### 0
pkill -9 sglang; sleep 3; ray stop
  --force; pkill -9 ray; pkill -9 miles; sleep 3; pkill -9 ray; pkill -9 miles; pkill -9 redis; true; 

### 1
export PYTHONUNBUFFERED=1 && ray start
  --head
  --node-ip-address 127.0.0.1
  --num-gpus 8
  --disable-usage-stats

### 2
ray status 2>/dev/null || true

### 3
sleep 5

### 4
ray status 2>/dev/null || true

### 5
sleep 5

### 6
ray status 2>/dev/null || true

### 7
sleep 5

### 8
ray status 2>/dev/null || true

### 9
sleep 5

### 10
ray status 2>/dev/null || true

### 11
sleep 5

### 12
ray status 2>/dev/null || true

### 13
sleep 5

### 14
ray status 2>/dev/null || true

### 15
sleep 5

### 16
ray status 2>/dev/null || true

### 17
sleep 5

### 18
ray status 2>/dev/null || true

### 19
sleep 5

### 20
ray status 2>/dev/null || true

### 21
sleep 5

### 22
ray status 2>/dev/null || true

### 23
sleep 5

### 24
ray status 2>/dev/null || true

### 25
sleep 5

### 26
ray status 2>/dev/null || true

### 27
sleep 5

### 28
ray status 2>/dev/null || true

### 29
sleep 5

### 30
ray status 2>/dev/null || true

### 31
sleep 5

### 32
ray status 2>/dev/null || true

### 33
sleep 5

### 34
ray status 2>/dev/null || true

### 35
sleep 5

### 36
ray status 2>/dev/null || true

### 37
sleep 5

### 38
ray status 2>/dev/null || true

### 39
sleep 5

### 40
ray status 2>/dev/null || true

### 41
sleep 5

### 42
ray status 2>/dev/null || true

### 43
sleep 5

### 44
ray status 2>/dev/null || true

### 45
sleep 5

### 46
ray status 2>/dev/null || true

### 47
sleep 5

### 48
ray status 2>/dev/null || true

### 49
sleep 5

### 50
ray status 2>/dev/null || true

### 51
sleep 5

### 52
ray status 2>/dev/null || true

### 53
sleep 5

### 54
ray status 2>/dev/null || true

### 55
sleep 5

### 56
ray status 2>/dev/null || true

### 57
sleep 5

### 58
ray status 2>/dev/null || true

### 59
sleep 5

### 60
ray status 2>/dev/null || true

### 61
sleep 5

### 62
ray status 2>/dev/null || true

### 63
sleep 5

### 64
ray status 2>/dev/null || true

### 65
sleep 5

### 66
ray status 2>/dev/null || true

### 67
sleep 5

### 68
ray status 2>/dev/null || true

### 69
sleep 5

### 70
ray status 2>/dev/null || true

### 71
sleep 5

### 72
ray status 2>/dev/null || true

### 73
sleep 5

### 74
ray status 2>/dev/null || true

### 75
sleep 5

### 76
ray status 2>/dev/null || true

### 77
sleep 5

### 78
ray status 2>/dev/null || true

### 79
sleep 5

### 80
ray status 2>/dev/null || true

### 81
sleep 5

### 82
ray status 2>/dev/null || true

### 83
sleep 5

### 84
ray status 2>/dev/null || true

### 85
sleep 5

### 86
ray status 2>/dev/null || true

### 87
sleep 5

### 88
ray status 2>/dev/null || true

### 89
sleep 5

### 90
ray status 2>/dev/null || true

### 91
sleep 5

### 92
ray status 2>/dev/null || true

### 93
sleep 5

### 94
ray status 2>/dev/null || true

### 95
sleep 5

### 96
ray status 2>/dev/null || true

### 97
sleep 5

### 98
ray status 2>/dev/null || true

### 99
sleep 5

### 100
ray status 2>/dev/null || true

### 101
sleep 5

### 102
ray status 2>/dev/null || true

### 103
sleep 5

### 104
ray status 2>/dev/null || true

### 105
sleep 5

### 106
ray status 2>/dev/null || true

### 107
sleep 5

### 108
ray status 2>/dev/null || true

### 109
sleep 5

### 110
ray status 2>/dev/null || true

### 111
sleep 5

### 112
ray status 2>/dev/null || true

### 113
sleep 5

### 114
ray status 2>/dev/null || true

### 115
sleep 5

