Initial release

Browse files

Files changed (9) hide show

.gitattributes +36 -0
README.md +187 -0
base_vs_ft.png +0 -0
config.json +48 -0
llama_bidirectional_model.py +362 -0
model.safetensors +3 -0
stack.png +0 -0
tokenizer.json +3 -0
tokenizer_config.json +15 -0

.gitattributes ADDED Viewed

	@@ -0,0 +1,36 @@

+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz filter=lfs diff=lfs merge=lfs -text
+*.zip filter=lfs diff=lfs merge=lfs -text
+*.zst filter=lfs diff=lfs merge=lfs -text
+*tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md ADDED Viewed

	@@ -0,0 +1,187 @@

+---
+language:
+- el
+- en
+license: other
+license_name: nvidia-open-model-license
+base_model: nvidia/llama-nemotron-rerank-1b-v2
+pipeline_tag: text-ranking
+library_name: transformers
+tags:
+- greek
+- english
+- reranker
+- cross-encoder
+- rag
+- nemotron
+model-index:
+- name: Sophea-Nemo-Reranker
+  results:
+  - task:
+      type: text-ranking
+      name: Greek+English reranking (5 domains, rerank top-50)
+    dataset:
+      type: retrieval
+      name: Internal Greek+English retrieval benchmark
+    metrics:
+    - type: nDCG@10
+      name: nDCG@10 (mean of 5 domains)
+      value: 0.8172
+    - type: recall@10
+      name: Recall@10 (mean of 5 domains)
+      value: 0.9267
+---
+<div style="display:flex;align-items:center;gap:18px;padding:16px 20px;margin-bottom:16px;border-radius:14px;background:linear-gradient(100deg,#e8f5ec 0%,#f7fbf8 62%);border:1px solid #c6e7d1"><img alt="KIEFERSA" style="flex:0 0 auto;height:58px;width:58px;border-radius:12px;object-fit:contain" src="data:image/webp;base64,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"/><div><div style="font-size:1.24rem;font-weight:700;color:#003713;line-height:1.2">Sophea-Nemo-Reranker</div><div style="font-size:.86rem;color:#3a5a47;margin-top:2px">Greek + English cross-encoder reranker — fine-tuned NVIDIA Llama-Nemotron-Rerank-1B-v2</div></div></div>
+**Sophea-Nemo-Reranker** is a Greek + English **cross-encoder reranker**, fine-tuned from
+**[NVIDIA llama-nemotron-rerank-1b-v2](https://huggingface.co/nvidia/llama-nemotron-rerank-1b-v2)** for
+Greek and English reranking. It is the second (precision) stage of a
+two-stage Greek RAG stack; the matching first stage is
+**[Sophea-Nemo-Embedding](https://huggingface.co/KIEFERSA/Sophea-Nemo-Embedding)**.
+- **Creator:** Kiefer SA
+- **Base model:** `nvidia/llama-nemotron-rerank-1b-v2` (`LlamaBidirectionalForSequenceClassification`,
+  **bidirectional**, single relevance logit; `trust_remote_code`)
+- **Languages:** Greek + English
+- **Input template:** a single sequence — `question:{q} \n \n passage:{p}`
+- **Training:** BCE on labeled (query, doc) pairs, 1 epoch, max-len 4096, bf16
+> **Greek is out-of-distribution for the base.** The base reranker lists 26 languages and **Greek is
+> not one of them** — this fine-tune adapts its trained relevance head to Greek.
+## Why it matters — base vs fine-tuned
+Reranking the fixed top-50 candidates over 5 domains (nDCG@10), the fine-tune improves on the base
+Nemotron reranker **on every domain**:
+![Greek+English reranking — base vs fine-tuned](base_vs_ft.png)
+<div style="overflow-x:auto">
+<table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
+<thead><tr>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">metric</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
+</tr></thead><tbody>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">nDCG@10 — base</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.793</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.741</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.966</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.702</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.651</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.770</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">nDCG@10 — fine-tuned</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.874</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.780</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.969</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.754</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.709</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.817</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Δ nDCG@10</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.081</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.039</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.004</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.052</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.058</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.047</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Recall@10 — base</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.953</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.847</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.993</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.853</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.773</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.884</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Recall@10 — fine-tuned</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.993</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.873</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.987</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.907</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.873</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.927</td></tr>
+</tbody></table></div>
+**Read.** Mean **nDCG@10 0.770 → 0.817
+(+0.047)**, better on all five domains, with the
+largest gains on the hardest ones (medical +0.06, finance +0.05, val +0.08). In the two-stage stack it
+is also the strongest reranker we measured — edging a fine-tuned Qwen3-0.6B cross-encoder and beating
+the base by a clear margin.
+## Size comparison — vs base Qwen3-Reranker (off-the-shelf)
+Reranking the same fixed top-50 candidates, same protocol (150 queries/domain), nDCG@10 — your
+**fine-tuned 1B** reranker **edges the base Qwen3-Reranker-8B** and clearly beats 4B / 0.6B, at a
+fraction of the size:
+<div style="overflow-x:auto">
+<table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
+<thead><tr>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">model</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
