iints-sdk-python35 1.0.0__tar.gz → 1.1.1__tar.gz

{iints_sdk_python35-1.0.0/src/iints_sdk_python35.egg-info → iints_sdk_python35-1.1.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: iints-sdk-python35
-Version: 1.0.0
+Version: 1.1.1
 Summary: A pre-clinical Edge-AI SDK for diabetes management validation.
 Author-email: Rune Bobbaers <rune.bobbaers@gmail.com>
 Project-URL: Homepage, https://github.com/python35/IINTS-SDK
@@ -46,7 +46,7 @@ Requires-Dist: h5py>=3.10.0; extra == "research"
 Requires-Dist: onnx>=1.16.0; extra == "research"
 Requires-Dist: onnxscript>=0.1.0; extra == "research"
 Provides-Extra: mdmp
-Requires-Dist: mdmp-protocol>=0.2.1; extra == "mdmp"
+Requires-Dist: mdmp-protocol>=0.3.0; extra == "mdmp"
 Dynamic: license-file
 
 # IINTS-AF SDK
@@ -81,6 +81,54 @@ cd iints_quickstart
 iints presets run --name baseline_t1d --algo algorithms/example_algorithm.py
 ```
 
+## AI Assistant (Ministral 3 Open-Weight via Ollama)
+
+The SDK now includes a research-only AI assistant layer for explanations and run summaries.
+It is gated by MDMP verification before any LLM call is allowed.
+
+Use an active virtual environment for the full flow:
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+python -m pip install -U pip
+python -m pip install -e ".[mdmp]"
+```
+
+Run the open local Mistral model locally with Ollama:
+
+```bash
+python -m pip install -e ".[mdmp]"
+ollama pull ministral-3:8b
+iints ai models
+```
+
+Recommended first-time setup:
+
+```bash
+ollama pull ministral-3:8b
+iints ai local-check --model ministral-3:8b
+```
+
+Example commands:
+
+```bash
+iints ai explain results/step.json \
+  --mdmp-cert results/report.signed.mdmp
+
+iints ai report results/simulation_run.json \
+  --mdmp-cert results/report.signed.mdmp \
+  --output results/ai_report.md
+```
+
+Notes:
+- AI analysis is blocked if the MDMP artifact is invalid.
+- Minimum required MDMP grade defaults to `research_grade`.
+- The SDK now targets the open local `Ministral 3` Ollama model by default.
+- Users can choose a larger or smaller local Mistral-family model with `--model ...`.
+- Large JSON payloads are clipped automatically before prompt generation to keep local inference stable.
+- Output is research-only and not medical advice.
+
 ## MDMP (Short)
 MDMP is the data-quality protocol used by IINTS.
 
@@ -118,16 +166,31 @@ mdmp registry push --registry registry/mdmp_registry.json --report results/mdmp_
 - MDMP repo: `python35/MDMP`
 
 Local helper scripts:
-- `tools/local/dual_repo_status.sh`
-- `tools/local/dual_repo_commit_push.sh`
+- `tools/dev/dual_repo_status.sh`
+- `tools/dev/dual_repo_commit_push.sh`
 
 Full process: `docs/DUAL_REPO_WORKFLOW.md`
 
 MDMP sync CI gate:
 - `.github/workflows/mdmp-sync.yml`
-- Pulls the public `python35/MDMP` repo directly.
+- Uses private-repo checkout when `MDMP_REPO_TOKEN` is configured.
+- Falls back to `mdmp-protocol` from PyPI when checkout is unavailable.
 - Auto dependency updates for MDMP are handled via Dependabot (`.github/dependabot.yml`).
 
+## Tools Layout
+
+Repository helpers are now grouped by purpose:
+
+- `scripts/`: simple user-facing shortcuts like test, lint, and demo entrypoints
+- `tools/ci/`: CI gates and policy checks
+- `tools/dev/`: maintainer workflows and multi-repo helpers
+- `tools/docs/`: manual and documentation builders
+- `tools/data/`: dataset import and conversion utilities
+- `tools/analysis/`: plotting, diagnostics, and report helpers
+- `tools/assets/`: branding and asset generation helpers
+
+Reference: `tools/README.md`
+
 ## Typical Workflow
 1. Prepare or import data.
 2. Validate data with MDMP.

{iints_sdk_python35-1.0.0 → iints_sdk_python35-1.1.1}/README.md

@@ -30,6 +30,54 @@ cd iints_quickstart
 iints presets run --name baseline_t1d --algo algorithms/example_algorithm.py
 ```
 
