Skip to content

slime troubleshooting

Known gotchas when training with the slime backend.

For reproducibility, here’s the exact environment this integration was validated against:

ComponentVersion / SHA
Instance type8 × NVIDIA H100 80GB HBM3
CUDA13.0
PyTorch2.11.0+cu130
slimecommit fa3c990af6f18efd3fd9922698bf4bf4048d1263
SGLang0.5.13
Megatron-LMcommit 1dcf0dafa884ad52ffb243625717a3471643e087

--norm-epsilon mismatch on Qwen2.5-32B-Instruct

Section titled “--norm-epsilon mismatch on Qwen2.5-32B-Instruct”

Cause: slime’s scripts/models/qwen2.5-32B.sh hardcodes --norm-epsilon 1e-5 (matching Qwen2.5-32B base), but Qwen2.5-32B-Instruct uses 1e-6.

Fix: Edit the slime model script to --norm-epsilon 1e-6, or pass an override through SlimeRunner(extra_flags=["--norm-epsilon", "1e-6"]). Qwen2.5-0.5B / 1.5B / 3B / 7B Instruct variants match their base-model norm epsilons, so this only affects Qwen2.5-32B.