| license: apache-2.0 | |
| library_name: onnx | |
| pipeline_tag: text-ranking | |
| base_model: | |
| - cross-encoder/ms-marco-MiniLM-L2-v2 | |
| - cross-encoder/ms-marco-MiniLM-L4-v2 | |
| - cross-encoder/ms-marco-MiniLM-L6-v2 | |
| - cross-encoder/ms-marco-MiniLM-L12-v2 | |
| # Marlowe memory rerankers | |
| > **Not intended for individual download.** These files are fetched automatically by the | |
| > marlowe-cowork harness: its first launch offers to install them, and it checks every file against a | |
| > sha256 pinned in its code. They are tuned for its memory pipeline and are not a general-purpose | |
| > reranker. | |
| Fine-tuned MS MARCO MiniLM cross-encoders (ONNX) that marlowe-cowork uses to rerank recalled memories. | |
| | Folder | Use | | |
| |---|---| | |
| | `ms-marco-MiniLM-L-2-v2-ft-session-j` | the shipped reranker (CPU) | | |
| | `ms-marco-MiniLM-L-6-v2-ft-session-j`, `ms-marco-MiniLM-L-{2,4,6,12}-v2-ft-w1` | the GPU rerank cascade (CUDA builds) | | |
| Each folder holds `model.onnx` and the shared `tokenizer.json`. The embedder is the upstream | |
| `jinaai/jina-embeddings-v2-small-en`. | |