Use OpenMed runtime guidance

README.md

@@ -38,8 +38,8 @@ including local clinical-document and chart workflows on Mac, iPhone, and iPad.
 These repositories contain MLX conversions of
 [`CohereLabs/North-Micro-Vision-Instruct`](https://huggingface.co/CohereLabs/North-Micro-Vision-Instruct),
 a compact 2.4B-parameter Cohere Compass vision-language model released under
-Apache 2.0. OpenMed owns the Python and Swift runtime paths described here; no
-`mlx-vlm` installation or model-repository Python code is required for use.
+Apache 2.0. OpenMed provides the Python and Swift runtime paths described here
+and consumes the repositories as data-only model artifacts.
 
 The same byte-identical README is published across all five precision
 variants. The repository name, `config.json`, and `openmed-mlx.json` identify
@@ -62,12 +62,10 @@ while reducing the decoder footprint.
 
 ## Python through OpenMed
 
-Install an OpenMed revision that contains the native Compass runtime. Until the
-linked implementation PR is merged and released, install its tested branch:
+Install OpenMed with Apple MLX support:
 
 ```bash
-python -m pip install -U \
-  "openmed[mlx] @ git+https://github.com/maziyarpanahi/openmed.git@feature/cohere-compass-runtime"
+uv pip install "openmed[mlx]"
 ```
 
 Image plus text:
@@ -237,12 +235,9 @@ incorrect, or fabricated. A qualified human must verify consequential use.
 - Source: [`CohereLabs/North-Micro-Vision-Instruct`](https://huggingface.co/CohereLabs/North-Micro-Vision-Instruct)
 - Pinned source revision: `373bda96ac70bf89f99f7048f420cf00dc07c149`
 - OpenMed / OpenMedKit: [github.com/maziyarpanahi/openmed](https://github.com/maziyarpanahi/openmed)
-- Independent conversion reference: the Cohere Compass port contributed to
-  MLX-VLM at revision `dd79a5d8caf3edafd6fa9e6326d7ce4977ddcbfc`
 
-Thank you to Cohere for releasing North Micro Vision and to the MLX and
-MLX-VLM contributors whose prior Compass work provided a useful independent
-reference while OpenMed implemented and tested its own Python and Swift paths.
+Thank you to Cohere for releasing North Micro Vision and to the Apple MLX
+contributors who make private on-device inference possible.
 
 The converted weights retain the source model's Apache 2.0 license. OpenMed's
 SDK source is separately licensed under Apache 2.0.

openmed-mlx.json

@@ -38,11 +38,5 @@
   "weights": {
     "format": "safetensors",
     "path": "model.safetensors"
-  },
-  "provenance": {
-    "independent_conversion_reference": {
-      "library": "mlx-vlm",
-      "revision": "dd79a5d8caf3edafd6fa9e6326d7ce4977ddcbfc"
-    }
   }
 }

openmed-runtime-validation.json

@@ -20,7 +20,7 @@
       "huggingface-hub": "1.27.0",
       "pillow": "12.3.0"
     },
-    "test": "tests/integration/test_mlx_vlm_compass.py"
+    "test": "tests/integration/test_mlx_vision_language_compass.py"
   },
   "swift": {
     "passed": true,