+## AI Assistant (Ministral 3 Open-Weight via Ollama)
+
+The SDK now includes a research-only AI assistant layer for explanations and run summaries.
+It is gated by MDMP verification before any LLM call is allowed.
+
+Use an active virtual environment for the full flow:
+
+```bash
+python3 -m venv .venv
+source .venv/bin/activate
+python -m pip install -U pip
+python -m pip install -e ".[mdmp]"
+```
+
+Run the open local Mistral model locally with Ollama:
+
+```bash
+python -m pip install -e ".[mdmp]"
+ollama pull ministral-3:8b
+iints ai models
+```
+
+Recommended first-time setup:
+
+```bash
+ollama pull ministral-3:8b
+iints ai local-check --model ministral-3:8b
+```
+
+Example commands:
+
+```bash
+iints ai explain results/step.json \
+  --mdmp-cert results/report.signed.mdmp
+
+iints ai report results/simulation_run.json \
+  --mdmp-cert results/report.signed.mdmp \
+  --output results/ai_report.md
+```
+
+Notes:
+- AI analysis is blocked if the MDMP artifact is invalid.
+- Minimum required MDMP grade defaults to `research_grade`.
+- The SDK now targets the open local `Ministral 3` Ollama model by default.
+- Users can choose a larger or smaller local Mistral-family model with `--model ...`.
+- Large JSON payloads are clipped automatically before prompt generation to keep local inference stable.
+- Output is research-only and not medical advice.
+
 ## MDMP (Short)
 MDMP is the data-quality protocol used by IINTS.
 
@@ -67,16 +115,31 @@ mdmp registry push --registry registry/mdmp_registry.json --report results/mdmp_
 - MDMP repo: `python35/MDMP`
 
 Local helper scripts:
-- `tools/local/dual_repo_status.sh`
-- `tools/local/dual_repo_commit_push.sh`
+- `tools/dev/dual_repo_status.sh`
+- `tools/dev/dual_repo_commit_push.sh`
 
 Full process: `docs/DUAL_REPO_WORKFLOW.md`
 
 MDMP sync CI gate:
 - `.github/workflows/mdmp-sync.yml`
-- Pulls the public `python35/MDMP` repo directly.
+- Uses private-repo checkout when `MDMP_REPO_TOKEN` is configured.
+- Falls back to `mdmp-protocol` from PyPI when checkout is unavailable.
 - Auto dependency updates for MDMP are handled via Dependabot (`.github/dependabot.yml`).
 
+## Tools Layout
+
+Repository helpers are now grouped by purpose:
+
+- `scripts/`: simple user-facing shortcuts like test, lint, and demo entrypoints
+- `tools/ci/`: CI gates and policy checks
+- `tools/dev/`: maintainer workflows and multi-repo helpers
+- `tools/docs/`: manual and documentation builders
+- `tools/data/`: dataset import and conversion utilities
+- `tools/analysis/`: plotting, diagnostics, and report helpers
+- `tools/assets/`: branding and asset generation helpers
+
+Reference: `tools/README.md`
+
 ## Typical Workflow
 1. Prepare or import data.
 2. Validate data with MDMP.

