# Modifications from the upstream model

`ettin-150m-memory-reranker-ft-v1` is not an unmodified copy of
`cross-encoder/ettin-reranker-150m-v1` (revision
`025501c4e0f9bbeb4c5b198318e0089ff061cc14`). It is a Daecore-modified
derivative and is not endorsed by the upstream authors, by Johns Hopkins
University, whose `jhu-clsp/ettin-encoder-150m` the upstream reranker builds
on, or by Hugging Face.

What changed:

- **Fine-tune.** The base weights were trained further for fixed top-50
  memory reranking using graded ListNet and adjacent-grade RankNet supervision
  with rank-16 LoRA+. The adapter was merged into the model weights.
  Relevance grades were frontier-model judgments under a frozen protocol,
  not human annotations.
- **Export.** The merged model was exported to ONNX graphs taking
  `input_ids` and `attention_mask` and emitting one score per query-passage
  pair. CUDA uses FP16. DirectML widens the same stored weights and floating
  calculations to FP32, and uses compatible inferred reshapes. Both graphs
  share one external parameter file; the CUDA graph and weights are unchanged.
- **Serving contracts.** `serving.json` and `serving.directml.json` use the
  same descriptor format for each graph: 1,153 total pair tokens and batch
  ceilings of 16 on CUDA and 8 on DirectML. CPU handles permitted control
  operations, not a fallback execution of the whole reranker.

Files modified or generated by Daecore:

- `model.onnx`: the generated ONNX graph of the fine-tuned reranker;
- `model.directml.onnx`: the FP32 calculation graph for DirectML;
- `model.onnx.data`: the fine-tuned parameters shared by both graphs;
- `config.json`: the export configuration of the modified reranker;
- `serving.json` and `serving.directml.json`: the runtime and sequence contracts.

Unchanged: `tokenizer.json` and `tokenizer_config.json` preserve the upstream
tokenizer contract.

This distribution is subject to the Apache License 2.0 (`LICENSE`).