+</tr></thead><tbody>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Sophea-Nemo-Reranker (ours · 1B · fine-tuned)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.874</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.780</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.969</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.754</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.709</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.817</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Reranker-8B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.858</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.796</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.975</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.763</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.672</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.813</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Reranker-4B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.865</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.782</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.966</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.735</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.647</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.799</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Qwen3-Reranker-0.6B (base, off-the-shelf)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.812</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.746</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.960</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.692</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.634</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.769</td></tr>
+</tbody></table></div>
+**Read.** Mean nDCG@10 **0.817** (ours, 1B) vs **0.813**
+(Qwen3-Reranker-8B), **0.799** (4B), **0.769** (0.6B) — and
+ours is strongest on the hardest domain (medical 0.709 vs 8B's 0.672).
+A small, Greek-tuned model beats a general reranker 8× its size.
+## End-to-end two-stage pipeline
+The two models are designed to run **together**: the embedder retrieves the top-50 candidates, the
+reranker re-orders them. Measured end-to-end on the same 5-domain Greek+English eval (150 queries/set,
+correct `query:`/`passage:` prompts):
+![End-to-end two-stage stack — embedder alone vs + reranker](stack.png)
+<div style="overflow-x:auto">
+<table style="width:100%;border-collapse:collapse;font-size:13px;font-variant-numeric:tabular-nums;font-family:-apple-system,BlinkMacSystemFont,'Segoe UI',Roboto,sans-serif">
+<thead><tr>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:left;">metric</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Val</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Energy</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Legal</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Finance</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Medical</th>
+<th style="padding:10px 8px;font-weight:500;border-bottom:2px solid #0a8043;color:#0a8043;font-size:14px;text-align:center;">Mean</th>
+</tr></thead><tbody>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">nDCG@10 — embedder alone</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.863</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.802</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.939</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.702</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.689</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">0.799</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">nDCG@10 — + reranker (stack)</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.867</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.780</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.976</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.763</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.703</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.818</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);">Δ from reranking</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.004</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">-0.022</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.037</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.061</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.013</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);">+0.019</td></tr>
+<tr><td style="padding:7px 8px;padding-left:20px;text-align:left;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">Recall@10 — full stack</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.987</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.873</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.993</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.920</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.860</td><td style="padding:7px 8px;text-align:center;border-bottom:1px solid rgba(128,128,128,.15);font-weight:700;color:#0a8043;">0.927</td></tr>
+</tbody></table></div>
+**Read.** The full stack reaches **mean nDCG@10 0.818** (Recall@10 0.927) — the
+strongest configuration measured. The reranker adds the most where the embedder is weakest
+(**finance +0.061, legal +0.037**), i.e. it sharpens precision on the hard
+cases. On **energy** it slightly regresses (-0.022) and on `val` the gain is marginal,
+because the embedder alone is already very strong there and leaves little headroom — for the easiest,
+already-high-recall domains you can skip the rerank stage if latency matters.
+## Usage (transformers)
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForSequenceClassification
+name = "KIEFERSA/Sophea-Nemo-Reranker"
+tok = AutoTokenizer.from_pretrained(name, trust_remote_code=True)
+model = AutoModelForSequenceClassification.from_pretrained(
+    name, num_labels=1, dtype=torch.bfloat16, trust_remote_code=True).eval()
+def score(query, passage):
+    text = f"question:{query} \n \n passage:{passage}"
+    enc = tok(text, return_tensors="pt", truncation=True, max_length=4096)
+    with torch.no_grad():
+        return model(**enc).logits.squeeze(-1).item()   # higher = more relevant
+# rank a candidate list by descending score
+```
+Feed the base's own `question:/passage:` template — it has a **trained** relevance head, so matching
+its template adapts that head rather than fighting it. Needs `trust_remote_code=True`.
+## Serving with vLLM
+```bash
+vllm serve KIEFERSA/Sophea-Nemo-Reranker --runner pooling --trust-remote-code
+```
+Exposes `/rerank`, `/v1/rerank` (Jina), `/v2/rerank` (Cohere) and `/score`:
+```bash
+curl http://localhost:8000/rerank -H 'Content-Type: application/json' -d '{
+  "model": "KIEFERSA/Sophea-Nemo-Reranker",
+  "query": "Ποια είναι η πρωτεύουσα της Ελλάδας;",
+  "documents": ["Η Αθήνα είναι η πρωτεύουσα της Ελλάδας.", "Το Παρίσι είναι στη Γαλλία."]
+}'
+```
+## License
+Fine-tune of `nvidia/llama-nemotron-rerank-1b-v2` — governed by the **NVIDIA Open Model License** and
+the **Llama 3.2 Community License**; commercial use permitted under those terms (note: **not**
+Apache-2.0). Review the base model's license before use.

