Skip to content

Commit 42440c6

Browse files
Dando18Daniel Nichols
Dando18
and
Daniel Nichols
authored
Updates and Optimizations to Generation Scripts (#9)
* updates and optimizations to generation scripts --------- Co-authored-by: Daniel Nichols <dnicho@login01.chn>
1 parent 4c9276c commit 42440c6

28 files changed

Lines changed: 147717 additions & 54 deletions

‎.gitignore‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,6 +4,7 @@
44
# python
55
*.pyc
66
__pycache__
7+
.env
78

89
# cpp
910
*.out

‎generate.py‎

Lines changed: 0 additions & 54 deletions
This file was deleted.

‎generate/generate-codellama-13b-prompted.sbatch‎

Lines changed: 40 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,40 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 4
4+
#SBATCH --mem=64000
5+
#SBATCH -t 23:59:59
6+
#SBATCH -p gpu
7+
#SBATCH --gpus=a100:1
8+
#SBATCH -J generate-codellama-13b-prompted
9+
#SBATCH -o generate-codellama-13b-prompted-%A.out
10+
11+
# settings
12+
MODEL="codellama/CodeLlama-13b-hf"
13+
TEMP=0.2
14+
TOPP=0.95
15+
MAX_NEW_TKNS=1024
16+
SAMPLES_PER_PROMPT=20
17+
BATCH_SIZE=4
18+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
19+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_prompted_temp${TEMP}.json"
20+
echo "Writing to $OUTPUT"
21+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
22+
23+
# setup
24+
ml cuda/11.8.0
25+
source .env/bin/activate
26+
export HF_HOME=~/scratch/.cache/huggingface
27+
export OMP_NUM_THREADS=4
28+
29+
# generate
30+
srun python generate.py \
31+
--model $MODEL \
32+
--prompts ../prompts/generation-prompts.json \
33+
--output $OUTPUT \
34+
--temperature $TEMP \
35+
--top_p $TOPP \
36+
--do_sample \
37+
--max_new_tokens $MAX_NEW_TKNS \
38+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
39+
--batch_size $BATCH_SIZE \
40+
--prompted

‎generate/generate-codellama-13b.sbatch‎

Lines changed: 39 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,39 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 4
4+
#SBATCH --mem=64000
5+
#SBATCH -t 23:59:59
6+
#SBATCH -p gpu
7+
#SBATCH --gpus=a100:1
8+
#SBATCH -J generate-codellama-13b-bs4
9+
#SBATCH -o generate-codellama-13b-bs4-%A.out
10+
11+
# settings
12+
MODEL="codellama/CodeLlama-13b-hf"
13+
TEMP=0.2
14+
TOPP=0.95
15+
MAX_NEW_TKNS=1024
16+
SAMPLES_PER_PROMPT=20
17+
BATCH_SIZE=4
18+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
19+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_temp${TEMP}.json"
20+
echo "Writing to $OUTPUT"
21+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
22+
23+
# setup
24+
ml cuda/11.8.0
25+
source .env/bin/activate
26+
export HF_HOME=~/scratch/.cache/huggingface
27+
export OMP_NUM_THREADS=4
28+
29+
# generate
30+
srun python generate.py \
31+
--model $MODEL \
32+
--prompts ../prompts/generation-prompts.json \
33+
--output $OUTPUT \
34+
--temperature $TEMP \
35+
--top_p $TOPP \
36+
--do_sample \
37+
--max_new_tokens $MAX_NEW_TKNS \
38+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
39+
--batch_size $BATCH_SIZE