{iints_sdk_python35-1.0.0 → iints_sdk_python35-1.1.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "iints-sdk-python35" 
-version = "1.0.0"
+version = "1.1.1"
 authors = [
   { name="Rune Bobbaers", email="rune.bobbaers@gmail.com" },
 ]
@@ -60,7 +60,7 @@ research = [
   "onnxscript>=0.1.0",
 ]
 mdmp = [
-  "mdmp-protocol>=0.2.1",
+  "mdmp-protocol>=0.3.0",
 ]
 
 [project.scripts]

{iints_sdk_python35-1.0.0 → iints_sdk_python35-1.1.1}/src/iints/__init__.py

@@ -61,6 +61,7 @@ from .data.synthetic_mirror import generate_synthetic_mirror, SyntheticMirrorArt
 from .analysis.metrics import generate_benchmark_metrics # Added for benchmark
 from .analysis.reporting import ClinicalReportGenerator
 from .analysis.edge_efficiency import EnergyEstimate, estimate_energy_per_decision
+from .ai import AIResponse, IINTSAssistant, MDMPGuard
 from .highlevel import run_simulation, run_full, run_population
 from .scenarios import ScenarioGeneratorConfig, generate_random_scenario
 
@@ -169,6 +170,9 @@ __all__ = [
     "ClinicalReportGenerator",
     "EnergyEstimate",
     "estimate_energy_per_decision",
+    "AIResponse",
+    "IINTSAssistant",
+    "MDMPGuard",
     # Reporting
     "generate_report",
     "generate_quickstart_report",

iints_sdk_python35-1.1.1/src/iints/ai/__init__.py

@@ -0,0 +1,16 @@
+from .assistant import AIResponse, IINTSAssistant
+from .backends import DEFAULT_MINISTRAL_MODEL, DEFAULT_OLLAMA_HOST, OllamaBackend
+from .mdmp_guard import GuardResult, MDMPGuard
+from .model_catalog import LocalMistralModelProfile, list_local_mistral_models
+
+__all__ = [
+    "AIResponse",
+    "IINTSAssistant",
+    "DEFAULT_MINISTRAL_MODEL",
+    "DEFAULT_OLLAMA_HOST",
+    "OllamaBackend",
+    "GuardResult",
+    "MDMPGuard",
+    "LocalMistralModelProfile",
+    "list_local_mistral_models",
+]

