划重点: 1)SPLADEv1 (SIGIR 21’) -> SPLADEv2、SPLADEv2.1(SIGIR22’)是搜索召回方向的系列工作。搜索引擎因为索引量大,需要高效的检索方式。传统的搜索召回是通过query分词之后查询倒排索引(inverted Inde…
SPLADE模型的名字, 有效的值有 naver/splade-cocondenser-ensembledistil (默认), naver/splade_v2_max, naver/splade_v2_distil, naver/splade-cocondenser-selfdistil. 更多请参考:GitHub - naver/splade: SPLADE: sparse neural search (SIGIR21, SIGIR22) device (string)要使用的设备,对于CPU使用cpu,...
We provide several base configurations which correspond to the experiments in the v2bis and "efficiency" papers. Please note that these are suited for our hardware setting, i.e. 4 GPUs Tesla V100 with 32GB memory. In order to train models with e.g. one GPU, you need to decrease the bat...
SPLADE v2 network takes at most 512 tokens, but most stories tokenize into 30000+ tokens. For a document that is larger than 512 tokens, the network is given 512 tokens at a time and the resulting vectors are combined using proximity pooling. The network will never see the start and the...
naver/splade_v2_distil (v2 HF) 36.8 naver/splade-cocondenser-selfdistil (SPLADE++, HF) 37.6 naver/splade-cocondenser-ensembledistil (SPLADE++, HF) 38.3 naver/efficient-splade-V-large-doc (HF) + naver/efficient-splade-V-large-query (HF) (efficient SPLADE) 38.8 naver/efficient-splade...