‎generate/generate-codellama-34b-prompted.sbatch‎

Lines changed: 44 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,44 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 8
4+
#SBATCH --ntasks-per-node=1
5+
#SBATCH --gpus-per-task=1
6+
#SBATCH --mem=128000
7+
#SBATCH -t 23:59:59
8+
#SBATCH -A m2404
9+
#SBATCH -C gpu&hbm80g
10+
#SBATCH -q regular
11+
#SBATCH -J generate-codellama-34b-prompted
12+
#SBATCH -o generate-codellama-34b-prompted-%A.out
13+
14+
# settings
15+
MODEL="codellama/CodeLlama-34b-hf"
16+
TEMP=0.2
17+
TOPP=0.95
18+
MAX_NEW_TKNS=1024
19+
SAMPLES_PER_PROMPT=20
20+
BATCH_SIZE=8
21+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
22+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_prompted_temp${TEMP}.json"
23+
echo "Writing to $OUTPUT"
24+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
25+
26+
# setup
27+
ml cuda/11.8.0
28+
source .env/bin/activate
29+
export HF_HOME=/pscratch/sd/d/dnicho/.cache/huggingface
30+
export OMP_NUM_THREADS=8
31+
#export SLURM_CPU_BIND="cores"
32+
33+
# generate
34+
srun python generate.py \
35+
--model $MODEL \
36+
--prompts ../prompts/generation-prompts.json \
37+
--output $OUTPUT \
38+
--temperature $TEMP \
39+
--top_p $TOPP \
40+
--do_sample \
41+
--max_new_tokens $MAX_NEW_TKNS \
42+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
43+
--batch_size $BATCH_SIZE \
44+
--prompted

‎generate/generate-codellama-34b.sbatch‎

Lines changed: 43 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,43 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 8
4+
#SBATCH --ntasks-per-node=1
5+
#SBATCH --gpus-per-task=1
6+
#SBATCH --mem=128000
7+
#SBATCH -t 23:59:59
8+
#SBATCH -A m2404
9+
#SBATCH -C gpu&hbm80g
10+
#SBATCH -q regular
11+
#SBATCH -J generate-codellama-34b-bs8
12+
#SBATCH -o generate-codellama-34b-bs8-%A.out
13+
14+
# settings
15+
MODEL="codellama/CodeLlama-34b-hf"
16+
TEMP=0.2
17+
TOPP=0.95
18+
MAX_NEW_TKNS=1024
19+
SAMPLES_PER_PROMPT=20
20+
BATCH_SIZE=8
21+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
22+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_temp${TEMP}.json"
23+
echo "Writing to $OUTPUT"
24+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
25+
26+
# setup
27+
ml cuda/11.8.0
28+
source .env/bin/activate
29+
export HF_HOME=/pscratch/sd/d/dnicho/.cache/huggingface
30+
export OMP_NUM_THREADS=8
31+
#export SLURM_CPU_BIND="cores"
32+
33+
# generate
34+
srun python generate.py \
35+
--model $MODEL \
36+
--prompts ../prompts/generation-prompts.json \
37+
--output $OUTPUT \
38+
--temperature $TEMP \
39+
--top_p $TOPP \
40+
--do_sample \
41+
--max_new_tokens $MAX_NEW_TKNS \
42+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
43+
--batch_size $BATCH_SIZE

‎generate/generate-codellama-7b-prompted.sbatch‎

Lines changed: 40 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,40 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 4
4+
#SBATCH --mem=64000
5+
#SBATCH -t 23:59:59
6+
#SBATCH -p gpu
7+
#SBATCH --gpus=a100:1
8+
#SBATCH -J generate-codellama-7b-prompted
9+
#SBATCH -o generate-codellama-7b-prompted-%A.out
10+
11+
# settings
12+
MODEL="codellama/CodeLlama-7b-hf"
13+
TEMP=0.2
14+
TOPP=0.95
15+
MAX_NEW_TKNS=1024
16+
SAMPLES_PER_PROMPT=20
17+
BATCH_SIZE=8
18+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
19+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_prompted_temp${TEMP}.json"
20+
echo "Writing to $OUTPUT"
21+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
22+
23+
# setup
24+
ml cuda/11.8.0
25+
source .env/bin/activate
26+
export HF_HOME=~/scratch/.cache/huggingface
27+
export OMP_NUM_THREADS=4
28+
29+
# generate
30+
srun python generate.py \
31+
--model $MODEL \
32+
--prompts ../prompts/generation-prompts.json \
33+
--output $OUTPUT \
34+
--temperature $TEMP \
35+
--top_p $TOPP \
36+
--do_sample \
37+
--max_new_tokens $MAX_NEW_TKNS \
38+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
39+
--batch_size $BATCH_SIZE \
40+
--prompted

