Hugging Face's logo

jasonfan
/
FST_code

Safetensors
Model card Files
xet
 Community
FST_code
File size: 4,678 Bytes
3b2d368
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
#!/bin/bash

# Optional arguments [MODEL_NAME] [NUM_GPUS] [NUM_NODES]
NUM_GPUS=${1:-4}                # default 4 GPUs per node
NUM_NODES=${2:-1}               # default 1 node

# Clear first 2 args to avoid hydra issues
shift 2

# CPUs and memory automatically determined 
CPUS_PER_TASK=8                             # 8 CPUs per GPU
TOTAL_CPUS=$((NUM_GPUS * CPUS_PER_TASK))
TOTAL_MEM=$((NUM_GPUS * 32))                # 32gb ram per GPU

# Extract model name from hydra overrides
MODEL_NAME="transformer"
for arg in "$@"; do
  if [[ "$arg" == model=* ]]; then
    MODEL_NAME="${arg#model=}"
    break
  fi
done

JOB_NAME="${MODEL_NAME}_train_medium"

DATE=$(date +%F)

echo "Launching training job: $JOB_NAME"
echo "Nodes: ${NUM_NODES}, GPUs per node: ${NUM_GPUS}"
echo "Resources per node: ${TOTAL_CPUS} CPUs, ${TOTAL_MEM} GB memory"

SBATCH_ARGS="--job-name=${JOB_NAME} \
             --nodes=${NUM_NODES} \
             --gres=gpu:h200:${NUM_GPUS} \
             --ntasks-per-node=${NUM_GPUS} \
             --cpus-per-task=${CPUS_PER_TASK} \
             --mem=${TOTAL_MEM}G \
             --output=../../logs/${MODEL_NAME}/${DATE}/%x-%j.out \
             --error=../../logs/${MODEL_NAME}/${DATE}/%x-%j.err"

# bash  /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_1_3b.sh \
    # size=1_3b \
    # training=1_3b \
    # dataset=tinygsm_resume \
    # "$@"

# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=transformer_1_3B


# cp /work/jf381/code/lm-research/main_1_2.py /work/jf381/code/lm-research/main.py
# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M_bert.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm \
#     "$@" \
#     model=transformer_353M_bert
bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M_bert.sh \
    size=medium \
    training=medium \
    dataset=slimpajama_6b \
    training.lr=2e-6 \
    "$@" \
    model=transformer_353M_bert

bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M_bert1.sh \
    size=medium \
    training=medium \
    dataset=slimpajama_6b \
    training.lr=2e-6 \
    "$@" \
    model=transformer_353M_bert
  
bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M_bert2.sh \
    size=medium \
    training=medium \
    dataset=slimpajama_6b \
    training.lr=2e-6 \
    "$@" \
    model=transformer_353M_bert
# bert_pretrain
# ork/jf381/code/lm-research/main_1_2.py /work/jf381/code/lm-research/main.py
# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M_bert.sh \
#     size=medium \
#     training=medium \
#     dataset=slimpajama_6b \
#     "$@" \
#     model=transformer_353M_bert

# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_1_3B.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=transformer_1_3B

# cp /work/jf381/code/lm-research/main_1_7.py /work/jf381/code/lm-research/main.py
# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_1_3B.sh \
#     size=1_3b \
#     training=1_3b \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=fst_1_3B


# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_353M.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=fst_353M

# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_353M.sh \
#     size=1_3b \
#     training=1_3b \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=fst_1_3B

# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_353M.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm_resume \
#     "$@" \
#     model=transformer_353M
# bash  /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_1_3b.sh \
#     size=1_3b \
#     training=1_3b \
#     dataset=tinygsm_resume \
#     "$@"

# bash  /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_generate.sh \
#     size=medium \
#     training=medium \
#     dataset=tinygsm_resume \
#     "$@"
  ## left_small 02-
  ## 04 big-
  ## 03 fst big

# # cat << 'EOF' > 1_3b_transformer.txt
# bash /work/jf381/code/lm-research/scripts/jobs/run_job_h200_bash_resume_transformer_1_3b_generate.sh \
#     size=1_3b \
#     training=1_3b \
#     dataset=tinygsm_resume \
#     "$@"
# EOF