jevkit-runtime 0.3.0__tar.gz → 0.3.1__tar.gz

jevkit_runtime-0.3.1/CHANGELOG.md

@@ -0,0 +1,42 @@
+# Changelog
+
+`jevkit-runtime` is 0.x: a minor version may break the API. The JevKit tools pin `<0.(n+1)` so a
+new minor release never changes an installed tool.
+
+## 0.3.1
+
+- Documentation and package metadata only; the code is the same as 0.3.0.
+- The DiffusionGemma and Laya guides report what to expect from the four tools' benchmarks.
+- Add this changelog and a security policy, and list the repository and changelog on PyPI.
+- README links point to GitHub, so they also work on PyPI.
+
+## 0.3.0
+
+- Add the local decision servers `diffusiongemma` (OpenJev) and `laya` (laya-mlx) to the catalog.
+  They need no key, are chosen only by name and never in place of a configured hosted provider, and
+  are priced at zero API fees.
+- A provider may carry its own price; `JEV_PRICE_PER_MTOK` overrides it and `Settings.list_price`
+  is the fallback.
+- Providers with joint reads key every answer on the whole ordered batch and are re-sent whole when
+  any slot is missing.
+- Setup guides for both servers in `docs/`, and the Laya server adapter in `scripts/`.
+
+## 0.2.0
+
+Breaking redesign: one request pipeline, one answer store, one provider catalog.
+
+- `Client.ask` owns the whole pipeline: request sharing, hedging, cache-only runs, charge callbacks
+  and provenance are per-call keyword arguments. It returns `Answers`, a dict with an `origins`
+  attribute.
+- The import is now `jevkit_runtime`, matching the distribution.
+- The answer store keeps answer and provenance in one row under a versioned schema; caches from
+  0.1 are reset and re-asked.
+- HTTP/2 is used whenever the `http2` extra is installed.
+- Removed: the legacy key recipes, the two-table cache, the `DecisionClient` seam, and the unused
+  local providers.
+
+## 0.1.0
+
+- First release: shared transport, settings, answer cache and metering for jgrep, jsort, jlink,
+  jselect and jcol, published as `jevkit-runtime` (the PyPI name `jevkit-core` belongs to another
+  project).

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/PKG-INFO

@@ -1,8 +1,10 @@
 Metadata-Version: 2.5
 Name: jevkit-runtime
-Version: 0.3.0
+Version: 0.3.1
 Summary: Shared transport, configuration, caching, and accounting for JevKit tools
 Project-URL: Homepage, https://github.com/keltokhy/jevkit-core
+Project-URL: Repository, https://github.com/keltokhy/jevkit-core
+Project-URL: Changelog, https://github.com/keltokhy/jevkit-core/blob/main/CHANGELOG.md
 Project-URL: Issues, https://github.com/keltokhy/jevkit-core/issues
 Author: Khaled Eltokhy
 License-Expression: MIT
@@ -82,13 +84,13 @@ in place of a configured hosted provider, need no key, and are metered at zero A
 DiffusionGemma reads every question in a batch together, so the runtime keys each of its answers
 on the whole ordered batch and re-sends a batch whole when any slot is missing.
 
-No package ships the models. [docs/diffusiongemma.md](docs/diffusiongemma.md) and
-[docs/laya.md](docs/laya.md) explain how to run the servers, and `scripts/laya_server.py` is the
+No package ships the models. [docs/diffusiongemma.md](https://github.com/keltokhy/jevkit-core/blob/main/docs/diffusiongemma.md) and
+[docs/laya.md](https://github.com/keltokhy/jevkit-core/blob/main/docs/laya.md) explain how to run the servers, and `scripts/laya_server.py` is the
 adapter the Laya guide starts.
 
 ## Development
 
-Keep the six checkouts as siblings. Each consumer depends on `jevkit-runtime>=0.2.0,<0.3.0` and
+Keep the six checkouts as siblings. Each consumer depends on `jevkit-runtime>=0.3.0,<0.4.0` and
 overrides it for development with `jevkit-runtime = { path = "../jevkit-core", editable = true }`
 under `[tool.uv.sources]`.
 

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/README.md

@@ -67,13 +67,13 @@ in place of a configured hosted provider, need no key, and are metered at zero A
 DiffusionGemma reads every question in a batch together, so the runtime keys each of its answers
 on the whole ordered batch and re-sends a batch whole when any slot is missing.
 
-No package ships the models. [docs/diffusiongemma.md](docs/diffusiongemma.md) and
-[docs/laya.md](docs/laya.md) explain how to run the servers, and `scripts/laya_server.py` is the
+No package ships the models. [docs/diffusiongemma.md](https://github.com/keltokhy/jevkit-core/blob/main/docs/diffusiongemma.md) and
+[docs/laya.md](https://github.com/keltokhy/jevkit-core/blob/main/docs/laya.md) explain how to run the servers, and `scripts/laya_server.py` is the
 adapter the Laya guide starts.
 
 ## Development
 
-Keep the six checkouts as siblings. Each consumer depends on `jevkit-runtime>=0.2.0,<0.3.0` and
+Keep the six checkouts as siblings. Each consumer depends on `jevkit-runtime>=0.3.0,<0.4.0` and
 overrides it for development with `jevkit-runtime = { path = "../jevkit-core", editable = true }`
 under `[tool.uv.sources]`.
 

jevkit_runtime-0.3.1/SECURITY.md

@@ -0,0 +1,18 @@
+# Security
+
+## Reporting a vulnerability
+
+Please report security problems privately through GitHub:
+[Report a vulnerability](https://github.com/keltokhy/jevkit-core/security/advisories/new).
+Do not open a public issue. This is maintained by one person in spare time; expect a first reply
+within a week.
+
+Only the latest release is supported. Fixes ship as a new release, not as patches to older ones.
+
+## What to know when using it
+
+- Text you pass in is sent to the model provider you configure (TypeSafe, OpenRouter, a gateway,
+  or a server on your own machine). Do not send data you are not allowed to share with that provider.
+- API keys are read from environment variables or `~/.config/jev/<provider>.key` and are sent only
+  to the provider they belong to. They are never written to the answer cache or to logs.
+- Answers are cached on disk in `~/.cache/jev`. Delete it to remove stored answers.

