openaivec 1.0.10__tar.gz → 1.0.12__tar.gz

openaivec-1.0.12/.github/workflows/publish.yml

@@ -0,0 +1,59 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*.*.*"
+
+permissions:
+  contents: write
+  id-token: write
+
+jobs:
+  build-and-publish:
+    runs-on: ubuntu-latest
+    environment: pypi
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v7
+
+      - name: Set up Python
+        run: uv python install 3.10
+
+      - name: Install dependencies via uv
+        run: uv sync --all-extras --dev
+
+      - name: Build with uv
+        run: uv build
+
+      - name: Install cosign
+        uses: sigstore/cosign-installer@v3
+
+      - name: Sign artifacts with cosign (OIDC)
+        run: |
+          set -euo pipefail
+          ls -la dist
+          mkdir -p signatures
+          for f in dist/*; do
+            if [ -f "$f" ]; then
+              base="$(basename "$f")"
+              cosign sign-blob --yes \
+                --output-signature "signatures/${base}.sig" \
+                --output-certificate "signatures/${base}.pem" \
+                "$f"
+            fi
+          done
+
+      - name: Create GitHub Release
+        uses: softprops/action-gh-release@v2
+        with:
+          files: |
+            dist/*
+            signatures/*
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@release/v1

{openaivec-1.0.10 → openaivec-1.0.12}/.github/workflows/test-pr.yml

@@ -6,6 +6,7 @@ on:
 
 permissions:
   contents: read
+  issues: write
   pull-requests: read
 
 jobs:
@@ -23,6 +24,19 @@ jobs:
       OPENAI_API_KEY: ${{ secrets.OPENAI_API_KEY }}
 
     steps:
+      - name: Comment workflow run link
+        shell: bash
+        run: |
+          set -euo pipefail
+          COMMENTS_URL="${{ github.event.issue.comments_url }}"
+          RUN_URL="https://github.com/${{ github.repository }}/actions/runs/${{ github.run_id }}"
+          export RUN_URL
+          PAYLOAD="$(python -c 'import json,os; print(json.dumps({\"body\": f\"Integration workflow started: {os.environ[\\\"RUN_URL\\\"]}\"}))')"
+          curl -sS -H "Authorization: Bearer ${{ secrets.GITHUB_TOKEN }}" \
+               -H "Accept: application/vnd.github+json" \
+               -d "$PAYLOAD" \
+               "$COMMENTS_URL"
+
       - name: Fetch PR head SHA
         id: pr
         shell: bash

{openaivec-1.0.10 → openaivec-1.0.12}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: openaivec
-Version: 1.0.10
+Version: 1.0.12
 Summary: Generative mutation for tabular calculation
 Project-URL: Homepage, https://microsoft.github.io/openaivec/
 Project-URL: Repository, https://github.com/microsoft/openaivec
@@ -63,7 +63,7 @@ print(sentiment.tolist())
 # Output: ['Positive sentiment', 'Negative sentiment']
 ```
 
-**Try it live:** https://microsoft.github.io/openaivec/examples/pandas/
+**Pandas tutorial (GitHub Pages):** https://microsoft.github.io/openaivec/examples/pandas/
 
 ## Benchmarks
 
@@ -82,6 +82,7 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Contents
 
 - [Why openaivec?](#why-openaivec)
+- [Overview](#overview)
 - [Core Workflows](#core-workflows)
 - [Using with Apache Spark UDFs](#using-with-apache-spark-udfs)
 - [Building Prompts](#building-prompts)
@@ -93,14 +94,13 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Why openaivec?
 
 - Drop-in `.ai` and `.aio` accessors keep pandas analysts in familiar tooling.
-- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, and keep column order stable.
-- Smart batching (`BatchingMapProxy`/`AsyncBatchingMapProxy`) dedupes prompts, preserves order, and releases waiters on failure.
+- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, preserve order, and release waiters on failure.
 - Reasoning support mirrors the OpenAI SDK; structured outputs accept Pydantic `response_format`.
 - Built-in caches and retries remove boilerplate; helpers reuse caches across pandas, Spark, and async flows.
 - Spark UDFs and Microsoft Fabric guides move notebooks into production-scale ETL.
 - Prompt tooling (`FewShotPromptBuilder`, `improve`) and the task library ship curated prompts with validated outputs.
 
-# Overview
+## Overview
 
 Vectorized OpenAI batch processing so you handle many inputs per call instead of one-by-one. Batching proxies dedupe inputs, enforce ordered outputs, and unblock waiters even on upstream errors. Cache helpers (`responses_with_cache`, Spark UDF builders) plug into the same layer so expensive prompts are reused across pandas, Spark, and async flows. Reasoning models honor SDK semantics. Requires Python 3.10+.
 
@@ -186,7 +186,7 @@ result = df.assign(
 
 ### Using with reasoning models
 
-Reasoning models (o1-preview, o1-mini, o3-mini, etc.) work without special flags. `reasoning` mirrors the OpenAI SDK.
+Reasoning models (o1-preview, o1-mini, o3-mini, etc.) follow OpenAI SDK semantics. Pass `reasoning` when you want to override model defaults.
 