base_vs_ft.png ADDED Viewed

config.json ADDED Viewed

	@@ -0,0 +1,48 @@

+{
+  "architectures": [
+    "LlamaBidirectionalForSequenceClassification"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "llama_bidirectional_model.LlamaBidirectionalConfig",
+    "AutoModelForSequenceClassification": "llama_bidirectional_model.LlamaBidirectionalForSequenceClassification"
+  },
+  "bos_token_id": 128000,
+  "dtype": "bfloat16",
+  "eos_token_id": 128001,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "id2label": {
+    "0": "LABEL_0"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "label2id": {
+    "LABEL_0": 0
+  },
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "llama_bidirec",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 16,
+  "num_key_value_heads": 8,
+  "pad_token_id": 128001,
+  "pooling": "avg",
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_parameters": {
+    "factor": 32.0,
+    "high_freq_factor": 4.0,
+    "low_freq_factor": 1.0,
+    "original_max_position_embeddings": 8192,
+    "rope_theta": 500000.0,
+    "rope_type": "llama3"
+  },
+  "temperature": 1.0,
+  "tie_word_embeddings": true,
+  "transformers_version": "5.13.0",
+  "use_cache": false,
+  "vocab_size": 128256
+}

llama_bidirectional_model.py ADDED Viewed

	@@ -0,0 +1,362 @@

+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0.
+"""
+Bidirectional Llama model for cross-encoder reranking.
+Modifies LlamaModel to use bidirectional (non-causal) attention so each token
+attends to all others — required for cross-encoder scoring of query-document pairs.
+Provides three classes:
+    - LlamaBidirectionalConfig: Adds pooling and temperature to LlamaConfig.
+    - LlamaBidirectionalModel: LlamaModel with causal masking replaced by
+      bidirectional masking. Overrides forward() to support transformers >=4.44.
+    - LlamaBidirectionalForSequenceClassification: Pools hidden states and
+      projects to a relevance score via a linear head.
+Transformers version compatibility (>=4.44 including 5.0+):
+    The forward() implementation handles these API changes at import time via
+    inspect.signature() on LlamaDecoderLayer and DynamicCache:
+    < 4.53:  _update_causal_mask exists on LlamaModel (not used here).
+    4.53+:   Masking moved to masking_utils; requires full forward() override.
+    < 4.54:  Decoder layer returns a tuple.
+    4.54+:   Decoder layer returns a tensor.
+    < 4.56:  Cache kwarg is ``past_key_value`` (singular).
+    4.56+:   Cache kwarg is ``past_key_values`` (plural); DynamicCache accepts config.
+    5.0+:    Native ``create_bidirectional_mask`` in masking_utils.
+"""
+import inspect
+from typing import Optional, Union, Tuple, List
+import torch
+import torch.nn as nn
+from torch.nn import BCEWithLogitsLoss, CrossEntropyLoss, MSELoss
+from transformers.modeling_outputs import SequenceClassifierOutputWithPast
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.modeling_outputs import BaseModelOutputWithPast
+from transformers.models.llama.configuration_llama import LlamaConfig
+from transformers.models.llama.modeling_llama import (
+    LlamaDecoderLayer,
+    LlamaModel,
+    LlamaPreTrainedModel,
+)
+from transformers.utils import logging
+logger = logging.get_logger(__name__)
+# Check if native create_bidirectional_mask exists (transformers >= 5.0)
+try:
+    from transformers.masking_utils import create_bidirectional_mask
+    _HAS_NATIVE_BIDIRECTIONAL_MASK = True
+except ImportError:
+    from transformers.modeling_attn_mask_utils import _prepare_4d_attention_mask
+    _HAS_NATIVE_BIDIRECTIONAL_MASK = False
+# Detect API differences via introspection
+_decoder_forward_params = inspect.signature(LlamaDecoderLayer.forward).parameters
+_dynamic_cache_init_params = inspect.signature(DynamicCache.__init__).parameters