iints_sdk_python35-1.1.1/src/iints/ai/assistant.py

@@ -0,0 +1,124 @@
+from __future__ import annotations
+
+from dataclasses import dataclass
+from pathlib import Path
+from typing import Any
+
+from .backends import DEFAULT_MINISTRAL_MODEL, CompletionBackend, MistralAPIBackend, OllamaBackend
+from .mdmp_guard import GuardResult, MDMPGuard
+from .prompts import TaskName, build_prompt
+
+
+@dataclass(frozen=True)
+class AIResponse:
+    task: str
+    text: str
+    backend: str
+    model: str
+    certification: GuardResult
+
+    def to_dict(self) -> dict[str, Any]:
+        return {
+            "task": self.task,
+            "text": self.text,
+            "backend": self.backend,
+            "model": self.model,
+            "certification": self.certification.to_dict(),
+        }
+
+
+class IINTSAssistant:
+    """Research-only LLM assistant gated by MDMP certification."""
+
+    def __init__(
+        self,
+        mdmp_cert: str | Path,
+        *,
+        mode: str = "auto",
+        model: str = DEFAULT_MINISTRAL_MODEL,
+        minimum_grade: str = "research_grade",
+        public_key_path: str | Path | None = None,
+        trust_store_path: str | Path | None = None,
+        ollama_host: str | None = None,
+        timeout_seconds: float = 120.0,
+        backend: CompletionBackend | None = None,
+        guard: MDMPGuard | None = None,
+    ) -> None:
+        self.guard = guard or MDMPGuard(
+            mdmp_cert,
+            minimum_grade=minimum_grade,
+            public_key_path=public_key_path,
+            trust_store_path=trust_store_path,
+        )
+        self.backend = backend or self._detect_backend(
+            mode=mode,
+            model=model,
+            ollama_host=ollama_host,
+            timeout_seconds=timeout_seconds,
+        )
+
+    def _detect_backend(
+        self,
+        *,
+        mode: str,
+        model: str,
+        ollama_host: str | None,
+        timeout_seconds: float,
+    ) -> CompletionBackend:
+        requested = mode.strip().lower()
+        if requested in {"auto", "local", "ollama"}:
+            ollama_backend = OllamaBackend(
+                model_name=model,
+                base_url=ollama_host,
+                timeout_seconds=timeout_seconds,
+            )
+            local_backend: CompletionBackend = ollama_backend
+            if not ollama_backend.available():
+                raise RuntimeError(
+                    "No local Ollama backend is available. "
+                    f"Could not reach {ollama_backend.base_url}. "
+                    "Start Ollama and try again."
+                )
+            ollama_backend.ensure_model_ready()
+            return local_backend
+        if requested == "api":
+            api_backend: CompletionBackend = MistralAPIBackend()
+            if api_backend.available():
+                return api_backend
+            raise RuntimeError(
+                "Cloud API fallback is not enabled in this SDK build yet. "
+                "Use mode='local' with Ollama."
+            )
+        raise ValueError(f"Unsupported AI mode: {mode}")
+
+    def _run_task(self, task: TaskName, payload: Any) -> AIResponse:
+        certification = self.guard.check()
+        system_prompt, user_prompt = build_prompt(task, payload)
+        text = self.guard.wrap(
+            self.backend.complete(system_prompt=system_prompt, user_prompt=user_prompt)
+        )
+        resolved_model = getattr(self.backend, "resolved_model_name", None)
+        response_model = (
+            str(resolved_model)
+            if isinstance(resolved_model, str) and resolved_model.strip()
+            else str(getattr(self.backend, "model_name", DEFAULT_MINISTRAL_MODEL))
+        )
+        return AIResponse(
+            task=task,
+            text=text,
+            backend=getattr(self.backend, "backend_name", type(self.backend).__name__),
+            model=response_model,
+            certification=certification,
+        )
+
+    def explain_decision(self, step: dict[str, Any]) -> AIResponse:
+        return self._run_task("explain_decision", step)
+
+    def analyze_trends(self, glucose_payload: list[Any] | dict[str, Any]) -> AIResponse:
+        return self._run_task("analyze_trends", glucose_payload)
+
+    def detect_anomalies(self, results: dict[str, Any]) -> AIResponse:
+        return self._run_task("detect_anomalies", results)
+
+    def generate_report(self, run: dict[str, Any]) -> AIResponse:
+        return self._run_task("generate_report", run)

iints_sdk_python35-1.1.1/src/iints/ai/backends/__init__.py

@@ -0,0 +1,11 @@
+from .base import CompletionBackend
+from .mistral_api import MistralAPIBackend
+from .ollama import DEFAULT_MINISTRAL_MODEL, DEFAULT_OLLAMA_HOST, OllamaBackend
+
+__all__ = [
+    "CompletionBackend",
+    "DEFAULT_MINISTRAL_MODEL",
+    "DEFAULT_OLLAMA_HOST",
+    "OllamaBackend",
+    "MistralAPIBackend",
+]

iints_sdk_python35-1.1.1/src/iints/ai/backends/base.py

@@ -0,0 +1,14 @@
+from __future__ import annotations
+
+from typing import Protocol
+
+
+class CompletionBackend(Protocol):
+    backend_name: str
+    model_name: str
+
+    def available(self) -> bool:
+        ...
+
+    def complete(self, *, system_prompt: str, user_prompt: str) -> str:
+        ...

iints_sdk_python35-1.1.1/src/iints/ai/backends/mistral_api.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+
+class MistralAPIBackend:
+    backend_name = "mistral_api"
+
+    def __init__(self, *args, **kwargs) -> None:
+        self.model_name = "mistral_api_unconfigured"
+
+    def available(self) -> bool:
+        return False
+
+    def complete(self, *, system_prompt: str, user_prompt: str) -> str:
+        raise RuntimeError(
+            "Cloud fallback is not enabled in this SDK build yet. "
+            "Use mode='local' with Ollama for the open Ministral 3 model."
+        )

