Scriptwriter dataset sync

restart-train.sh

@@ -7,8 +7,8 @@ tmux kill-session -t scriptwriter-train 2>/dev/null || true
 tmux kill-server 2>/dev/null || true
 pkill -9 -f 'train_runpod.py' 2>/dev/null || true
 pkill -9 -f 'scripts/train_runpod' 2>/dev/null || true
+pkill -9 -f torchrun 2>/dev/null || true
 sleep 2
-pgrep -af 'train_runpod' >/dev/null 2>&1 && pkill -9 -f train_runpod || true
 echo "Old session cleared."
 
 DEST=/workspace/scriptwriter-runpod
@@ -18,6 +18,9 @@ rm -rf "$DEST"
 mkdir -p "$DEST"
 tar xzf "$TGZ" -C "$DEST" --strip-components=1 --no-same-owner --no-same-permissions
 
+NPROC="$(python3 -c 'import torch; print(torch.cuda.device_count())' 2>/dev/null || echo 2)"
+if [[ "${NPROC}" -lt 1 ]]; then NPROC=1; fi
+
 export WORKDIR="$DEST"
 export DATA_DIR=/workspace/data/processed
 export OUTPUT_DIR=/workspace/models/lora
@@ -25,13 +28,15 @@ export HF_DATASET_REPO="${HF_DATASET_REPO:-datamatters24/scriptwriter-corpus-ia}
 export HF_MODEL_REPO="${HF_MODEL_REPO:-datamatters24/scriptwriter-lora-ia}"
 export BASE_MODEL="${BASE_MODEL:-meta-llama/Llama-3.2-3B-Instruct}"
 export RUNPOD_SKIP_FORCE_PIP=1
-export CUDA_VISIBLE_DEVICES=0
+unset CUDA_VISIBLE_DEVICES || true
 export MAX_SEQ_LENGTH=2048
 export MAX_SEQ_LENGTH_CAP=2048
 export PER_DEVICE_TRAIN_BATCH_SIZE=1
-export GRADIENT_ACCUMULATION_STEPS=8
+export GRADIENT_ACCUMULATION_STEPS=4
 export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
-export RESUME_FROM_CHECKPOINT=1
+# Fresh DDP run — old 1-GPU checkpoints won't load cleanly across 2 ranks
+export RESUME_FROM_CHECKPOINT=0
+export NPROC_PER_NODE="${NPROC}"
 
 if [[ ! -f /workspace/data/processed/train.jsonl ]]; then
   echo "train.jsonl missing — use fix-and-train.sh instead"
@@ -46,33 +51,33 @@ ln -sfn /workspace/data/processed "$DEST/data/processed"
 apt-get update -qq && DEBIAN_FRONTEND=noninteractive apt-get install -y -qq tmux >/dev/null || true
 
 WORKER=/workspace/logs/_train_worker.sh
-cat > "$WORKER" <<'INNER'
+cat > "$WORKER" <<INNER
 #!/bin/bash
 set -euo pipefail
 cd /workspace/scriptwriter-runpod
-export HF_TOKEN="${HF_TOKEN:-}"
-export HUGGING_FACE_HUB_TOKEN="${HF_TOKEN:-}"
-export HF_HOME="${HF_HOME:-/workspace/.cache/huggingface}"
+export HF_TOKEN="\${HF_TOKEN:-}"
+export HUGGING_FACE_HUB_TOKEN="\${HF_TOKEN:-}"
+export HF_HOME="\${HF_HOME:-/workspace/.cache/huggingface}"
 export CONFIG_PATH=/workspace/scriptwriter-runpod/config/training.yaml
-export BASE_MODEL="${BASE_MODEL:-meta-llama/Llama-3.2-3B-Instruct}"
+export BASE_MODEL="\${BASE_MODEL:-meta-llama/Llama-3.2-3B-Instruct}"
 export OUTPUT_DIR=/workspace/models/lora
-export HF_MODEL_REPO="${HF_MODEL_REPO:-datamatters24/scriptwriter-lora-ia}"
-export CUDA_VISIBLE_DEVICES=0
+export HF_MODEL_REPO="\${HF_MODEL_REPO:-datamatters24/scriptwriter-lora-ia}"
+unset CUDA_VISIBLE_DEVICES || true
 export MAX_SEQ_LENGTH=2048
 export MAX_SEQ_LENGTH_CAP=2048
 export PER_DEVICE_TRAIN_BATCH_SIZE=1
-export GRADIENT_ACCUMULATION_STEPS=8
+export GRADIENT_ACCUMULATION_STEPS=4
 export PYTORCH_CUDA_ALLOC_CONF=expandable_segments:True
-export RESUME_FROM_CHECKPOINT=1
+export RESUME_FROM_CHECKPOINT=0
+NPROC=${NPROC}
 exec > >(tee -a /workspace/logs/train_live.log) 2>&1
-echo "========== TRAIN START $(date -Is) =========="
-echo "OOM-safe: seq=2048 batch=1 accum=8 CUDA_VISIBLE_DEVICES=0"
-ls /workspace/models/lora/checkpoint-* 2>/dev/null | tail -5 || echo "no prior checkpoint"
+echo "========== TRAIN START \$(date -Is) =========="
+echo "DDP on \${NPROC} GPUs: seq=2048 batch=1/gpu accum=4"
 nvidia-smi || true
-python3 scripts/train_runpod.py
-echo "========== UPLOAD $(date -Is) =========="
-python3 scripts/sync_hf.py upload-model --repo "$HF_MODEL_REPO" --folder /workspace/models/lora
-echo "========== DONE $(date -Is) =========="
+torchrun --standalone --nproc_per_node=\${NPROC} scripts/train_runpod.py
+echo "========== UPLOAD \$(date -Is) =========="
+python3 scripts/sync_hf.py upload-model --repo "\$HF_MODEL_REPO" --folder /workspace/models/lora
+echo "========== DONE \$(date -Is) =========="
 echo "STOP THE POD"
 INNER
 chmod +x "$WORKER"
@@ -80,5 +85,5 @@ chmod +x "$WORKER"
 tmux new-session -d -s scriptwriter-train -n train "bash $WORKER"
 tmux new-window -t scriptwriter-train -n monitor
 tmux send-keys -t scriptwriter-train:monitor "watch -n 15 'nvidia-smi --query-gpu=index,memory.used,utilization.gpu --format=csv; echo; tail -25 /workspace/logs/train_live.log'" Enter
-echo "Started tmux scriptwriter-train — attach: tmux attach -t scriptwriter-train"
-echo "OOM fix: 2048 tokens, batch 1, gradient checkpointing; resumes checkpoint-100 if present"
+echo "Started tmux scriptwriter-train on ${NPROC} GPU(s) via torchrun DDP"
+echo "Attach: tmux attach -t scriptwriter-train"

scriptwriter-runpod-ready.tgz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e5d2813d7c694fd3e3a48474e5660c76faf114720a5be44eecd4e92afe723e28
-size 10140
+oid sha256:0c65297a24a5afce18c60ab0d70d718c3d3b220f39efe64798ed9ea011d3b638
+size 10317