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/docs/diffusiongemma.md

@@ -80,7 +80,19 @@ same URL and model, use `--no-cache` or a separate `XDG_CACHE_HOME`.
 
 ## What to expect
 
-A September 2026 jgrep comparison over 10,000 public-data decisions found news classification
-close to hosted Jev but more false positives on SMS spam at a 0.9 cutoff. For code, judge whole
-diff hunks or functions; isolated diff lines recalled poorly on a small synthetic fixture. Keep
-it experimental and evaluate its cutoff on your own data.
+On September 22, 2026, DiffusionGemma (`openjev-0.1`) ran each tool's benchmarks on an Apple M3
+Ultra with 96 GiB of unified memory, beside Jev 1.13's recorded runs. It is a reasonable
+substitute for Jev when the text must stay on your machine and each decision is about one record,
+or about long ones. In
+[jgrep](https://github.com/keltokhy/jgrep/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+its spam F1 was 0.87 against Jev's 0.91, with the same news accuracy; in
+[jlink](https://github.com/keltokhy/jlink/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+it came within 0.03 of Jev's F1 on four of five benchmarks; in
+[jcol](https://github.com/keltokhy/jcol/blob/main/benchmarks/README.md#local-models-2026-09-22)
+it agreed with the product label on 78 of 100 complaints against Jev's 75; and in
+[jsort](https://github.com/keltokhy/jsort/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+its scores for whole FOMC statements correlated 0.89 with Jev's. It is not a substitute for
+sorting single short lines, where its comparisons had a reliability of 0.62 and 0.39 against
+Jev's 0.97 and 0.96, or for judging isolated diff lines. The server runs one call at a time on
+the GPU: about a third of a second per short record, and 14 to 18 minutes per 3,000 record pairs.
+Keep it experimental and evaluate its cutoff on your own data.

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/docs/laya.md

@@ -74,7 +74,20 @@ or a separate `XDG_CACHE_HOME`.
 
 ## What to expect
 
-A September 2026 jgrep experiment on an M3 Ultra found fast, strong short-text classification
-(92% top-1 on a news set) but weak precision on spam at the default 0.5 cutoff (53%, rising to
-79% at 0.9) and weak code judgments. Treat it as a fast classifier for short texts, and evaluate
-thresholds on your own labeled inputs before relying on the scores.
+On September 22, 2026, Laya (`laya-421m`) ran each tool's benchmarks on an Apple M3 Ultra with
+96 GiB of unified memory, beside Jev 1.13's recorded runs. It is a fast classifier of short texts
+into broad topics, and not a substitute for Jev elsewhere. In
+[jgrep](https://github.com/keltokhy/jgrep/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+it had the best news accuracy of the three models, 92% against Jev's 87%, and finished 2,000
+messages in 46 s; at the default cutoff it also flagged 235 messages that were not spam, against
+Jev's 41. Its 512-token window, question included, decides what it can read: the adapter refuses
+longer requests with HTTP 422, which the tools report as failed decisions, so in
+[jcol](https://github.com/keltokhy/jcol/blob/main/benchmarks/README.md#local-models-2026-09-22)
+it refused 27 of 100 complaint narratives and in
+[jsort](https://github.com/keltokhy/jsort/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+every whole FOMC statement. On what it could read, its pairwise comparisons in jsort did not
+follow Jev's order, and in
+[jlink](https://github.com/keltokhy/jlink/blob/main/docs/benchmarks/local-models-2026-09-22.md)
+it scored most candidate pairs as matches, for an F1 of 0.16 to 0.22 on firms, products and
+software where Jev's was 0.67 to 0.94. Evaluate thresholds on your own labeled inputs before
+relying on the scores.

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "jevkit-runtime"
-version = "0.3.0"
+version = "0.3.1"
 description = "Shared transport, configuration, caching, and accounting for JevKit tools"
 readme = "README.md"
 license = "MIT"
@@ -10,6 +10,8 @@ dependencies = ["httpx>=0.27"]
 
 [project.urls]
 Homepage = "https://github.com/keltokhy/jevkit-core"
+Repository = "https://github.com/keltokhy/jevkit-core"
+Changelog = "https://github.com/keltokhy/jevkit-core/blob/main/CHANGELOG.md"
 Issues = "https://github.com/keltokhy/jevkit-core/issues"
 
 [project.optional-dependencies]

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/src/jevkit_runtime/__init__.py

@@ -28,7 +28,7 @@ from .settings import DEFAULT_PRICE_PER_MTOK, Settings
 from .store import AnswerStore, Entry
 from .transport import FATAL, RETRYABLE, post
 
-__version__ = "0.3.0"
+__version__ = "0.3.1"
 __all__ = [
     "AnswerStore",
     "Answers",

{jevkit_runtime-0.3.0 → jevkit_runtime-0.3.1}/uv.lock

@@ -139,7 +139,7 @@ wheels = [
 
 [[package]]
 name = "jevkit-runtime"
-version = "0.3.0"
+version = "0.3.1"
 source = { editable = "." }
 dependencies = [
     { name = "httpx" },