gemma4-warmstart-test / train_command.txt
shanjiaz's picture
Upload folder using huggingface_hub
cdd3f4d verified
Raw
History Blame Contribute Delete
837 Bytes
# Timestamp: 2026-07-16T17:27:53.351860+00:00
# Git SHA: 0d28bc7f504925c3d47127f75cb6624d372d1f52
# World size: 4
# speculators: 0.7.0.dev107
# vllm: 0.25.1
# transformers: 5.13.1
# torch: 2.11.0
# compressed-tensors: 0.17.0
scripts/train.py --verifier-name-or-path google/gemma-4-31B-it --speculator-type dspark --from-pretrained RedHatAI/gemma-4-31B-it-speculator.dflash --data-path ./output/dspark_gemma4_31b_warmstart --vllm-endpoint http://localhost:8000/v1 --save-path ./output/dspark_gemma4_31b_warmstart/checkpoints --epochs 5 --lr 1e-4 --total-seq-len 16000 --max-anchors 512 --markov-rank 256 --markov-head-type vanilla --enable-confidence-head --confidence-head-with-markov --confidence-head-alpha 1.0 --loss-fn '{"ce": 0.1, "tv": 0.9}' --on-missing generate --no-sample-from-anchor --on-generate delete --checkpoint-freq 0.1