castorini/first_qwen3_14b

Reranking top 100 retrieval results of DL tracks with SPLADE++ EnsemBleDistil as first stage retriever.

nDCG@10	DL19	DL20	DL21	DL22	DL23	Average
First Stage
SPLADE++_EnsemBleDistil_ONNX	0.7308	0.7197	0.6851	0.5698	0.4727	0.6356
Zero Shot
Qwen3-0.6B think	0.7084	0.7047	0.6952	0.5723	0.4844	0.6330
Qwen3-0.6B no think	0.5816	0.5276	0.5971	0.4546	0.3833	0.5088
Qwen3-1.7B think	0.7377	0.7054	0.6849	0.5884	0.4896	0.6412
Qwen3-1.7B no think	0.7037	0.6144	0.6572	0.5455	0.3974	0.5836
Qwen3-4B think	0.7620	0.7544	0.7382	0.6210	0.5381	0.6827
Qwen3-4B no think	0.7613	0.7266	0.7259	0.6282	0.5535	0.6791
Qwen3-8B think	0.7499	0.7496	0.7539	0.6676	0.5959	0.7034
Qwen3-8B no think	0.7740	0.7585	0.7469	0.6471	0.5612	0.6975
Qwen3-14B think	0.7403	0.7638	0.7530	0.6854	0.6150	0.7115
Qwen3-14B no think	0.7745	0.7740	0.7600	0.6931	0.6090	0.7221
Qwen3-32B think	0.7458	0.7692	0.7510	0.6898	0.6337	0.7179
Qwen3-32B no think	0.7656	0.7680	0.7642	0.7113	0.5950	0.7208
Finetuned
FirstQwen3-0.6B	0.7764	0.7720	0.7263	0.6665	0.5579	0.6998
FirstQwen3-1.7B	0.7663	0.7934	0.7542	0.6986	0.5762	0.7177
FirstQwen3-4B	0.7699	0.7893	0.7524	0.7007	0.6184	0.7261
FirstQwen3-8B	0.7746	0.8104	0.7555	0.7156	0.6398	0.7392
FirstQwen3-14B	0.7654	0.8062	0.7718	0.7094	0.6083	0.7322