feat: auto-load HF ONNX artifacts on CPU

AGENTS.md

@@ -686,3 +686,12 @@ Entry format:
 - Details: Added repo purpose/runtime constraints, ownership table, safe-edit boundaries, required verification matrix, role-based agent cards, A2A-style task contract, mandatory handoff envelope, single-agent loop, actor-critic workflow, reusable lessons-learned section, worked examples, and explicit AGENTS review triggers. Critic concerns addressed in the update: keep single-agent as default to avoid unnecessary delegation, require executable evidence for actor-vs-critic disputes, and keep memory logging critical-only instead of turning the file into an activity log.
 - Verification: `sed -n '1,260p' AGENTS.md`; `rg -n "Module And File Ownership|Safe-Edit Boundaries|Required Verification Commands|Agent Cards|A2A-Style Task Contract|Actor-Critic|AGENTS Review Triggers|ML-20260317-001" AGENTS.md`
 - Links: `AGENTS.md`
+
+- ID: `ML-20260508-001`
+- Timestamp: `2026-05-08T23:08:05Z`
+- Type: `change`
+- Summary: Shared HF text serving now auto-selects ONNX artifacts for CPU-only runtime when the HF repo declares a compatible runtime manifest.
+- Criticality: Serving runtime architecture change affecting model loading, artifact downloads, output decoding, and API metadata across generic text-classification deployments.
+- Details: `ThHfModelBase` keeps Transformers/PT as the default GPU and fallback path, but CPU-only `HF_RUNTIME=auto` now loads `artifact_manifest.json`, selects a declared ONNX Runtime artifact, downloads only safe allow-patterns, loads schema and contract decoder from HF artifacts, and exposes the decoded artifact contract through the existing text-classifier flow. Business API response shaping now passes through generic model/runtime metadata emitted by serving.
+- Verification: `python3 -m unittest extensions.serving.test_th_hf_model_base extensions.serving.test_th_text_classifier extensions.serving.test_th_privacy_filter extensions.business.edge_inference_api.test_text_classifier_inference_api extensions.business.edge_inference_api.test_privacy_filter_inference_api`; `python3 -m py_compile extensions/serving/default_inference/nlp/th_hf_model_base.py extensions/business/edge_inference_api/text_classifier_inference_api.py`; required serving gate `python3 -m unittest extensions.serving.model_testing.test_llm_servings` currently fails at import with `ImportError: cannot import name 'Logger' from 'naeural_core'`.
+- Links: `extensions/serving/default_inference/nlp/th_hf_model_base.py`, `extensions/business/edge_inference_api/text_classifier_inference_api.py`, `extensions/serving/test_th_hf_model_base.py`

extensions/business/edge_inference_api/privacy_filter_inference_api.py

@@ -95,4 +95,14 @@ def _build_result_from_inference(  # pylint: disable=arguments-differ
       result_payload["tokenizer_name"] = inference["TOKENIZER_NAME"]
     if "PIPELINE_TASK" in inference:
       result_payload["pipeline_task"] = inference["PIPELINE_TASK"]
+    if "MODEL" in inference:
+      result_payload["model"] = inference["MODEL"]
+    if "MODEL_VERSION" in inference:
+      result_payload["model_version"] = inference["MODEL_VERSION"]
+    if "MODEL_REVISION" in inference:
+      result_payload["model_revision"] = inference["MODEL_REVISION"]
+    if "HF_RUNTIME" in inference:
+      result_payload["hf_runtime"] = inference["HF_RUNTIME"]
+    if "RUNTIME" in inference:
+      result_payload["runtime"] = inference["RUNTIME"]
     return result_payload