‎generate/generate-codellama-7b.sbatch‎

Lines changed: 39 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,39 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 4
4+
#SBATCH --mem=64000
5+
#SBATCH -t 23:59:59
6+
#SBATCH -p gpu
7+
#SBATCH --gpus=a100:1
8+
#SBATCH -J generate-codellama-7b-bs8
9+
#SBATCH -o generate-codellama-7b-bs8-%A.out
10+
11+
# settings
12+
MODEL="codellama/CodeLlama-7b-hf"
13+
TEMP=0.2
14+
TOPP=0.95
15+
MAX_NEW_TKNS=1024
16+
SAMPLES_PER_PROMPT=20
17+
BATCH_SIZE=8
18+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
19+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_temp${TEMP}.json"
20+
echo "Writing to $OUTPUT"
21+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
22+
23+
# setup
24+
ml cuda/11.8.0
25+
source .env/bin/activate
26+
export HF_HOME=~/scratch/.cache/huggingface
27+
export OMP_NUM_THREADS=4
28+
29+
# generate
30+
srun python generate.py \
31+
--model $MODEL \
32+
--prompts ../prompts/generation-prompts.json \
33+
--output $OUTPUT \
34+
--temperature $TEMP \
35+
--top_p $TOPP \
36+
--do_sample \
37+
--max_new_tokens $MAX_NEW_TKNS \
38+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
39+
--batch_size $BATCH_SIZE

‎generate/generate-phind-v2-prompted.sbatch‎

Lines changed: 44 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,44 @@
1+
#!/bin/bash
2+
#SBATCH -n 1
3+
#SBATCH -c 8
4+
#SBATCH --ntasks-per-node=1
5+
#SBATCH --gpus-per-task=1
6+
#SBATCH --mem=128000
7+
#SBATCH -t 23:59:59
8+
#SBATCH -A m2404
9+
#SBATCH -C gpu&hbm80g
10+
#SBATCH -q regular
11+
#SBATCH -J generate-phind-v2-prompted
12+
#SBATCH -o generate-phind-v2-prompted-%A.out
13+
14+
# settings
15+
MODEL="Phind/Phind-CodeLlama-34B-v2"
16+
TEMP=0.2
17+
TOPP=0.95
18+
MAX_NEW_TKNS=1024
19+
SAMPLES_PER_PROMPT=20
20+
BATCH_SIZE=8
21+
hash=$(md5sum ../prompts/generation-prompts.json | cut -d' ' -f1)
22+
OUTPUT="../outputs/output_${hash:0:8}_${MODEL//\//--}_prompted_temp${TEMP}.json"
23+
echo "Writing to $OUTPUT"
24+
echo "model=$MODEL MAX_NEW_TKNS=$MAX_NEW_TKNS SAMPLES_PER_PROMPT=$SAMPLES_PER_PROMPT BATCH_SIZE=$BATCH_SIZE"
25+
26+
# setup
27+
ml cuda/11.8.0
28+
source .env/bin/activate
29+
export HF_HOME=/pscratch/sd/d/dnicho/.cache/huggingface
30+
export OMP_NUM_THREADS=8
31+
#export SLURM_CPU_BIND="cores"
32+
33+
# generate
34+
srun python generate.py \
35+
--model $MODEL \
36+
--prompts ../prompts/generation-prompts.json \
37+
--output $OUTPUT \
38+
--temperature $TEMP \
39+
--top_p $TOPP \
40+
--do_sample \
41+
--max_new_tokens $MAX_NEW_TKNS \
42+
--num_samples_per_prompt $SAMPLES_PER_PROMPT \
43+
--batch_size $BATCH_SIZE \
44+
--prompted

0 commit comments

Comments
 (0)