| { | |
| "model_type": "kenga-embed", | |
| "architectures": [ | |
| "KengaEmbedEnc" | |
| ], | |
| "d": 768, | |
| "layers": 8, | |
| "heads": 12, | |
| "dff": 3072, | |
| "seq": 160, | |
| "v": 256, | |
| "k_attn": 192, | |
| "k_ff": 512, | |
| "rank": [ | |
| 192, | |
| 512 | |
| ], | |
| "instruct": true, | |
| "instruct_retr": "Instruct: Given a query, retrieve relevant passages\nQuery: ", | |
| "instruct_sts": "Instruct: Retrieve semantically similar text.\nQuery: ", | |
| "params": 41900288, | |
| "step": 10400, | |
| "acc_en": 0.8333333333333334, | |
| "acc_ru": 1.0, | |
| "method": "prophet", | |
| "note": "42M UTF-8 byte bidirectional encoder. Not Giga-480M. Not ruMTEB SOTA." | |
| } |