iints_sdk_python35-1.1.1/src/iints/ai/backends/ollama.py

@@ -0,0 +1,238 @@
+from __future__ import annotations
+
+import json
+import os
+from urllib import error, request
+
+
+DEFAULT_OLLAMA_HOST = "http://127.0.0.1:11434"
+DEFAULT_MINISTRAL_MODEL = "ministral-3:8b"
+LEGACY_MINISTRAL_MODEL = "mistral/ministral-8b-instruct"
+MIN_OLLAMA_VERSION_FOR_MINISTRAL_3 = (0, 13, 1)
+MINISTRAL_MODEL_ALIASES = (
+    DEFAULT_MINISTRAL_MODEL,
+    "ministral-3",
+    "ministral-3:latest",
+    "ministral-3:8b",
+    "ministral-3:8b-instruct",
+    "ministral",
+    "ministral-8b",
+    "ministral-8b-instruct",
+    LEGACY_MINISTRAL_MODEL,
+)
+
+
+class OllamaBackend:
+    backend_name = "ollama"
+
+    def __init__(
+        self,
+        *,
+        model_name: str = DEFAULT_MINISTRAL_MODEL,
+        base_url: str | None = None,
+        timeout_seconds: float = 120.0,
+    ) -> None:
+        self.model_name = model_name
+        self.base_url = (base_url or os.getenv("OLLAMA_HOST") or DEFAULT_OLLAMA_HOST).rstrip("/")
+        self.timeout_seconds = timeout_seconds
+        self.resolved_model_name: str | None = None
+
+    def _pull_hint(self) -> str:
+        return f"ollama pull {self.model_name}"
+
+    def _requires_ministral_3_runtime(self) -> bool:
+        requested = self.model_name.strip().lower()
+        return requested.startswith("ministral-3") or requested == "ministral"
+
+    @staticmethod
+    def _parse_version(raw_version: str) -> tuple[int, ...] | None:
+        value = raw_version.strip().lower().lstrip("v")
+        numeric_parts: list[int] = []
+        for part in value.split("."):
+            digits = ""
+            for char in part:
+                if char.isdigit():
+                    digits += char
+                else:
+                    break
+            if not digits:
+                break
+            numeric_parts.append(int(digits))
+        if not numeric_parts:
+            return None
+        return tuple(numeric_parts)
+
+    def _request_json(
+        self,
+        path: str,
+        payload: dict[str, object] | None = None,
+        *,
+        method: str = "POST",
+    ) -> dict[str, object]:
+        url = f"{self.base_url}{path}"
+        body = None
+        headers = {"Accept": "application/json"}
+        if payload is not None:
+            body = json.dumps(payload).encode("utf-8")
+            headers["Content-Type"] = "application/json"
+        req = request.Request(url, data=body, headers=headers, method=method)
+        try:
+            with request.urlopen(req, timeout=self.timeout_seconds) as response:
+                text = response.read().decode("utf-8")
+        except error.HTTPError as exc:
+            detail = exc.read().decode("utf-8", errors="replace").strip()
+            if exc.code == 404 and path == "/api/generate":
+                raise RuntimeError(
+                    f"Ollama model '{self.model_name}' is not available locally. "
+                    f"Run: ollama pull {self.model_name}"
+                ) from exc
+            raise RuntimeError(f"Ollama request failed ({exc.code}): {detail or exc.reason}") from exc
+        except error.URLError as exc:
+            raise RuntimeError(
+                f"Could not reach Ollama at {self.base_url}. "
+                "Start Ollama or set OLLAMA_HOST to the correct endpoint."
+            ) from exc
+
+        try:
+            payload_json = json.loads(text)
+        except json.JSONDecodeError as exc:
+            raise RuntimeError("Ollama returned invalid JSON.") from exc
+        if not isinstance(payload_json, dict):
+            raise RuntimeError("Ollama returned an unexpected response shape.")
+        return payload_json
+
+    def available(self) -> bool:
+        try:
+            self._request_json("/api/tags", method="GET")
+        except Exception:
+            return False
+        return True
+
+    def server_version(self) -> str | None:
+        try:
+            response = self._request_json("/api/version", method="GET")
+        except Exception:
+            return None
+        raw_version = response.get("version")
+        if isinstance(raw_version, str) and raw_version.strip():
+            return raw_version.strip()
+        return None
+
+    def version_supported(self) -> tuple[bool | None, str | None]:
+        version = self.server_version()
+        if version is None:
+            return None, None
+        if not self._requires_ministral_3_runtime():
+            return True, version
