Update pipeline tag and add paper link

README.md

@@ -1,6 +1,13 @@
 ---
+base_model: mixedbread-ai/mxbai-edge-colbert-v0-32m
+datasets:
+- reasonir/reasonir-data
+- hanhainebula/bge-reasoner-data
 language:
 - en
+library_name: PyLate
+license: cc-by-nc-4.0
+pipeline_tag: text-retrieval
 tags:
 - ColBERT
 - PyLate
@@ -11,17 +18,12 @@ tags:
 - reasoning-retrieval
 - edge
 - loss:CachedContrastive
-base_model: mixedbread-ai/mxbai-edge-colbert-v0-32m
-datasets:
-- reasonir/reasonir-data
-- hanhainebula/bge-reasoner-data
-pipeline_tag: sentence-similarity
-library_name: PyLate
-license: cc-by-nc-4.0
 ---
 
 # SmallReason-ColBERT (32M)
 
+**Paper**: [SmallReason-ColBERT: An Ultra-Small Late-Interaction Retriever for Reasoning Intensive Retrieval](https://huggingface.co/papers/2609.29652)
+
 An ultra-small late-interaction retriever for **reasoning-intensive** retrieval.
 32M parameters, plus a **129-parameter query-side importance head**.
 
@@ -178,8 +180,8 @@ head is a no-op against the frozen base at step zero.
 
 Not soft-IDF. Across ~199K BRIGHT query tokens the gate–IDF Spearman correlation is
 **ρ = −0.02** — statistically detectable, practically zero. Per-split mean gate sits in
-0.43–0.47 with std ≈ 0.10: the head is a soft re-weighting, not a selector. A fixed IDF
-gate on the same base reaches only 20.06, against 21.41 for the learned head.
+0.43–0.47 with std ≈ 0.10: the head is a soft re-weighting, not a selector. A fixed
+IDF gate on the same base reaches only 20.06, against 21.41 for the learned head.
 
 ---
 
@@ -240,4 +242,4 @@ the companion training/inference **code** is released under Apache-2.0 — but t
 Thanks to Antoine Chaffin (LightOn, Reason-ModernColBERT) for flagging the upstream
 `2_Dense/use_residual` config bug in `mxbai-edge-colbert-v0-32m` — the base weights were
 trained with a residual on that layer while the shipped config said otherwise. This
-model uses the patched config (`use_residual: true`).
+model uses the patched config (`use_residual: true`).