Hanrui / progress /github /SpecForge /examples /run_gemma3_1b_eagle3_online.sh
Lekr0's picture
Add files using upload-large-folder tool
212a146 verified
SCRIPT_DIR=$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )
ROOT_DIR=$(dirname $SCRIPT_DIR)
export TORCHINDUCTOR_CACHE_DIR=$ROOT_DIR/cache/compiled_kernels
# train eagle3 for gemma3-1b
NUM_GPUS=${1:-1}
TP_SIZE=${2:-1}
torchrun \
--standalone \
--nproc_per_node $NUM_GPUS \
$ROOT_DIR/scripts/train_eagle3.py \
--target-model-path google/gemma-3-1b-it \
--draft-model-config $ROOT_DIR/configs/gemma3-1b-eagle3.json \
--train-data-path $ROOT_DIR/cache/dataset/sharegpt_train.jsonl \
--output-dir $ROOT_DIR/outputs/gemma3-1b-eagle3-sharegpt \
--num-epochs 10 \
--batch-size 1 \
--tp-size $TP_SIZE \
--learning-rate 1e-4 \
--max-length 4096 \
--chat-template gemma \
--cache-dir $ROOT_DIR/cache \
--attention-backend sdpa \
--target-model-backend hf \
--log-interval 10