+# past_key_value (singular) in < 4.56, past_key_values (plural) in >= 4.56
+_USE_PLURAL_CACHE_PARAM = "past_key_values" in _decoder_forward_params
+# DynamicCache accepts config parameter in >= 4.56
+_DYNAMIC_CACHE_ACCEPTS_CONFIG = "config" in _dynamic_cache_init_params
+class LlamaBidirectionalConfig(LlamaConfig):
+    """Configuration for LlamaBidirectionalModel with pooling and temperature settings."""
+    model_type = "llama_bidirec"
+    def __init__(
+        self, pooling: str = "avg", temperature: float = 1.0, **kwargs
+    ) -> None:
+        """
+        Initialize bidirectional Llama configuration.
+        Args:
+            pooling: Pooling strategy for embeddings ("avg", "cls", "last", etc.)
+            temperature: Temperature scaling for embeddings
+            **kwargs: Additional arguments passed to LlamaConfig
+        """
+        self.pooling = pooling
+        self.temperature = temperature
+        super().__init__(**kwargs)
+class LlamaBidirectionalModel(LlamaModel):
+    """
+    LlamaModel modified to use bidirectional (non-causal) attention.
+    In standard Llama, each token can only attend to previous tokens (causal attention).
+    This model removes that restriction, allowing each token to attend to all tokens
+    in the sequence, which is useful for embedding tasks.
+    The key modifications are:
+        1. Setting is_causal=False on all attention layers
+        2. Using a bidirectional attention mask instead of causal mask
+    """
+    config_class = LlamaBidirectionalConfig
+    def __init__(self, config: LlamaConfig) -> None:
+        super().__init__(config)
+        for layer in self.layers:
+            layer.self_attn.is_causal = False
+    def _create_bidirectional_mask(
+        self,
+        input_embeds: torch.Tensor,
+        attention_mask: torch.Tensor | None,
+    ) -> torch.Tensor | None:
+        """
+        Create bidirectional attention mask.
+        Args:
+            input_embeds: Input embeddings tensor of shape (batch_size, seq_len, hidden_size)
+            attention_mask: Optional 2D attention mask of shape (batch_size, seq_len)
+                where 1 indicates tokens to attend to and 0 indicates masked tokens
+        Returns:
+            4D attention mask suitable for the attention implementation, or None
+            if no masking is needed
+        """
+        if attention_mask is None:
+            return None
+        if _HAS_NATIVE_BIDIRECTIONAL_MASK:
+            return create_bidirectional_mask(
+                self.config,
+                input_embeds,
+                attention_mask=attention_mask,
+            )
+        # Fallback for transformers < 5.0 without create_bidirectional_mask
+        # Flash attention handles 2D masks internally; only pass mask if there
+        # are actually masked tokens (zeros), otherwise return None for efficiency
+        if getattr(self.config, "_attn_implementation", None) == "flash_attention_2":
+            has_masked_tokens = (attention_mask == 0).any()
+            return attention_mask if has_masked_tokens else None
+        return _prepare_4d_attention_mask(attention_mask, input_embeds.dtype)
+    def forward(
+        self,
+        input_ids: torch.LongTensor | None = None,
+        attention_mask: torch.Tensor | None = None,
+        position_ids: torch.LongTensor | None = None,
+        past_key_values: Cache | None = None,
+        inputs_embeds: torch.FloatTensor | None = None,
+        cache_position: torch.LongTensor | None = None,
+        use_cache: bool | None = None,
+        **kwargs,
+    ) -> BaseModelOutputWithPast:
+        """
+        Forward pass with bidirectional attention.
+        Args:
+            input_ids: Input token IDs of shape (batch_size, seq_len)
+            attention_mask: Attention mask of shape (batch_size, seq_len)
+            position_ids: Position IDs for rotary embeddings
