qwen2-moe-a3b-2layer / generation_config.json
ELutris's picture
KD-distilled 2-layer Qwen2-MoE from hyper-accel/ci-random-qwen2-moe-a3b against Qwen1.5-MoE-A2.7B teacher (alpaca-cleaned, 1500 steps, T=2.0)
ef12344 verified
Raw
History Blame Contribute Delete
120 Bytes
{
"_from_model_config": true,
"bos_token_id": 151643,
"eos_token_id": 151643,
"transformers_version": "5.5.0"
}