Add CrossEncoder integration

1_LogitScore/config.json

@@ -0,0 +1,4 @@
+{
+    "true_token_id": 9693,
+    "false_token_id": 2152
+}

README.md

@@ -58,6 +58,22 @@ If the document is not relevant, the model outputs `no` and stops. No contributi
 
 ## Quickstart
 
+Two ways to call the model. Both produce the **same** relevance score `s(q, d) = σ(ℓ_yes − ℓ_no)`. Use **A** when you also want `<contribution>` / `<evidence>`. Use **B** when you only need a score and want a drop-in replacement for any other CrossEncoder reranker.
+
+We use one shared example throughout so you can compare the outputs side by side:
+
+```python
+QUERY = "What is the boiling point of water at sea level?"
+DOCUMENTS = [
+    "Water boils at 100 C (212 F) at standard atmospheric pressure (1 atm), "
+    "which corresponds to sea-level conditions.",
+    "Mount Everest is the highest mountain on Earth, with a peak elevation "
+    "of 8,848 meters above sea level.",
+]
+```
+
+### A. Transformers (full output: score + contribution + evidence)
+
 ```python
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
@@ -80,7 +96,7 @@ INSTRUCTION = (
     "- Concise: drop query-irrelevant background.\n"
     "- Verbatim (no translation): proper nouns, terms, abbreviations, "
     "numbers, dates, code, URLs.\n"
-    "- Output language: multilingual doc -> query's language; else doc's language."
+    "- Output language: multilingual doc → query's language; else doc's language."
     "</evidence>"
 )
 
@@ -138,27 +154,62 @@ def rerank(query: str, doc: str, max_new_tokens: int = 512):
     return {"score": score, "text": text}
 
 
-example = rerank(
-    query="What is the boiling point of water at sea level?",
-    doc=(
-        "Water boils at 100 C (212 F) at standard atmospheric pressure (1 atm), "
-        "which corresponds to sea-level conditions."
-    ),
-)
-print(example)
+for doc in DOCUMENTS:
+    print(rerank(QUERY, doc))
 ```
 
-Expected shape of the output:
+Expected output (one dict per document):
 
 ```text
-{
-  "score": 0.98,
-  "text": "yes\n<contribution>...</contribution>\n<evidence>...</evidence>"
-}
+{"score": 0.98, "text": "yes\n<contribution>...</contribution>\n<evidence>...</evidence>"}
+{"score": 0.01, "text": "no"}
 ```
 
 For irrelevant pairs the score is close to 0 and `text` is just `"no"`.
 
+### B. Sentence Transformers CrossEncoder (score only)
+
+If you only need the score and want a drop-in CrossEncoder, the same model works directly with `sentence-transformers >= 5.4.0`. **Note:** in this mode `<contribution>` and `<evidence>` are not produced — only the calibrated relevance score.
+
+The system prompt and instruction are baked into the model's `chat_template.jinja` and are **not configurable** — the model was trained with one fixed prompt and only that prompt produces calibrated scores. You only pass `(query, document)`; the rest is hardcoded.
+
+```python
+import torch
+from sentence_transformers import CrossEncoder
+
+MODEL_PATH = "infgrad/Prism-Qwen3.5-Reranker-4B"  # or any sibling repo above
+
+ce = CrossEncoder(MODEL_PATH, model_kwargs={"torch_dtype": torch.bfloat16})
+
+# 1) Score (q, d) pairs. The default activation is Sigmoid, so scores are in (0, 1)
+# and equal to s(q, d) = sigmoid(logit_yes - logit_no) — identical to path A above.
+pairs = [(QUERY, doc) for doc in DOCUMENTS]
+scores = ce.predict(pairs)
+print(scores)
+# array([0.98, 0.01], dtype=float32)
+
+# 2) Rank documents directly.
+ranked = ce.rank(QUERY, DOCUMENTS, return_documents=True)
+for r in ranked:
+    print(f"{r['score']:.3f}\t{r['corpus_id']}\t{r['text'][:80]}")
+```
+
+To get raw logit differences instead of [0, 1] probabilities, pass `activation_fn=torch.nn.Identity()` to `ce.predict(...)`.
+
+#### A note on numerical parity with path A
+
+In **fp32**, paths A and B produce the same score to within ~1e-6 (verified across all five checkpoints).
+
+In **bf16** with the default batched call (`batch_size > 1`), CE scores can drift from path A by **~1–3%** for individual pairs. The cause is bf16 SDPA: when CrossEncoder pads shorter sequences to the longest in the batch, the bf16 attention numerics differ by a few ULPs vs running each pair alone, and the difference accumulates across layers before the final sigmoid. **Ranking order is unaffected.** If you need bit-for-bit parity with path A:
+
+```python
+# Option 1: keep bf16, disable batching
+ce.predict(pairs, batch_size=1)
+
+# Option 2: use fp32 (slower, larger memory)
+ce = CrossEncoder(MODEL_PATH, model_kwargs={"torch_dtype": torch.float32})
+```
+
 
 ## Notes on usage
 
@@ -170,4 +221,4 @@ For irrelevant pairs the score is close to 0 and `text` is just `"no"`.
 
 ## Contact
 
-Dun Zhang — `dunnzhang0@gmail.com` (independent researcher).
+Dun Zhang — `dunnzhang0@gmail.com` (independent researcher).

chat_template.jinja

@@ -0,0 +1,22 @@
+{%- set query_text = messages | selectattr("role", "eq", "query") | map(attribute="content") | first -%}
+{%- set document_text = messages | selectattr("role", "eq", "document") | map(attribute="content") | first -%}
+<|im_start|>system
+Judge whether the Document meets the requirements based on the Query and the Instruct provided. <|im_end|>
+<|im_start|>user
+<Instruct>: Judge if the document is relevant to the query. Reply "yes" or "no".
+On "yes", also emit:
+<contribution>One sentence covering every core point the document contributes to the query, without elaboration.</contribution>
+<evidence>Self-contained rewrite of the query-relevant content. Rules:
+- Faithful: rephrase only; add or infer nothing.
+- Self-contained: evidence alone must fully answer the query.
+- Concise: drop query-irrelevant background.
+- Verbatim (no translation): proper nouns, terms, abbreviations, numbers, dates, code, URLs.
+- Output language: multilingual doc → query's language; else doc's language.</evidence>
+<Query>: {{ query_text }}
+<Document>: {{ document_text }}<|im_end|>
+<|im_start|>assistant
+<think>
+
+</think>
+
+

config_sentence_transformers.json

@@ -0,0 +1,7 @@
+{
+  "__version__": {
+    "sentence_transformers": "5.4.1"
+  },
+  "activation_fn": "torch.nn.modules.activation.Sigmoid",
+  "model_type": "CrossEncoder"
+}

modules.json

@@ -0,0 +1,14 @@
+[
+  {
+    "idx": 0,
+    "name": "0",
+    "path": "",
+    "type": "sentence_transformers.base.modules.transformer.Transformer"
+  },
+  {
+    "idx": 1,
+    "name": "1",
+    "path": "1_LogitScore",
+    "type": "sentence_transformers.cross_encoder.modules.logit_score.LogitScore"
+  }
+]

sentence_bert_config.json

@@ -0,0 +1,15 @@
+{
+    "transformer_task": "text-generation",
+    "modality_config": {
+        "text": {
+            "method": "forward",
+            "method_output_name": "logits"
+        },
+        "message": {
+            "method": "forward",
+            "method_output_name": "logits",
+            "format": "flat"
+        }
+    },
+    "module_output_name": "causal_logits"
+}