+            past_key_values: Cached key/value states for incremental decoding
+            inputs_embeds: Pre-computed input embeddings (alternative to input_ids)
+            cache_position: Position indices for cache updates
+            use_cache: Whether to return cached key/value states
+            **kwargs: Additional arguments passed to decoder layers
+        Returns:
+            BaseModelOutputWithPast containing last_hidden_state and past_key_values
+        """
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError(
+                "You must specify exactly one of input_ids or inputs_embeds"
+            )
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+        # Initialize cache if needed
+        if use_cache and past_key_values is None:
+            if _DYNAMIC_CACHE_ACCEPTS_CONFIG:
+                past_key_values = DynamicCache(config=self.config)
+            else:
+                past_key_values = DynamicCache()
+        if cache_position is None:
+            past_seen_tokens = (
+                past_key_values.get_seq_length() if past_key_values is not None else 0
+            )
+            cache_position = torch.arange(
+                past_seen_tokens,
+                past_seen_tokens + inputs_embeds.shape[1],
+                device=inputs_embeds.device,
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+        bidirectional_mask = self._create_bidirectional_mask(
+            inputs_embeds, attention_mask
+        )
+        hidden_states = inputs_embeds
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+        # Build decoder layer kwargs with correct cache parameter name
+        # (past_key_value in < 4.56, past_key_values in >= 4.56)
+        layer_kwargs = {
+            "attention_mask": bidirectional_mask,
+            "position_ids": position_ids,
+            "use_cache": use_cache,
+            "cache_position": cache_position,
+            "position_embeddings": position_embeddings,
+        }
+        if _USE_PLURAL_CACHE_PARAM:
+            layer_kwargs["past_key_values"] = past_key_values
+        else:
+            layer_kwargs["past_key_value"] = past_key_values
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            layer_outputs = decoder_layer(hidden_states, **layer_kwargs)
+            # Decoder returns tuple in < 4.54, tensor in >= 4.54
+            if isinstance(layer_outputs, tuple):
+                hidden_states = layer_outputs[0]
+            else:
+                hidden_states = layer_outputs
+        hidden_states = self.norm(hidden_states)
+        return BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+def pool(
+    last_hidden_states: torch.Tensor, attention_mask: torch.Tensor, pool_type: str
+) -> torch.Tensor:
+    last_hidden = last_hidden_states.masked_fill(~attention_mask[..., None].bool(), 0.0)
+    if pool_type == "avg":
+        emb = last_hidden.sum(dim=1) / attention_mask.sum(dim=1)[..., None]
+    elif pool_type == "weighted_avg":
+        emb = last_hidden.sum(dim=1)
+    elif pool_type == "cls":
+        emb = last_hidden[:, 0]
+    elif pool_type == "last":
+        left_padding = attention_mask[:, -1].sum() == attention_mask.shape[0]
+        if left_padding:
+            emb = last_hidden[:, -1]
+        else:
+            sequence_lengths = attention_mask.sum(dim=1) - 1
+            batch_size = last_hidden.shape[0]
+            emb = last_hidden[
+                torch.arange(batch_size, device=last_hidden.device), sequence_lengths
+            ]
+    else:
+        raise ValueError(f"pool_type {pool_type} not supported")
+    return emb
+class LlamaBidirectionalForSequenceClassification(LlamaPreTrainedModel):
+    config_class = LlamaBidirectionalConfig
+    def __init__(self, config):
+        super().__init__(config)
+        self.num_labels = config.num_labels
