← Back to Architecture

TrigramHash

Architecture
Used in
31 PRs
Best BPB
0.9850
Avg BPB
1.1658

Hyperparameters Across PRs

pr_numberparameters
292{"buckets":4096,"dim":32}
327{"buckets":8192,"dim":64}
344{"size":4096,"dim":128}
418{"buckets":2048,"dimensions":64}
440{"vocab_size":2048,"dim":48}
486{"buckets":4096,"dim":128}
562
571{"variants":[{"buckets":2048,"embed_dim":64},{"buckets":4096,"embed_dim":96}]}
635{"buckets":4096,"dim":128}
882{"buckets":8192,"n_gram":3}
884{"vocab_size":2048,"trigram_dim":48,"project_dim":512}
1089{"heads":2,"buckets":8192}
1098{"size":1024}
1117{"size":1024}
1118{"dimensions":1024}
1169{"heads":2,"buckets":8192}
1182{"buckets":1024,"dimensions":128}
1186{"order":3}
1200{"vocab_size":4096,"dimensions":128}
1201{"vocab_size":4096,"dimension":128}
1311{"enabled":false}
1370
1384{"buckets":65000}
1440{"buckets":3072,"heads":2}
1501
1544
1545
1553
1602
1632
1749