openmed-validation.json

@@ -3,29 +3,29 @@
     {
       "coherence_detail": "coherent surface form",
       "coherent": true,
-      "elapsed_seconds": 0.2304,
+      "elapsed_seconds": 0.423,
       "fact_detail": "privacy/locality concepts present",
       "facts_correct": true,
       "fixture": null,
-      "generation_tokens": 26,
+      "generation_tokens": 37,
       "id": "text_privacy",
       "passed": true,
-      "peak_memory_gb": 3.231338698,
+      "peak_memory_gb": 3.23761089,
       "prompt": "In one concise sentence, explain how running a vision-language model entirely on-device can improve privacy for clinical documents.",
       "prompt_tokens": 30,
-      "response": "Running a vision-language model entirely on-device reduces the need for cloud storage and transmission of sensitive clinical data, thereby enhancing privacy."
+      "response": "Running a vision-language model entirely on-device can improve privacy for clinical documents by reducing the need to transmit sensitive data over the internet, thus minimizing the risk of data breaches and unauthorized access."
     },
     {
       "coherence_detail": "coherent surface form",
       "coherent": true,
-      "elapsed_seconds": 0.0389,
+      "elapsed_seconds": 0.0375,
       "fact_detail": "all expected facts present",
       "facts_correct": true,
       "fixture": null,
-      "generation_tokens": 2,
+      "generation_tokens": 1,
       "id": "text_fact_extraction",
       "passed": true,
-      "peak_memory_gb": 3.26314305,
+      "peak_memory_gb": 3.281032518,
       "prompt": "A synthetic note states: \"The follow-up appointment is scheduled for Tuesday at 10:30 AM.\" What day is the follow-up? Answer with only the day.",
       "prompt_tokens": 44,
       "response": "Tuesday"
@@ -33,29 +33,29 @@
     {
       "coherence_detail": "coherent surface form",
       "coherent": true,
-      "elapsed_seconds": 0.9995,
+      "elapsed_seconds": 1.2672,
       "fact_detail": "all expected facts present",
       "facts_correct": true,
       "fixture": "synthetic_clinical_document.png",
-      "generation_tokens": 32,
+      "generation_tokens": 31,
       "id": "image_clinical_document",
       "passed": true,
-      "peak_memory_gb": 5.850670814,
+      "peak_memory_gb": 5.843571694,
       "prompt": "This is synthetic test data. In one concise sentence, report the exact patient name, record ID, medication with dose and frequency, and allergy shown in the image.",
       "prompt_tokens": 1161,
-      "response": "The synthetic test data includes patient Alex Rivera, record ID SYN-2048, medication Metformin 500 mg twice daily, and allergy to Penicillin."
+      "response": "The synthetic test data includes patient Alex Rivera (Record ID: SYN-2048), prescribed Metformin 500 mg twice daily, with a penicillin allergy."
     },
     {
       "coherence_detail": "coherent surface form",
       "coherent": true,
-      "elapsed_seconds": 0.7322,
+      "elapsed_seconds": 0.8766,
       "fact_detail": "all expected facts present",
       "facts_correct": true,
       "fixture": "synthetic_clinic_chart.png",
-      "generation_tokens": 5,
+      "generation_tokens": 4,
       "id": "image_chart",
       "passed": true,
-      "peak_memory_gb": 5.850670814,
+      "peak_memory_gb": 5.843571694,
       "prompt": "Which category has the tallest bar, and what exact value is printed above it? Answer concisely.",
       "prompt_tokens": 1053,
       "response": "Screening, 42"
@@ -69,16 +69,16 @@
     "memory_size": 549755813888,
     "resource_limit": 499000
   },
-  "load_seconds": 0.8468,
-  "mlx_vlm_revision": "dd79a5d8caf3edafd6fa9e6326d7ce4977ddcbfc",
-  "model_path": "OpenMed/North-Micro-Vision-Instruct-8bit-mlx",
+  "load_seconds": 2.9559,
+  "model_path": "/Users/maziyar/Developer/openmed-mlx-export/artifacts/north-micro-vision/North-Micro-Vision-Instruct-8bit-mlx",
   "passed": true,
+  "runtime": "openmed.mlx.OpenMedMLXVisionLanguageModel",
   "runtime_versions": {
     "huggingface-hub": "1.27.0",
     "mlx": "0.32.0",
     "mlx-lm": "0.31.3",
     "mlx-metal": "0.32.0",
-    "mlx-vlm": "0.6.10",
+    "openmed": "1.0.0",
     "transformers": "5.15.0"
   },
   "schema_version": 1,