extensions/business/edge_inference_api/test_privacy_filter_inference_api.py

@@ -55,6 +55,11 @@ def test_build_result_from_inference_uses_findings_key(self):
         "FINDINGS_COUNT": 1,
         "MODEL_NAME": "openai/privacy-filter",
         "PIPELINE_TASK": "token-classification",
+        "MODEL": {"model_key": "privacy_filter", "model_version": "2026.05.09"},
+        "MODEL_VERSION": "2026.05.09",
+        "MODEL_REVISION": "rev-privacy",
+        "HF_RUNTIME": "pt",
+        "RUNTIME": "transformers",
       },
       metadata={},
       request_data={"metadata": {}, "parameters": {"text": "example text"}},
@@ -73,6 +78,14 @@ def test_build_result_from_inference_uses_findings_key(self):
     self.assertEqual(result_payload["findings_count"], 1)
     self.assertEqual(result_payload["model_name"], "openai/privacy-filter")
     self.assertEqual(result_payload["pipeline_task"], "token-classification")
+    self.assertEqual(
+      result_payload["model"],
+      {"model_key": "privacy_filter", "model_version": "2026.05.09"},
+    )
+    self.assertEqual(result_payload["model_version"], "2026.05.09")
+    self.assertEqual(result_payload["model_revision"], "rev-privacy")
+    self.assertEqual(result_payload["hf_runtime"], "pt")
+    self.assertEqual(result_payload["runtime"], "transformers")
 
 
 if __name__ == "__main__":

extensions/business/edge_inference_api/test_text_classifier_inference_api.py

@@ -152,6 +152,33 @@ def test_build_result_from_inference_preserves_classifier_output(self):
     self.assertEqual(result_payload["model_name"], "openai/privacy-filter")
     self.assertEqual(result_payload["pipeline_task"], "token-classification")
 
+  def test_build_result_from_inference_preserves_runtime_model_metadata(self):
+    plugin = TextClassifierInferenceApiPlugin()
+
+    result_payload = plugin._build_result_from_inference(  # pylint: disable=protected-access
+      request_id="req-onnx",
+      inference={
+        "REQUEST_ID": "req-onnx",
+        "TEXT": "example text",
+        "result": {"prediction": "safe"},
+        "MODEL": {"key": "generic_text_classifier", "version": "2026.05.09"},
+        "MODEL_VERSION": "2026.05.09",
+        "HF_RUNTIME": "onnx_fp32",
+        "RUNTIME": "onnxruntime",
+      },
+      metadata={},
+      request_data={"metadata": {}, "parameters": {"text": "example text"}},
+    )
+
+    self.assertEqual(result_payload["classification"], {"prediction": "safe"})
+    self.assertEqual(
+      result_payload["model"],
+      {"key": "generic_text_classifier", "version": "2026.05.09"},
+    )
+    self.assertEqual(result_payload["model_version"], "2026.05.09")
+    self.assertEqual(result_payload["hf_runtime"], "onnx_fp32")
+    self.assertEqual(result_payload["runtime"], "onnxruntime")
+
   def test_handle_inferences_falls_back_to_payload_request_id(self):
     plugin = TextClassifierInferenceApiPlugin()
     plugin._requests = {"req-1": {"status": "pending"}}  # pylint: disable=protected-access

extensions/business/edge_inference_api/text_classifier_inference_api.py

@@ -405,6 +405,16 @@ def _build_result_from_inference(
         result_payload["tokenizer_name"] = inference["TOKENIZER_NAME"]
       if "PIPELINE_TASK" in inference:
         result_payload["pipeline_task"] = inference["PIPELINE_TASK"]
+      if "MODEL" in inference:
+        result_payload["model"] = inference["MODEL"]
+      if "MODEL_VERSION" in inference:
+        result_payload["model_version"] = inference["MODEL_VERSION"]
+      if "MODEL_REVISION" in inference:
+        result_payload["model_revision"] = inference["MODEL_REVISION"]
+      if "HF_RUNTIME" in inference:
+        result_payload["hf_runtime"] = inference["HF_RUNTIME"]
+      if "RUNTIME" in inference:
+        result_payload["runtime"] = inference["RUNTIME"]
       return result_payload
 
     def handle_inference_for_request(