Killed Training Command | Generated by AI
Killed. Here’s the command to paste into tmux:
cd /root/nanochat && source .venv/bin/activate && \
export OMP_NUM_THREADS=1 && \
export NANOCHAT_BASE_DIR="$HOME/.cache/nanochat" && \
export WANDB_MODE=disabled && \
export HIP_FORCE_DEV_KERNARG=1 && \
export HSA_OVERRIDE_GFX_VERSION=9.4.2 && \
export PYTORCH_ALLOC_CONF=expandable_segments:True && \
python -m scripts.base_train \
--depth=24 \
--device-batch-size=32 \
--total-batch-size=524288 \
--max-seq-len=2048 \
--window-pattern SSSL \
--num-iterations=29000 \
--eval-every=1000 \
--eval-tokens=1048576 \
--sample-every=5000 \
--save-every=5000 \
--core-metric-every=5000 \
--core-metric-max-per-task=200 \
--tracker=mlflow \
--fp8 \
--run=mi300x-d24-760m-fa2-fp8 \
2>&1 | tee /root/nanochat/run_mi300x_d24_fa2_fp8.log
If it crashes, resume with:
--resume-from-step=<LAST_STEP> \
(check the log for last completed step number)
Stats: ~99,500 tok/sec, 36% MFU, ~5.3s/step, ~42.5 hours total.