+        parsed = self._parse_version(version)
+        if parsed is None:
+            return None, version
+        return parsed >= MIN_OLLAMA_VERSION_FOR_MINISTRAL_3, version
+
+    def list_models(self) -> list[str]:
+        response = self._request_json("/api/tags", method="GET")
+        raw_models = response.get("models", [])
+        if not isinstance(raw_models, list):
+            raise RuntimeError("Ollama returned an unexpected model list.")
+
+        discovered: list[str] = []
+        for entry in raw_models:
+            if not isinstance(entry, dict):
+                continue
+            name = entry.get("name")
+            if isinstance(name, str) and name.strip():
+                discovered.append(name.strip())
+        return discovered
+
+    def resolve_model_name(self) -> str | None:
+        installed = self.list_models()
+        installed_lookup = {name.lower(): name for name in installed}
+
+        if self.model_name.lower() in installed_lookup:
+            return installed_lookup[self.model_name.lower()]
+
+        requested = self.model_name.strip().lower()
+        if requested in {
+            "ministral",
+            "ministral-3",
+            "ministral-3:latest",
+            "ministral-3:8b",
+            "ministral-3:8b-instruct",
+            "ministral-8b",
+            "ministral-8b-instruct",
+        }:
+            for alias in MINISTRAL_MODEL_ALIASES:
+                resolved = installed_lookup.get(alias.lower())
+                if resolved is not None:
+                    return resolved
+
+        for installed_name in installed:
+            lowered = installed_name.lower()
+            if "ministral-3" in lowered and "8b" in lowered:
+                return installed_name
+        for installed_name in installed:
+            lowered = installed_name.lower()
+            if "ministral" in lowered and "8b" in lowered:
+                return installed_name
+
+        return None
+
+    def ensure_model_ready(self) -> str:
+        version_ok, version = self.version_supported()
+        if version_ok is False:
+            required_version = ".".join(str(part) for part in MIN_OLLAMA_VERSION_FOR_MINISTRAL_3)
+            raise RuntimeError(
+                "The open Ministral 3 local model requires a newer Ollama runtime.\n"
+                f"Detected Ollama: {version}\n"
+                f"Required Ollama: >= {required_version}"
+            )
+        try:
+            resolved = self.resolve_model_name()
+        except RuntimeError:
+            raise
+        except Exception as exc:
+            raise RuntimeError(f"Failed to inspect local Ollama models: {exc}") from exc
+
+        if resolved is None:
+            self.resolved_model_name = None
+            installed = self.list_models()
+            installed_hint = ", ".join(installed) if installed else "none"
+            raise RuntimeError(
+                "Ollama is running, but the requested Ministral model is not installed locally.\n"
+                f"Requested: {self.model_name}\n"
+                f"Installed: {installed_hint}\n"
+                f"Run: {self._pull_hint()}"
+            )
+        self.resolved_model_name = resolved
+        return resolved
+
+    def healthcheck(self) -> dict[str, object]:
+        version_ok, version = self.version_supported()
+        installed = self.list_models()
+        resolved = self.resolve_model_name() if installed else None
+        return {
+            "available": True,
+            "base_url": self.base_url,
+            "requested_model": self.model_name,
+            "resolved_model": resolved,
+            "installed_models": installed,
+            "ready": resolved is not None,
+            "pull_command": None if resolved is not None else self._pull_hint(),
+            "timeout_seconds": self.timeout_seconds,
+            "server_version": version,
+            "version_ok": version_ok,
+        }
+
+    def complete(self, *, system_prompt: str, user_prompt: str) -> str:
+        resolved_model = self.ensure_model_ready()
+        payload = {
+            "model": resolved_model,
+            "system": system_prompt,
+            "prompt": user_prompt,
+            "stream": False,
+        }
+        response = self._request_json("/api/generate", payload)
+        text = response.get("response")
+        if not isinstance(text, str) or not text.strip():
+            raise RuntimeError("Ollama returned an empty completion.")
+        return text.strip()