amd-quark/internal-testing-qwen3_0.6b-fp8-tuned-orthogonal

Generated with:

CUDA_VISIBLE_DEVICES=5 python train_rotation.py \
    --model_dir /models/Qwen_Qwen3-0.6B \
    --quant_scheme fp8 \
    --eval_batch_size 16 \
    --num_samples 4000 \
    --learning_rate 1.5 \
    --rotation_algo_config_file ${ROTATION_TRAIN_CFG} \
    --output_dir qwen3-0.6b-fp8-tuned-orthogonal \
    --max_steps 5 \
    --loss_type kl_top_1000 \
    --model_attn_implementation sdpa \
    --train_batch_size 8 \
    --model_export hf_format \
    --export_weight_format real_quantized