feat(server): relations from GLiNER relex models, plus GLiNER bi-encoder v2 and PII models

README.md

@@ -45,7 +45,7 @@ One SIE cluster runs the inference behind a whole agent. Each task is a handful
 |---|---|---|
 | **Search** | Embed, match, and rerank to retrieve the right context. | [`bge-m3`](packages/sie_server/models/BAAI__bge-m3.yaml), [`splade-v3`](packages/sie_server/models/naver__splade-v3.yaml), [`colbertv2`](packages/sie_server/models/colbert-ir__colbertv2.0.yaml), [`qwen3-reranker`](packages/sie_server/models/Qwen__Qwen3-Reranker-4B.yaml) |
 | **Document to markdown** | PDFs, Office files, and scans become clean markdown. | [`lightonocr`](packages/sie_server/models/lightonai__LightOnOCR-2-1B.yaml), [`glm-ocr`](packages/sie_server/models/zai-org__GLM-OCR.yaml), [`mineru`](packages/sie_server/models/opendatalab__MinerU2.5-Pro-2604-1.2B.yaml), [`paddleocr-vl`](packages/sie_server/models/PaddlePaddle__PaddleOCR-VL-1.5.yaml), [`docling`](packages/sie_server/models/docling.yaml) |
-| **Structured output** | Schema-valid JSON, extracted or generated. | [`gliner2`](packages/sie_server/models/fastino__gliner2-large-v1.yaml), [`nuner-zero`](packages/sie_server/models/numind__NuNER_Zero.yaml), [`qwen3.8-27b`](packages/sie_server/models/Qwen__Qwen3.8-27B-FP8.yaml), [`qwen3.6-27b`](packages/sie_server/models/Qwen__Qwen3.6-27B.yaml) |
+| **Structured output** | Schema-valid JSON, extracted or generated. | [`gliner2`](packages/sie_server/models/fastino__gliner2-large-v1.yaml), [`gliner-relex`](packages/sie_server/models/knowledgator__gliner-relex-large-v1.0.yaml), [`nuner-zero`](packages/sie_server/models/numind__NuNER_Zero.yaml), [`qwen3.8-27b`](packages/sie_server/models/Qwen__Qwen3.8-27B-FP8.yaml), [`qwen3.6-27b`](packages/sie_server/models/Qwen__Qwen3.6-27B.yaml) |
 | **Decide** | Choice, yes/no, and score answers with probabilities to typed questions about a text or JSON state. | [`laya`](packages/sie_server/models/convaiinnovations__laya.yaml), [`laya-multilingual`](packages/sie_server/models/convaiinnovations__laya-multilingual.yaml), [`laya-typed-decisions`](packages/sie_server/models/convaiinnovations__laya-typed-decisions.yaml) |
 | **Classify** | Zero-shot labels, with several label groups answered in one call. The instruct models also follow a task instruction and few-shot examples. | [`gliclass-large-v3`](packages/sie_server/models/knowledgator__gliclass-large-v3.0.yaml), [`gliclass-instruct-large`](packages/sie_server/models/knowledgator__gliclass-instruct-large-v1.0.yaml), [`gliclass-multilang-mini`](packages/sie_server/models/knowledgator__gliclass-multilang-mini.yaml) |
 | **Guard content** | A safety verdict: Yes/No with the threshold set in the model config, or safe/unsafe and policy-label scores with the threshold chosen per request. | [`granite-guardian-2b`](packages/sie_server/models/ibm-granite__granite-guardian-3.0-2b.yaml), [`opir-multitask-large`](packages/sie_server/models/knowledgator__opir-multitask-large-v1.0.yaml), [`opir-edge`](packages/sie_server/models/knowledgator__opir-edge-v1.0.yaml) |

