Hugging Face's logo

alexandersdata
/
gemma-4-E4B-hebrew-first

Transformers
Safetensors
Generated from Trainer
hf_jobs
sft
trl
Model card Files
xet
 Community
gemma-4-E4B-hebrew-first

Commit History

  1. Model save

    ssdata analysis
    Verified signature
    e938a9d

  1. Fix: add assistant_only_loss=False to prevent all labels being masked to -100

    ssdata analysis
    Verified signature
    76bf694

  2. Training in progress, step 500

    ssdata analysis
    Verified signature
    b1581a4

  3. Add adapter resume support: loads previous adapter from hub if available, 48h timeout ready

    ssdata analysis
    Verified signature
    12dd270

  4. Training in progress, step 500

    ssdata analysis
    Verified signature
    a122a25

  5. Optimal config: high-quality Hebrew data, constant LR, packing disabled, lora_dropout=0.1

    ssdata analysis
    Verified signature
    b88a92b

  6. Switch to packing=True, batch=4, grad_acc=4, step-based checkpoints for speed

    ssdata analysis
    Verified signature
    38eb889

  1. Disable eval to fix OOM on A10G

    ssdata analysis
    Verified signature
    779c4ca

  2. initial commit

    ssdata analysis
    Verified signature
    3ea3a00