Hugging Face's logo

2ira
/
Byte-lingua-code

Model card Files
xet
 Community
Byte-lingua-code
File size: 8,247 Bytes
72c0672
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
#!/bin/bash

DATA=${1:-"ocp_subsampled_5G"}

if [[ $DATA == "ocp_subsampled_5G" ]]; then
  input_dir="ocp_subsampled_5G"
  splits_dir="ocp_subsampled_5G_entropy87_splits"
  compression_dir="ocp_subsampled_5G_entropy87_enumerative_compression"

  #### Step 1: setup subsampled data for evaluation
  if [[ ! -d "$input_dir" ]]; then
    mkdir $input_dir
    head -n 1050000 /mnt/hdfs/user/linzheng/data/opencoder_python/opencoder_python.chunk.1.jsonl > temp.jsonl
    split -n r/8 --suffix-length=1 --numeric-suffixes=1 --additional-suffix=.jsonl temp.jsonl ${input_dir}/ocp.chunk.
    rm temp.jsonl
  else
    echo "Directory '$input_dir' already exists."
  fi

elif [[ $DATA == "opencoder_subsampled_5G" ]]; then
  input_dir="opencoder_subsampled_5G"
  splits_dir="opencoder_subsampled_5G_entropy87_splits"
  compression_dir="opencoder_subsampled_5G_entropy87_enumerative_compression"

  #### Step 1: setup subsampled data for evaluation
  if [[ ! -d "$input_dir" ]]; then
    mkdir $input_dir
    head -n 1050000 /mnt/hdfs/user/linzheng/data/opencoder/chunk.1.jsonl > temp.jsonl
    split -n r/8 --suffix-length=1 --numeric-suffixes=1 --additional-suffix=.jsonl temp.jsonl ${input_dir}/ocp.chunk.
    rm temp.jsonl
  else
    echo "Directory '$input_dir' already exists."
  fi 

elif [[ $DATA == "ocp_subsampled_50G" ]]; then
  input_dir="/mnt/hdfs/user/linzheng/data/ocpython_subsampled_50G"
  splits_dir="ocpython_subsampled_50G_entropy87_splits"
  compression_dir="ocpython_subsampled_50G_entropy87_enumerative_compression"

else
  echo "Unknown $DATA."
  exit 0
fi

entropy_model_path=/mnt/bn/tiktok-mm-5/aiic/users/linzheng/artifacts/m1_checkpoints/m1_40M_lr1e-3_steps200k_bs8_seqlen2048_full/checkpoints/0000200000
compression_model_path=/mnt/bn/tiktok-mm-5/aiic/users/linzheng/artifacts/m1_checkpoints/m1_40M_lr1e-3_steps200k_bs8_seqlen2048_full/checkpoints/0000200000

NUM_GPUS=4
total_jsonls=8
total_jobs=1

wait 

# for JSONL_IDX in $(seq 1 4); do
#     for index in $(seq 0 $((total_jobs - 1))); do
#         echo "Starting job $index..."

#         GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
#         CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
#             --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
#             --output_dir "ocpython_subsampled_50G_entropy90_splits_linesplit" \
#             --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
#             --data_batch_size 256 --max_entropy_batch_size 8192 \
#             --num_workers 1 --process_id $index --num_processes $total_jobs \
#             --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
#             --apply_line_split > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_lines.log 2>&1 &
#     done
# done
# wait
# for JSONL_IDX in $(seq 5 $total_jsonls); do
#     for index in $(seq 0 $((total_jobs - 1))); do
#         echo "Starting job $index..."

#         GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
#         CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
#             --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
#             --output_dir "ocpython_subsampled_50G_entropy90_splits_linesplit" \
#             --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
#             --data_batch_size 256 --max_entropy_batch_size 8192 \
#             --num_workers 1 --process_id $index --num_processes $total_jobs \
#             --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
#             --apply_line_split > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_lines.log 2>&1 &
#     done
# done

##############################################################
##############################################################

# wait 

# for JSONL_IDX in $(seq 1 4); do
#     for index in $(seq 0 $((total_jobs - 1))); do
#         echo "Starting job $index..."

#         GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
#         CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
#             --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
#             --output_dir "ocpython_subsampled_50G_entropy90_splits_chunk256" \
#             --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
#             --data_batch_size 256 --max_entropy_batch_size 4096 \
#             --num_workers 1 --process_id $index --num_processes $total_jobs \
#             --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
#             --chunk_size 256 > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_chunk256.log 2>&1 &
#     done
# done
# wait
# for JSONL_IDX in $(seq 5 $total_jsonls); do
#     for index in $(seq 0 $((total_jobs - 1))); do
#         echo "Starting job $index..."

#         GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
#         CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
#             --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
#             --output_dir "ocpython_subsampled_50G_entropy90_splits_chunk256" \
#             --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
#             --data_batch_size 256 --max_entropy_batch_size 4096 \
#             --num_workers 1 --process_id $index --num_processes $total_jobs \
#             --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
#             --chunk_size 256 > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_chunk256.log 2>&1 &
#     done
# done

##############################################################
##############################################################

wait 

for JSONL_IDX in $(seq 1 4); do
    for index in $(seq 0 $((total_jobs - 1))); do
        echo "Starting job $index..."

        GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
        CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
            --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
            --output_dir "ocpython_subsampled_50G_entropy90_splits_chunk512" \
            --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
            --data_batch_size 256 --max_entropy_batch_size 2048 \
            --num_workers 1 --process_id $index --num_processes $total_jobs \
            --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
            --chunk_size 512 > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_chunk512.log 2>&1 &
    done
done
wait
for JSONL_IDX in $(seq 5 $total_jsonls); do
    for index in $(seq 0 $((total_jobs - 1))); do
        echo "Starting job $index..."

        GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
        CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_split.py \
            --input_file ${input_dir}/ocp.chunk.${JSONL_IDX}.jsonl \
            --output_dir "ocpython_subsampled_50G_entropy90_splits_chunk512" \
            --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
            --data_batch_size 256 --max_entropy_batch_size 2048 \
            --num_workers 1 --process_id $index --num_processes $total_jobs \
            --base_global_quantile 0.90 --base_monotonic_quantile 0.90 \
            --chunk_size 512 > split_jsonl${JSONL_IDX}_process${index}_total${total_jobs}_chunk512.log 2>&1 &
    done
done

# wait

# NUM_GPUS=8
# total_jsonls=8
# total_jobs=3

# for JSONL_IDX in $(seq 1 $total_jsonls); do
#     for index in $(seq 0 $((total_jobs - 1))); do
#         echo "Starting job $index..."

#         GPU_IDX=$(( (JSONL_IDX - 1) % NUM_GPUS ))
#         CUDA_VISIBLE_DEVICES=${GPU_IDX} python3 offline_entropy_window_compress.py \
#             --input_file ${splits_dir}/ocp.chunk.${JSONL_IDX}_out_0.jsonl \
#             --output_dir $compression_dir \
#             --entropy_model_path $entropy_model_path --compression_model_path $compression_model_path \
#             --data_batch_size 1024 --max_compression_batch_size 3072 \
#             --num_workers 1 --process_id $index --num_processes $total_jobs > compress_jsonl${JSONL_IDX}_process${index}_total${total_jobs}.log 2>&1 &
#     done
# done