packages/sie_sdk/README.md

@@ -34,6 +34,30 @@ for entry in scores["scores"]:
     print(entry["item_id"], entry["score"])
 ```
 
+## Entities and relations
+
+GLiNER models extract entity spans for the labels you pass. Joint
+entity-relation models such as `knowledgator/gliner-relex-large-v1.0` also
+return relations between those entities when you name relation types in
+`options`:
+
+```python
+result = client.extract(
+    "knowledgator/gliner-relex-large-v1.0",
+    Item(text="Steve Jobs founded Apple in Cupertino."),
+    labels=["person", "organization", "location"],
+    options={"relation_labels": ["founded", "located in"], "relation_threshold": 0.7},
+)
+for relation in result["relations"]:
+    print(relation["head"], relation["relation"], relation["tail"], relation["score"])
+```
+
+Without `relation_labels` these models return entities only. Relations are
+scored among at most 100 entity candidates per item, taken in document order,
+so a very long document may not get relations for its later entities; its
+entities are still all returned. `threshold` and `relation_threshold` must be
+between 0 and 1, and these models need a `threshold` of at least 0.1.
+
 ## Zero-shot classification
 
 GLiClass models (`knowledgator/gliclass-*` and the `knowledgator/opir-*`

packages/sie_server/bundles/default.yaml

@@ -73,8 +73,8 @@ deps:
   pillow: '==12.3.0'
   # grounding_dino
   requests: ''
-  # gliner, gliner_bi
-  gliner: '>=0.2,<1'
+  # gliner, gliner_bi (0.2.26: first release that loads the relex checkpoints)
+  gliner: '>=0.2.26,<1'
   # gliner2
   gliner2: '>=1.3.1,<2'
   # glirel

packages/sie_server/models/knowledgator__gliner-bi-base-v2.0.yaml

@@ -1,6 +1,8 @@
 sie_id: knowledgator/gliner-bi-base-v2.0
 hf_id: knowledgator/gliner-bi-base-v2.0
 hf_revision: 3fdae58db008a8b4e0a803a920ce9bf6566aceed
+hf_tokenizer_dependencies:
+  BAAI/bge-small-en-v1.5: 5c38ec7c405ec4b44b94cc5a9bb96e735b38267a
 inputs:
   text: true
   image: false

packages/sie_server/models/knowledgator__gliner-bi-edge-v2.0.yaml

@@ -0,0 +1,25 @@
+sie_id: knowledgator/gliner-bi-edge-v2.0
+hf_id: knowledgator/gliner-bi-edge-v2.0
+hf_revision: 3a401e4902d93df04971c795fb1f42f37801aced
+hf_tokenizer_dependencies:
+  sentence-transformers/all-MiniLM-L6-v2: 1110a243fdf4706b3f48f1d95db1a4f5529b4d41
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 1024
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: float16
+    adapter_path: sie_server.adapters.gliner_bi:GLiNERBiAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3
+        precompute_labels: true

packages/sie_server/models/knowledgator__gliner-bi-large-v2.0.yaml

@@ -0,0 +1,25 @@
+sie_id: knowledgator/gliner-bi-large-v2.0
+hf_id: knowledgator/gliner-bi-large-v2.0
+hf_revision: 71eec1d5fe57684cd6669ad8a1e1e3e7401a9a55
+hf_tokenizer_dependencies:
+  BAAI/bge-base-en-v1.5: a5beb1e3e68b9ab74eb54cfd186867f64f240e1a
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 1024
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: float16
+    adapter_path: sie_server.adapters.gliner_bi:GLiNERBiAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3
+        precompute_labels: true

packages/sie_server/models/knowledgator__gliner-bi-small-v2.0.yaml

@@ -0,0 +1,25 @@
+sie_id: knowledgator/gliner-bi-small-v2.0
+hf_id: knowledgator/gliner-bi-small-v2.0
+hf_revision: 3d74c1bf459b8b1c0be1ecbddd679416ce005418
+hf_tokenizer_dependencies:
+  sentence-transformers/all-MiniLM-L12-v2: a50ef00143b4d5391434df20ae11632588ac25be
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 1024
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: float16
+    adapter_path: sie_server.adapters.gliner_bi:GLiNERBiAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3
+        precompute_labels: true

packages/sie_server/models/knowledgator__gliner-pii-base-v1.0.yaml

@@ -0,0 +1,22 @@
+sie_id: knowledgator/gliner-pii-base-v1.0
+hf_id: knowledgator/gliner-pii-base-v1.0
+hf_revision: 61726e0ad791dcab3e29339bbec3ad42ded65641
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 2048
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: null
+    adapter_path: sie_server.adapters.gliner:GLiNERAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3

packages/sie_server/models/knowledgator__gliner-pii-edge-v1.0.yaml

@@ -0,0 +1,22 @@
+sie_id: knowledgator/gliner-pii-edge-v1.0
+hf_id: knowledgator/gliner-pii-edge-v1.0
+hf_revision: 9b7f39b0a2da971a5beea78d35f1539d4009c891
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 2048
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: null
+    adapter_path: sie_server.adapters.gliner:GLiNERAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3

packages/sie_server/models/knowledgator__gliner-pii-large-v1.0.yaml

@@ -0,0 +1,22 @@
+sie_id: knowledgator/gliner-pii-large-v1.0
+hf_id: knowledgator/gliner-pii-large-v1.0
+hf_revision: f847f54fbc97ad6e78bfa20ed9c5e5d5c43327b9
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 768
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: null
+    adapter_path: sie_server.adapters.gliner:GLiNERAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3

packages/sie_server/models/knowledgator__gliner-pii-small-v1.0.yaml

@@ -0,0 +1,22 @@
+sie_id: knowledgator/gliner-pii-small-v1.0
+hf_id: knowledgator/gliner-pii-small-v1.0
+hf_revision: d21aad5b4a7ec82b3d0970fd1ac74a12c087d85e
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 2048
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: null
+    adapter_path: sie_server.adapters.gliner:GLiNERAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        threshold: 0.3

packages/sie_server/models/knowledgator__gliner-relex-large-v1.0.yaml

@@ -0,0 +1,23 @@
+sie_id: knowledgator/gliner-relex-large-v1.0
+hf_id: knowledgator/gliner-relex-large-v1.0
+hf_revision: 4aedc9226a5ac9e2f6b5ea3e91c1ee577c88a290
+inputs:
+  text: true
+  image: false
+  audio: false
+  video: false
+tasks:
+  encode: null
+  score: null
+  extract: {}
+max_sequence_length: 2048
+profiles:
+  default:
+    max_batch_tokens: 16384
+    compute_precision: null
+    adapter_path: sie_server.adapters.gliner:GLiNERAdapter
+    adapter_options:
+      loadtime: {}
+      runtime:
+        # Model card: relation scores are most precise at 0.7-0.9.
+        relation_threshold: 0.7

packages/sie_server/pyproject.toml

@@ -44,7 +44,8 @@ dependencies = [
     # authority: only what this distribution declares reaches a resolved lock
     # and therefore a built image. gliner2 lived in the bundle alone and so
     # never shipped, killing three extract cells at model load (#2873).
-    "gliner>=0.2,<1",
+    # 0.2.26 is the first release that loads the joint entity-relation checkpoints.
+    "gliner>=0.2.26,<1",
     "gliner2>=1.3.1,<2",
     "glirel>=1.0,<2",
     # 0.1.17 adds the cross-attention scorer and pass-through pooling used by