-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy patheval_time.sh
More file actions
37 lines (30 loc) · 1.11 KB
/
eval_time.sh
File metadata and controls
37 lines (30 loc) · 1.11 KB
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
set -x
PAPER_TABLE=coco2017_cap_val
LOG_DIR=./logs_final
RUN_NAME=divprune_llava_1.5_7b_time
CUDA_VISIBLE_DEVICES=0,1,2,3 BASELINE=OURS LAYER_INDEX=0 SUBSET_RATIO=0.098 EVAL_TIME=TRUE python3 -m accelerate.commands.launch \
--num_processes=4 \
-m lmms_eval \
--model llava \
--model_args pretrained="liuhaotian/llava-v1.6-vicuna-7b" \
--tasks $PAPER_TABLE \
--batch_size 1 \
--log_samples \
--log_samples_suffix $RUN_NAME \
--output_path $LOG_DIR/$RUN_NAME | tee ./log_divprune.log
RUN_NAME=orig_llava_1.5_7b_time
CUDA_VISIBLE_DEVICES=0,1,2,3 BASELINE=ORIG EVAL_TIME=TRUE python3 -m accelerate.commands.launch \
--num_processes=4 \
-m lmms_eval \
--model llava \
--model_args pretrained="liuhaotian/llava-v1.6-vicuna-7b" \
--tasks $PAPER_TABLE \
--batch_size 1 \
--log_samples \
--log_samples_suffix $RUN_NAME \
--output_path $LOG_DIR/$RUN_NAME | tee ./log_orig.log
echo "===== Summary ====="
echo "Dviprune memory/latency"
python3 ./extract_time.py --path ./log_divprune.log
echo "Original model memory/latency"
python3 ./extract_time.py --path ./log_orig.log