Safetensors
qwen2
max_seq_length=16384
lr=2e-5
batch_size=1
grad_accum=16
epochs=3
qlora
quantize=4bit_nf4
lora_rank=64
lora_alpha=128
lora_dropout=0.05
completion_only_loss
eval_size=0.1
cosine_schedule
warmup=0.05
bf16
run_name=sft-deepseek-game24
hub_model_id=tinyllms/deepseek-r1-distill-qwen-7b-sft-game24
ray_job=raysubmit_tYQfn6Lm6DKqkMXU