### 116
ray status 2>/dev/null || true

### 117
sleep 5

### 118
ray status 2>/dev/null || true

### 119
sleep 5

### 120
ray status 2>/dev/null || true

### 121
sleep 5

### 122
ray status 2>/dev/null || true

### 123
sleep 5

### 124
ray status 2>/dev/null || true

### 125
sleep 5

### 126
ray status 2>/dev/null || true

### 127
sleep 5

### 128
ray status 2>/dev/null || true

### 129
sleep 5

### 130
ray status 2>/dev/null || true

### 131
sleep 5

### 132
ray status 2>/dev/null || true

### 133
sleep 5

### 134
ray status 2>/dev/null || true

### 135
sleep 5

### 136
ray status 2>/dev/null || true

### 137
sleep 5

### 138
ray status 2>/dev/null || true

### 139
sleep 5

### 140
ray status 2>/dev/null || true

### 141
sleep 5

### 142
ray status 2>/dev/null || true

### 143
sleep 5

### 144
ray status 2>/dev/null || true

### 145
sleep 5

### 146
ray status 2>/dev/null || true

### 147
sleep 5

### 148
ray status 2>/dev/null || true

### 149
sleep 5

### 150
ray status 2>/dev/null || true

### 151
sleep 5

### 152
ray status 2>/dev/null || true

### 153
sleep 5

### 154
ray status 2>/dev/null || true

### 155
sleep 5

### 156
ray status 2>/dev/null || true

### 157
sleep 5

### 158
ray status 2>/dev/null || true

### 159
sleep 5

### 160
ray status 2>/dev/null || true

### 161
sleep 5

### 162
ray status 2>/dev/null || true

### 163
sleep 5

### 164
ray status 2>/dev/null || true

### 165
sleep 5

### 166
ray status 2>/dev/null || true

### 167
sleep 5

### 168
ray status 2>/dev/null || true

### 169
sleep 5

### 170
ray status 2>/dev/null || true

### 171
sleep 5

### 172
ray status 2>/dev/null || true

### 173
sleep 5

### 174
ray status 2>/dev/null || true

### 175
sleep 5

### 176
ray status 2>/dev/null || true

### 177
sleep 5

### 178
ray status 2>/dev/null || true

### 179
sleep 5

### 180
ray status 2>/dev/null || true

### 181
sleep 5

### 182
ray status 2>/dev/null || true

### 183
sleep 5

### 184
ray status 2>/dev/null || true

### 185
sleep 5

### 186
ray status 2>/dev/null || true

### 187
sleep 5

### 188
ray status 2>/dev/null || true

### 189
sleep 5

### 190
ray status 2>/dev/null || true

### 191
sleep 5

### 192
ray status 2>/dev/null || true

### 193
sleep 5

### 194
ray status 2>/dev/null || true

### 195
sleep 5

### 196
ray status 2>/dev/null || true

### 197
sleep 5

### 198
ray status 2>/dev/null || true

### 199
sleep 5

### 200
ray status 2>/dev/null || true

### 201
sleep 5

### 202
ray status 2>/dev/null || true

### 203
sleep 5

### 204
ray status 2>/dev/null || true

### 205
sleep 5

### 206
ray status 2>/dev/null || true

### 207
sleep 5

### 208
ray status 2>/dev/null || true

### 209
sleep 5

### 210
ray status 2>/dev/null || true

### 211
sleep 5

### 212
ray status 2>/dev/null || true

### 213
sleep 5

### 214
ray status 2>/dev/null || true

### 215
sleep 5

### 216
ray status 2>/dev/null || true

### 217
sleep 5

### 218
ray status 2>/dev/null || true

### 219
sleep 5

### 220
ray status 2>/dev/null || true

### 221
sleep 5

### 222
ray status 2>/dev/null || true

### 223
sleep 5

### 224
ray status 2>/dev/null || true

### 225
sleep 5

### 226
ray status 2>/dev/null || true

### 227
sleep 5

### 228
ray status 2>/dev/null || true

### 229
sleep 5

### 230
ray status 2>/dev/null || true

### 231
sleep 5

### 232
ray status 2>/dev/null || true

### 233
sleep 5

### 234
ray status 2>/dev/null || true