+        self.score = nn.Linear(config.hidden_size, self.num_labels, bias=False)
+        self.model = LlamaBidirectionalModel(config)
+        # Initialize weights and apply final processing
+        self.post_init()
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        **kwargs,
+    ) -> Union[Tuple, SequenceClassifierOutputWithPast]:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size,)`, *optional*):
+            Labels for computing the sequence classification/regression loss. Indices should be in `[0, ...,
+            config.num_labels - 1]`. If `config.num_labels == 1` a regression loss is computed (Mean-Square loss), If
+            `config.num_labels > 1` a classification loss is computed (Cross-Entropy).
+        """
+        return_dict = (
+            return_dict if return_dict is not None else self.config.use_return_dict
+        )
+        transformer_outputs = self.model(
+            input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            return_dict=return_dict,
+            **kwargs,
+        )
+        hidden_states = transformer_outputs[0]
+        pooled_hidden_states = pool(
+            last_hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            pool_type=self.config.pooling,
+        )
+        pooled_logits = self.score(pooled_hidden_states)
+        pooled_logits = pooled_logits / self.config.temperature
+        loss = None
+        if labels is not None:
+            labels = labels.to(pooled_logits.device)
+            if self.config.problem_type is None:
+                if self.num_labels == 1:
+                    self.config.problem_type = "regression"
+                elif self.num_labels > 1 and (
+                    labels.dtype == torch.long or labels.dtype == torch.int
+                ):
+                    self.config.problem_type = "single_label_classification"
+                else:
+                    self.config.problem_type = "multi_label_classification"
+            if self.config.problem_type == "regression":
+                loss_fct = MSELoss()
+                if self.num_labels == 1:
+                    loss = loss_fct(pooled_logits.squeeze(), labels.squeeze())
+                else:
+                    loss = loss_fct(pooled_logits, labels)
+            elif self.config.problem_type == "single_label_classification":
+                loss_fct = CrossEntropyLoss()
+                loss = loss_fct(
+                    pooled_logits.view(-1, self.num_labels), labels.view(-1)
+                )
+            elif self.config.problem_type == "multi_label_classification":
+                loss_fct = BCEWithLogitsLoss()
+                loss = loss_fct(pooled_logits, labels)
+        if not return_dict:
+            output = (pooled_logits,) + transformer_outputs[1:]
+            return ((loss,) + output) if loss is not None else output
+        return SequenceClassifierOutputWithPast(
+            loss=loss,
+            logits=pooled_logits,
+            past_key_values=transformer_outputs.past_key_values,
+            hidden_states=transformer_outputs.hidden_states,
+            attentions=transformer_outputs.attentions,
+        )

model.safetensors ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:f05ed7b48677a1c574b2c33c732867414eb70fdb7df1039a9962824ce658b655
+size 2471649792

stack.png ADDED Viewed

tokenizer.json ADDED Viewed

	@@ -0,0 +1,3 @@

+version https://git-lfs.github.com/spec/v1
+oid sha256:8df7c1e572298d8e41141627b38c75e5103e131d16e1c334f365bb0a930b0260
+size 17210189

tokenizer_config.json ADDED Viewed

	@@ -0,0 +1,15 @@

+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "is_local": true,
+  "local_files_only": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 4096,
+  "pad_token": "<|end_of_text|>",
+  "tokenizer_class": "TokenizersBackend"
+}