full_FT YES --user_msg rationale --epoch 50 --lr 5e-5 --use_caption --use_generate --prompt_format QCM-E --output_dir experiments --final_eval
* Now, run showFinalResult-B.py to generate (print) the formatted outputs.