Rebuild a half-present Python install; run r2 from scratch
Browse files- control/run.env +5 -9
- scripts/pipeline.sh +6 -1
control/run.env
CHANGED
|
@@ -1,13 +1,9 @@
|
|
| 1 |
# KEY=VALUE settings passed to scripts/pipeline.sh by scripts/supervise.py
|
| 2 |
-
#
|
| 3 |
-
#
|
|
|
|
| 4 |
GEN_TOKENS=4000000
|
| 5 |
-
GEN_MINUTES=
|
| 6 |
-
|
| 7 |
-
EXTRACT_MINUTES=25
|
| 8 |
-
EXTRACT_APPEND=1
|
| 9 |
TRAIN_MINUTES=40
|
| 10 |
-
RUN_NAME=main2
|
| 11 |
-
TRAIN_INIT=runs/main/sub_bf16
|
| 12 |
EVAL_ALT=0
|
| 13 |
-
TRAIN_INIT_HF=mudassiruddin21/EMC:submissions/r1/main-bf16
|
|
|
|
| 1 |
# KEY=VALUE settings passed to scripts/pipeline.sh by scripts/supervise.py
|
| 2 |
+
# run r2 on a fresh machine: everything from scratch with twice the data of r1 (4M generated
|
| 3 |
+
# tokens), training from the baseline head; the bf16 variant is not evaluated
|
| 4 |
+
RUN_ID_FORCE=r2
|
| 5 |
GEN_TOKENS=4000000
|
| 6 |
+
GEN_MINUTES=55
|
| 7 |
+
EXTRACT_MINUTES=35
|
|
|
|
|
|
|
| 8 |
TRAIN_MINUTES=40
|
|
|
|
|
|
|
| 9 |
EVAL_ALT=0
|
|
|
scripts/pipeline.sh
CHANGED
|
@@ -25,6 +25,7 @@ export TORCHINDUCTOR_CACHE_DIR="$WORK/cache/inductor" TRITON_CACHE_DIR="$WORK/ca
|
|
| 25 |
# greedy decoding only: use the built-in sampler instead of compiling FlashInfer kernels at start-up
|
| 26 |
export VLLM_USE_FLASHINFER_SAMPLER=0
|
| 27 |
[ -f "$STATE/run_id" ] || echo "${RUN_ID:-$(date +%Y%m%d-%H%M)}" > "$STATE/run_id"
|
|
|
|
| 28 |
# training run name (runs/<name>), its starting weights and whether to also evaluate the bf16 variant
|
| 29 |
RUN_NAME="${RUN_NAME:-main}"; TRAIN_INIT="${TRAIN_INIT:-$BASELINE}"; EVAL_ALT="${EVAL_ALT:-1}"
|
| 30 |
case "$TRAIN_INIT" in /*) ;; *) TRAIN_INIT="$EMC_ROOT/$TRAIN_INIT" ;; esac
|
|
@@ -64,7 +65,11 @@ kit() { # the official kit commands (scripts/kit.py only maps its container pat
|
|
| 64 |
s_env() {
|
| 65 |
nvidia-smi || true
|
| 66 |
df -h "$EMC_ROOT" || true; free -g || true; nproc || true
|
| 67 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 68 |
"$UV" pip install --python "$PY" "vllm==0.31.0" accelerate datasets
|
| 69 |
"$UV" cache clean >/dev/null 2>&1 || true
|
| 70 |
"$PY" - <<'EOF'
|
|
|
|
| 25 |
# greedy decoding only: use the built-in sampler instead of compiling FlashInfer kernels at start-up
|
| 26 |
export VLLM_USE_FLASHINFER_SAMPLER=0
|
| 27 |
[ -f "$STATE/run_id" ] || echo "${RUN_ID:-$(date +%Y%m%d-%H%M)}" > "$STATE/run_id"
|
| 28 |
+
if [ -n "${RUN_ID_FORCE:-}" ]; then echo "$RUN_ID_FORCE" > "$STATE/run_id"; fi # e.g. a new run on a fresh machine
|
| 29 |
# training run name (runs/<name>), its starting weights and whether to also evaluate the bf16 variant
|
| 30 |
RUN_NAME="${RUN_NAME:-main}"; TRAIN_INIT="${TRAIN_INIT:-$BASELINE}"; EVAL_ALT="${EVAL_ALT:-1}"
|
| 31 |
case "$TRAIN_INIT" in /*) ;; *) TRAIN_INIT="$EMC_ROOT/$TRAIN_INIT" ;; esac
|
|
|
|
| 65 |
s_env() {
|
| 66 |
nvidia-smi || true
|
| 67 |
df -h "$EMC_ROOT" || true; free -g || true; nproc || true
|
| 68 |
+
# a Python install or environment left half-present by a machine reset is rebuilt
|
| 69 |
+
for d in "$UV_PYTHON_INSTALL_DIR"/cpython-3.12*; do
|
| 70 |
+
if [ -d "$d" ] && [ ! -x "$d/bin/python3.12" ]; then echo "removing incomplete $d"; rm -rf "$UV_PYTHON_INSTALL_DIR"; break; fi
|
| 71 |
+
done
|
| 72 |
+
if ! "$PY" -c "import sys" >/dev/null 2>&1; then rm -rf "$VENV"; "$UV" venv --python 3.12 "$VENV"; fi
|
| 73 |
"$UV" pip install --python "$PY" "vllm==0.31.0" accelerate datasets
|
| 74 |
"$UV" cache clean >/dev/null 2>&1 || true
|
| 75 |
"$PY" - <<'EOF'
|