 ```python
 pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
@@ -194,7 +194,7 @@ pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
 result = df.assign(
     analysis=lambda df: df.text.ai.responses(
         "Analyze this text step by step",
-        reasoning={"effort": "none"}  # Optional: mirrors the OpenAI SDK argument
+        reasoning={"effort": "none"},  # Optional: mirrors the OpenAI SDK argument
     )
 )
 ```
@@ -254,7 +254,7 @@ df = pd.DataFrame({"text": [
 async def process_data():
     return await df["text"].aio.responses(
         "Analyze sentiment and classify as positive/negative/neutral",
-        reasoning={"effort": "none"},  # Required for gpt-5.1
+        reasoning={"effort": "none"},  # Recommended for reasoning models
         max_concurrency=12    # Allow up to 12 concurrent requests
     )
 

{openaivec-1.0.10 → openaivec-1.0.12}/README.md

@@ -37,7 +37,7 @@ print(sentiment.tolist())
 # Output: ['Positive sentiment', 'Negative sentiment']
 ```
 
-**Try it live:** https://microsoft.github.io/openaivec/examples/pandas/
+**Pandas tutorial (GitHub Pages):** https://microsoft.github.io/openaivec/examples/pandas/
 
 ## Benchmarks
 
@@ -56,6 +56,7 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Contents
 
 - [Why openaivec?](#why-openaivec)
+- [Overview](#overview)
 - [Core Workflows](#core-workflows)
 - [Using with Apache Spark UDFs](#using-with-apache-spark-udfs)
 - [Building Prompts](#building-prompts)
@@ -67,14 +68,13 @@ Batching alone removes most HTTP overhead, and letting batching overlap with con
 ## Why openaivec?
 
 - Drop-in `.ai` and `.aio` accessors keep pandas analysts in familiar tooling.
-- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, and keep column order stable.
-- Smart batching (`BatchingMapProxy`/`AsyncBatchingMapProxy`) dedupes prompts, preserves order, and releases waiters on failure.
+- OpenAI batch-optimized: `BatchingMapProxy`/`AsyncBatchingMapProxy` coalesce requests, dedupe prompts, preserve order, and release waiters on failure.
 - Reasoning support mirrors the OpenAI SDK; structured outputs accept Pydantic `response_format`.
 - Built-in caches and retries remove boilerplate; helpers reuse caches across pandas, Spark, and async flows.
 - Spark UDFs and Microsoft Fabric guides move notebooks into production-scale ETL.
 - Prompt tooling (`FewShotPromptBuilder`, `improve`) and the task library ship curated prompts with validated outputs.
 
-# Overview
+## Overview
 
 Vectorized OpenAI batch processing so you handle many inputs per call instead of one-by-one. Batching proxies dedupe inputs, enforce ordered outputs, and unblock waiters even on upstream errors. Cache helpers (`responses_with_cache`, Spark UDF builders) plug into the same layer so expensive prompts are reused across pandas, Spark, and async flows. Reasoning models honor SDK semantics. Requires Python 3.10+.
 
@@ -160,7 +160,7 @@ result = df.assign(
 
 ### Using with reasoning models
 
-Reasoning models (o1-preview, o1-mini, o3-mini, etc.) work without special flags. `reasoning` mirrors the OpenAI SDK.
+Reasoning models (o1-preview, o1-mini, o3-mini, etc.) follow OpenAI SDK semantics. Pass `reasoning` when you want to override model defaults.
 
 ```python
 pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
@@ -168,7 +168,7 @@ pandas_ext.set_responses_model("o1-mini")  # Set your reasoning model
 result = df.assign(
     analysis=lambda df: df.text.ai.responses(
         "Analyze this text step by step",
-        reasoning={"effort": "none"}  # Optional: mirrors the OpenAI SDK argument
+        reasoning={"effort": "none"},  # Optional: mirrors the OpenAI SDK argument
     )
 )
 ```
@@ -228,7 +228,7 @@ df = pd.DataFrame({"text": [
 async def process_data():
     return await df["text"].aio.responses(
         "Analyze sentiment and classify as positive/negative/neutral",
-        reasoning={"effort": "none"},  # Required for gpt-5.1
+        reasoning={"effort": "none"},  # Recommended for reasoning models
         max_concurrency=12    # Allow up to 12 concurrent requests
     )
 

