horizon-len can not be modified when finetuneing

[richardzhangy26]

peft/finetine.sh

it is finetine.sh, i want to change horizon-len to finetine the dataset.

export TF_CPP_MIN_LOG_LEVEL=2 XLA_PYTHON_CLIENT_PREALLOCATE=false

python3 finetune.py \
    --model-name="google/timesfm-1.0-200m" \
    --backend="cpu" \
    --horizon-len=512 \
    --context-len=1440 \
    --freq="15min" \
    --data-path="../datasets/ETT-small/ETTm2.csv" \
    --num-epochs=100 \
    --learning-rate=1e-3 \
    --adam-epsilon=1e-7 \
    --adam-clip-threshold=1e2 \
    --early-stop-patience=10 \
    --datetime-col="date" \
    --use-lora \
    --lora-rank=2 \
    --lora-target-modules="all" \
    --use-dora \
    --cos-initial-decay-value=1e-4 \
    --cos-decay-steps=40000 \
    --cos-final-decay-value=1e-5 \
    --ema-decay=0.9999

And i got an error about TypeError: sub got incompatible shapes for broadcasting: (112, 128), (112, 512). how to fix it.

Whether horizon-len is fixed when finetuneing.