kipimo 0.2.0__tar.gz → 0.2.2__tar.gz

{kipimo-0.2.0/src/kipimo.egg-info → kipimo-0.2.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kipimo
-Version: 0.2.0
+Version: 0.2.2
 Summary: Swahili agent-task evaluation suite for the East Africa coordination stack — model-agnostic seed benchmark (46 tasks: server routing, term grounding, cascade routing).
 Author: AfriKaziOS maintainers
 License-Expression: MIT
@@ -41,6 +41,10 @@ kipimo score preds.jsonl          # per-type + overall report
 
 The harness never calls a model — you generate predictions with whatever system you're evaluating; kipimo only scores. Any lab can publish comparable numbers.
 
+## Data sovereignty
+
+The models deployable under African data-sovereignty constraints are open-weight and self-hostable. kipimo scores that tier on equal terms with frontier APIs so builders can test whether the model they are *allowed* to run is good enough — see [docs/DATA_SOVEREIGNTY.md](docs/DATA_SOVEREIGNTY.md) and `examples/generate_predictions.py`.
+
 ## Honesty box
 - **v0.1 is a seed set.** 46 tasks establish the format and scoring; breadth comes from contributions.
 - Swahili phrasing is simple-register and **pending native-speaker review** — that is [issue #1](https://github.com/gabrielmahia/kipimo/issues), and corrections are the most valuable contribution possible.

{kipimo-0.2.0 → kipimo-0.2.2}/README.md

@@ -23,6 +23,10 @@ kipimo score preds.jsonl          # per-type + overall report
 
 The harness never calls a model — you generate predictions with whatever system you're evaluating; kipimo only scores. Any lab can publish comparable numbers.
 
+## Data sovereignty
+
+The models deployable under African data-sovereignty constraints are open-weight and self-hostable. kipimo scores that tier on equal terms with frontier APIs so builders can test whether the model they are *allowed* to run is good enough — see [docs/DATA_SOVEREIGNTY.md](docs/DATA_SOVEREIGNTY.md) and `examples/generate_predictions.py`.
+
 ## Honesty box
 - **v0.1 is a seed set.** 46 tasks establish the format and scoring; breadth comes from contributions.
 - Swahili phrasing is simple-register and **pending native-speaker review** — that is [issue #1](https://github.com/gabrielmahia/kipimo/issues), and corrections are the most valuable contribution possible.

{kipimo-0.2.0 → kipimo-0.2.2}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "kipimo"
-version = "0.2.0"
+version = "0.2.2"
 description = "Swahili agent-task evaluation suite for the East Africa coordination stack — model-agnostic seed benchmark (46 tasks: server routing, term grounding, cascade routing)."
 readme = "README.md"
 requires-python = ">=3.10"

{kipimo-0.2.0 → kipimo-0.2.2}/src/kipimo/targets.py

@@ -40,13 +40,21 @@ TARGETS: list[dict] = [
         "notes": "reference frontier closed model",
     },
     # --- open weights (frontier-adjacent, downloadable) ---
+    {
+        "id": "kimi-k2.6",
+        "label": "Kimi K2.6 (Moonshot)",
+        "family": "open-weight",
+        "params_b": 1000,
+        "endpoint_env": ["KIPIMO_BASE_URL", "KIPIMO_API_KEY"],
+        "notes": "weights public (HF, ~900k downloads); 1T MoE; strong agentic/tool-use; self-hostable via vLLM/SGLang/Ollama",
+    },
     {
         "id": "kimi-k3",
         "label": "Kimi K3 (Moonshot)",
         "family": "open-weight",
         "params_b": 2800,
         "endpoint_env": ["KIPIMO_BASE_URL", "KIPIMO_API_KEY"],
-        "notes": "weights public from 2026-07-27; MoE",
+        "notes": "2.8T MoE, 104B active; weights PUBLIC on HF (moonshotai/Kimi-K3, native multimodal, 1M ctx) under the custom \"Kimi K3 License\" — read LICENSE before redistribution (NOT MIT/Apache); strong agentic/MCP scores make it a valid frontier reference",
     },
     {
         "id": "glm-5.2",
@@ -97,6 +105,33 @@ TARGETS: list[dict] = [
         "endpoint_env": ["KIPIMO_BASE_URL", "KIPIMO_API_KEY"],
         "notes": "runnable on a single consumer GPU",
     },
+    {
+        "id": "tiny-aya-earth",
+        "label": "Tiny Aya Earth (Cohere Labs)",
+        "family": "small-open",
+        "params_b": 3.35,
+        "endpoint_env": ["KIPIMO_BASE_URL", "KIPIMO_API_KEY"],
+        "notes": (
+            "Cohere Labs Tiny Aya (Feb 2026), Earth variant tuned for West-Asian + "
+            "African languages incl. Swahili; 70+ langs, 8K ctx, mobile/edge-runnable, "
+            "CC-BY-NC-4.0. Sits between InkubaLM 0.4B and the ~12B tier. Verify exact "
+            "HF repo id before use."
+        ),
+    },
+    {
+        "id": "inkubalm-0.4b",
+        "label": "InkubaLM 0.4B (Lelapa AI)",
+        "family": "small-open",
+        "params_b": 0.4,
+        "endpoint_env": ["KIPIMO_BASE_URL", "KIPIMO_API_KEY"],
+        "notes": (
+            "African-built SLM (hf: lelapa/InkubaLM-0.4B) covering Swahili + "
+            "isiZulu/Yoruba/Hausa/isiXhosa; MobileLLM-class, mobile/edge-runnable "
+            "(community variants down to ~40M). License CC BY-NC 4.0 — evaluatable, "
+            "non-commercial redistribution. The floor of the deployment tier this "
+            "benchmark exists to measure."
+        ),
+    },
 ]
 
 

{kipimo-0.2.0 → kipimo-0.2.2/src/kipimo.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: kipimo
-Version: 0.2.0
+Version: 0.2.2
 Summary: Swahili agent-task evaluation suite for the East Africa coordination stack — model-agnostic seed benchmark (46 tasks: server routing, term grounding, cascade routing).
 Author: AfriKaziOS maintainers
 License-Expression: MIT
@@ -41,6 +41,10 @@ kipimo score preds.jsonl          # per-type + overall report
 
 The harness never calls a model — you generate predictions with whatever system you're evaluating; kipimo only scores. Any lab can publish comparable numbers.
 
+## Data sovereignty
+
+The models deployable under African data-sovereignty constraints are open-weight and self-hostable. kipimo scores that tier on equal terms with frontier APIs so builders can test whether the model they are *allowed* to run is good enough — see [docs/DATA_SOVEREIGNTY.md](docs/DATA_SOVEREIGNTY.md) and `examples/generate_predictions.py`.
+
 ## Honesty box
 - **v0.1 is a seed set.** 46 tasks establish the format and scoring; breadth comes from contributions.
 - Swahili phrasing is simple-register and **pending native-speaker review** — that is [issue #1](https://github.com/gabrielmahia/kipimo/issues), and corrections are the most valuable contribution possible.