Delete test_long_niah.sh
Browse files- test_long_niah.sh +0 -77
test_long_niah.sh
DELETED
@@ -1,77 +0,0 @@
|
|
1 |
-
PARTITION=${PARTITION:-"Intern5"}
|
2 |
-
GPUS=${GPUS:-16}
|
3 |
-
GPUS_PER_NODE=${GPUS_PER_NODE:-8}
|
4 |
-
GPUS_PER_TASK=${GPUS_PER_TASK:-1}
|
5 |
-
QUOTA_TYPE=${QUOTA_TYPE:-"reserved"}
|
6 |
-
|
7 |
-
|
8 |
-
set -x
|
9 |
-
|
10 |
-
CHECKPOINT=${1}
|
11 |
-
|
12 |
-
JOB_FOLDER=$(dirname "$CHECKPOINT")
|
13 |
-
files=(
|
14 |
-
"$JOB_FOLDER/configuration_intern_vit.py"
|
15 |
-
"$JOB_FOLDER/configuration_internlm2.py"
|
16 |
-
"$JOB_FOLDER/configuration_internvl_chat.py"
|
17 |
-
"$JOB_FOLDER/conversation.py"
|
18 |
-
"$JOB_FOLDER/modeling_intern_vit.py"
|
19 |
-
"$JOB_FOLDER/modeling_internlm2.py"
|
20 |
-
"$JOB_FOLDER/modeling_internvl_chat.py"
|
21 |
-
"$JOB_FOLDER/tokenization_internlm2_fast.py"
|
22 |
-
"$JOB_FOLDER/tokenization_internlm2.py"
|
23 |
-
"test_long_niah.sh"
|
24 |
-
)
|
25 |
-
for file in "${files[@]}"; do
|
26 |
-
dest_file="$CHECKPOINT/$(basename "$file")"
|
27 |
-
if [ ! -f "$dest_file" ]; then
|
28 |
-
cp "$file" "$CHECKPOINT"
|
29 |
-
fi
|
30 |
-
done
|
31 |
-
ARGS=("$@")
|
32 |
-
|
33 |
-
declare -a tasks=( \
|
34 |
-
'retrieval-image-test-long-subset' \
|
35 |
-
'retrieval-image-test-long-128k' \
|
36 |
-
)
|
37 |
-
|
38 |
-
# declare -a tasks=(\
|
39 |
-
# 'retrieval-image-test-long-1M'\
|
40 |
-
# )
|
41 |
-
declare -a tasks=(\
|
42 |
-
'retrieval-image-test-long-800k' \
|
43 |
-
'retrieval-image-test-long-512k' \
|
44 |
-
)
|
45 |
-
|
46 |
-
model_name="internvl"
|
47 |
-
for STRIDE in 1; do
|
48 |
-
LOG_DIR=$CHECKPOINT/eval_mm_niah_ring_attn_$STRIDE
|
49 |
-
mkdir -p $LOG_DIR
|
50 |
-
|
51 |
-
for ((j=0; j<${#tasks[@]}; j++)); do
|
52 |
-
task=${tasks[j]}
|
53 |
-
|
54 |
-
echo "$(date) ${model_name}_${task}_stride_${STRIDE}"
|
55 |
-
|
56 |
-
srun -p ${PARTITION} \
|
57 |
-
--gres=gpu:${GPUS_PER_NODE} \
|
58 |
-
--ntasks=$((GPUS / GPUS_PER_TASK)) \
|
59 |
-
--ntasks-per-node=$((GPUS_PER_NODE / GPUS_PER_TASK)) \
|
60 |
-
--quotatype=${QUOTA_TYPE} \
|
61 |
-
--job-name="${STRIDE}${task}" \
|
62 |
-
-o "${LOG_DIR}/${task}_stride_${STRIDE}.log" \
|
63 |
-
-e "${LOG_DIR}/${task}_stride_${STRIDE}.log" \
|
64 |
-
--async \
|
65 |
-
python -u eval/mm_niah/eval_mm_niah_long.py \
|
66 |
-
--checkpoint $CHECKPOINT \
|
67 |
-
--outputs-dir $LOG_DIR \
|
68 |
-
--task $task \
|
69 |
-
--num-gpus-per-rank ${GPUS_PER_TASK} "${ARGS[@]:1}" \
|
70 |
-
--rope_pos_id_version 'v5' \
|
71 |
-
--ring_attn \
|
72 |
-
--rope_pos_id_stride $STRIDE \
|
73 |
-
|
74 |
-
|
75 |
-
sleep 0.2
|
76 |
-
done
|
77 |
-
done
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|