### 235
sleep 5

### 236
ray status 2>/dev/null || true

### 237
sleep 5

### 238
ray status 2>/dev/null || true

### 239
sleep 5

### 240
ray status 2>/dev/null || true

### 241
sleep 5

### 242
ray status

### 243
nvidia-smi topo -m 2>/dev/null | grep -o 'NV[0-9][0-9]*' | wc -l

### 244
export no_proxy=127.0.0.1 && export PYTHONUNBUFFERED=1 && ray job submit
  --address="http://127.0.0.1:8265"
  --runtime-env-json='{"env_vars": {"PYTHONUNBUFFERED": "1", "CUDA_DEVICE_MAX_CONNECTIONS": "1", "NCCL_NVLS_ENABLE": "0", "no_proxy": "127.0.0.1,127.0.0.1", "MASTER_ADDR": "127.0.0.1", "PYTHONPATH": "<REPO_ROOT>:/root/Megatron-LM:/frozen/pythonpath"}}'
  -- python3 <REPO_ROOT>/train.py
  --disable-bias-linear
  --group-query-attention
  --num-attention-heads 32
  --num-query-groups 2
  --kv-channels 128
  --num-layers 88
  --hidden-size 4096
  --ffn-hidden-size 2688
  --normalization RMSNorm
  --position-embedding-type none
  --vocab-size 131072
  --make-vocab-size-divisible-by 128
  --untie-embeddings-and-output-weights
  --num-experts 512
  --moe-router-topk 22
  --moe-ffn-hidden-size 2688
  --moe-shared-expert-intermediate-size 5376
  --moe-latent-size 1024
  --moe-router-score-function sigmoid
  --moe-router-enable-expert-bias
  --moe-grouped-gemm
  --moe-router-dtype fp32
  --moe-router-num-groups 1
  --moe-router-group-topk 1
  --moe-router-topk-scaling-factor 5.0
  --moe-router-pre-softmax
  --moe-router-load-balancing-type seq_aux_loss
  --moe-router-bias-update-rate 0
  --moe-aux-loss-coeff 0
  --hf-checkpoint /root/models/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
  --ref-load /root/models/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
  --save /root/shared_data/checkpoints
  --save-interval 20
  --megatron-to-hf-mode bridge 
  --prompt-data /root/datasets/dapo-math-17k/dapo-math-17k.jsonl
  --input-key prompt
  --label-key label
  --apply-chat-template
  --rollout-shuffle
  --rm-type deepscaler
  --num-rollout 10
  --rollout-batch-size 32
  --n-samples-per-prompt 4
  --rollout-max-response-len 1024
  --rollout-temperature 1
  --global-batch-size 128
  --balance-data 
  --optimizer adam
  --lr 1e-6
  --lr-decay-style constant
  --weight-decay 0.1
  --adam-beta1 0.9
  --adam-beta2 0.98
  --optimizer-cpu-offload
  --overlap-cpu-optimizer-d2h-h2d
  --use-precision-aware-optimizer 
  --advantage-estimator grpo
  --use-kl-loss
  --kl-loss-coef 0.00
  --kl-loss-type low_var_kl
  --entropy-coef 0.00
  --eps-clip 0.2
  --eps-clip-high 0.28 
  --use-wandb
  --wandb-project miles-run_nemotron_3_super_120b_a12b
  --wandb-group 260101-000000-000
  --wandb-key 'frozen-wandb-api-key'
  --disable-wandb-random-suffix 
  --tensor-model-parallel-size 4
  --sequence-parallel
  --pipeline-model-parallel-size 2
  --context-parallel-size 1
  --expert-model-parallel-size 8
  --expert-tensor-parallel-size 1
  --recompute-granularity full
  --recompute-method uniform
  --recompute-num-layers 1
  --use-dynamic-batch-size
  --max-tokens-per-gpu 1024
  --log-probs-chunk-size 128 
  --rollout-num-gpus-per-engine 8
  --sglang-mem-fraction-static 0.7
  --use-rollout-routing-replay 
  --attention-dropout 0.0
  --hidden-dropout 0.0
  --accumulate-allreduce-grads-in-fp32
  --attention-softmax-in-fp32
  --attention-backend auto
  --colocate
  --actor-num-nodes 2
  --actor-num-gpus-per-node 8
  --num-gpus-per-node 8
  --rollout-num-gpus 16   
  --deploy-component all
