|
|
@@ -108,9 +108,9 @@ CUTE_DSL_ARCH = "sm_121a"
|
|
|
TORCH_CUDA_ARCH_LIST = "12.1a"
|
|
|
|
|
|
# ── JIT cache ──────────────────────────────────────────────────────────────
|
|
|
-# Uses launcher default: /root/.cache/huggingface
|
|
|
-# Set XDG_CACHE_HOME to override; mount a persistent volume to avoid
|
|
|
-# ~20 min recompilation on each cold boot.
|
|
|
+# Under ~/.cache so it persists inside the HF cache mount (-v ~/.cache/.../hf)
|
|
|
+# Avoids ~20 min recompilation on each cold boot.
|
|
|
+XDG_CACHE_HOME = "/root/.cache"
|
|
|
|
|
|
# ── NCCL overrides ─────────────────────────────────────────────────────────
|
|
|
# Cluster config provides NCCL_IB_HCA, NCCL_IB_GID_INDEX, NCCL_SOCKET_IFNAME
|