### 0
mkdir -p /root/models /root/datasets

### 1
hf download moonshotai/Kimi-K2.5
  --local-dir /root/models/Kimi-K2.5

### 2
hf download
  --repo-type dataset zhuzilin/dapo-math-17k
  --local-dir /root/datasets/dapo-math-17k

### 3
python <REPO_ROOT>/tools/convert_kimi_int4_to_bf16.py
  --model-dir /root/models/Kimi-K2.5
  --output-dir /root/models/Kimi-K2.5-bf16 

### 4
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; 

### 5
export PYTHONUNBUFFERED=1 && ray start
  --head
  --node-ip-address 127.0.0.1
  --num-gpus 8
  --disable-usage-stats

### 6
nvidia-smi topo -m 2>/dev/null | grep -o 'NV[0-9][0-9]*' | wc -l

### 7
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", "NCCL_TIMEOUT": "3600", "OPEN_TRAINING_INT4_FAKE_QAT_FLAG": "1", "OPEN_TRAINING_INT4_GROUP_SIZE": "32", "PYTHONPATH": "<REPO_ROOT>:/root/Megatron-LM:/frozen/pythonpath"}}'
  -- python3 <REPO_ROOT>/train.py
  --disable-bias-linear
  --num-layers 61
  --hidden-size 7168
  --ffn-hidden-size 18432
  --num-attention-heads 64
  --kv-channels 64
  --normalization RMSNorm
  --position-embedding-type rope
  --rope-type yarn
  --norm-epsilon 1e-5
  --swiglu
  --untie-embeddings-and-output-weights
  --vocab-size 163840
  --multi-latent-attention
  --q-lora-rank 1536
  --kv-lora-rank 512
  --qk-head-dim 128
  --qk-pos-emb-head-dim 64
  --v-head-dim 128
  --qk-layernorm
  --rotary-scaling-factor 64.0
  --rotary-base 50000
  --original-max-position-embeddings 4096
  --beta-fast 32
  --beta-slow 1
  --mscale 1.0
  --mscale-all-dim 1.0
  --attention-softmax-in-fp32
  --no-rope-fusion
  --num-experts 384
  --moe-layer-freq '[0,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1]'
  --moe-ffn-hidden-size 2048
  --moe-router-topk 8
  --moe-shared-expert-intermediate-size 2048
  --moe-router-pre-softmax
  --moe-router-score-function sigmoid
  --moe-router-enable-expert-bias
  --moe-router-load-balancing-type seq_aux_loss
  --moe-token-dispatcher-type alltoall
  --moe-aux-loss-coeff 0
  --moe-router-bias-update-rate 0
  --moe-router-group-topk 1
  --moe-router-num-groups 1
  --moe-grouped-gemm
  --moe-router-topk-scaling-factor 2.827
  --moe-router-dtype fp32
  --moe-permute-fusion
  --hf-checkpoint /root/models/Kimi-K2.5
  --ref-load /root/models/Kimi-K2.5-bf16
  --megatron-to-hf-mode bridge
  --model-name kimi_k25
  --load /root/shared_data/260101-000000-000/checkpoints
  --save /root/shared_data/260101-000000-000/checkpoints
  --save-interval 2 
  --prompt-data /root/datasets/dapo-math-17k/dapo-math-17k.jsonl
  --input-key prompt
  --label-key label
  --apply-chat-template
  --rollout-shuffle
  --balance-data
  --rm-type deepscaler
  --num-rollout 3000
  --rollout-batch-size 32
  --n-samples-per-prompt 8
  --rollout-max-response-len 100
  --rollout-temperature 1
  --global-batch-size 256
  --use-dynamic-global-batch-size 
  --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
  --use-distributed-optimizer 
  --advantage-estimator grpo
  --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_kimi_k25
  --wandb-group 260101-000000-000
  --wandb-key 'frozen-wandb-api-key'
  --disable-wandb-random-suffix 
  --tensor-model-parallel-size 8
  --sequence-parallel
  --pipeline-model-parallel-size 1
  --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 2048  
  --rollout-num-gpus-per-engine 8
  --sglang-mem-fraction-static 0.7
  --sglang-ep-size 8
  --sglang-server-concurrency 1024
  --sglang-cuda-graph-bs 1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128
  --use-rollout-routing-replay 
  --attention-dropout 0.0
  --hidden-dropout 0.0
  --accumulate-allreduce-grads-in-fp32
  --attention-softmax-in-fp32
  --attention-backend flash
  --no-check-for-nan-in-loss-and-grad
  --colocate
  --update-weight-buffer-size 2147483648
  --actor-num-nodes 1
  --actor-num-gpus-per-node 8
  --num-gpus-per-node 8   