{openaivec-1.0.10 → openaivec-1.0.12}/src/openaivec/_cache/optimize.py

@@ -21,7 +21,10 @@ class BatchSizeSuggester:
     min_batch_size: int = 10
     min_duration: float = 30.0
     max_duration: float = 60.0
-    step_ratio: float = 0.2
+    step_ratio_up: float = 0.1
+    step_ratio_down: float = 0.2
+    max_step: int | None = None
+    min_step: int = 1
     sample_size: int = 4
     _history: list[PerformanceMetric] = field(default_factory=list)
     _lock: threading.RLock = field(default_factory=threading.RLock, repr=False)
@@ -34,8 +37,14 @@ class BatchSizeSuggester:
             raise ValueError("current_batch_size must be >= min_batch_size")
         if self.sample_size <= 0:
             raise ValueError("sample_size must be > 0")
-        if self.step_ratio <= 0:
-            raise ValueError("step_ratio must be > 0")
+        if self.step_ratio_up <= 0:
+            raise ValueError("step_ratio_up must be > 0")
+        if self.step_ratio_down <= 0:
+            raise ValueError("step_ratio_down must be > 0")
+        if self.max_step is not None and self.max_step <= 0:
+            raise ValueError("max_step must be > 0")
+        if self.min_step <= 0:
+            raise ValueError("min_step must be > 0")
         if self.min_duration <= 0 or self.max_duration <= 0:
             raise ValueError("min_duration and max_duration must be > 0")
         if self.min_duration >= self.max_duration:
@@ -94,9 +103,15 @@ class BatchSizeSuggester:
             current_size = self.current_batch_size
 
             if average_duration < self.min_duration:
-                new_batch_size = int(current_size * (1 + self.step_ratio))
+                delta = max(self.min_step, int(current_size * self.step_ratio_up))
+                if self.max_step is not None:
+                    delta = min(delta, self.max_step)
+                new_batch_size = current_size + delta
             elif average_duration > self.max_duration:
-                new_batch_size = int(current_size * (1 - self.step_ratio))
+                delta = max(self.min_step, int(current_size * self.step_ratio_down))
+                if self.max_step is not None:
+                    delta = min(delta, self.max_step)
+                new_batch_size = current_size - delta
             else:
                 new_batch_size = current_size
 

{openaivec-1.0.10 → openaivec-1.0.12}/tests/_cache/test_optimize.py

@@ -33,21 +33,35 @@ class TestBatchSizeSuggester:
         assert suggester.min_batch_size == 10
         assert suggester.min_duration == 30.0
         assert suggester.max_duration == 60.0
-        assert suggester.step_ratio == 0.2
+        assert suggester.step_ratio_up == 0.1
+        assert suggester.step_ratio_down == 0.2
+        assert suggester.max_step is None
+        assert suggester.min_step == 1
         assert suggester.sample_size == 4
         assert len(suggester._history) == 0
         assert suggester._batch_size_changed_at is None
 
     def test_custom_initialization(self):
         suggester = BatchSizeSuggester(
-            current_batch_size=20, min_batch_size=5, min_duration=15.0, max_duration=45.0, step_ratio=0.2, sample_size=5
+            current_batch_size=20,
+            min_batch_size=5,
+            min_duration=15.0,
+            max_duration=45.0,
+            step_ratio_up=0.15,
+            step_ratio_down=0.25,
+            min_step=2,
+            max_step=5,
+            sample_size=5,
         )
 
         assert suggester.current_batch_size == 20
         assert suggester.min_batch_size == 5
         assert suggester.min_duration == 15.0
         assert suggester.max_duration == 45.0
-        assert suggester.step_ratio == 0.2
+        assert suggester.step_ratio_up == 0.15
+        assert suggester.step_ratio_down == 0.25
+        assert suggester.min_step == 2
+        assert suggester.max_step == 5
         assert suggester.sample_size == 5
 
     @pytest.mark.parametrize(
@@ -56,7 +70,10 @@ class TestBatchSizeSuggester:
             ({"min_batch_size": 0}, "min_batch_size must be > 0"),
             ({"current_batch_size": 5, "min_batch_size": 10}, "current_batch_size must be >= min_batch_size"),
             ({"sample_size": 0}, "sample_size must be > 0"),
-            ({"step_ratio": 0}, "step_ratio must be > 0"),
+            ({"step_ratio_up": 0}, "step_ratio_up must be > 0"),
+            ({"step_ratio_down": 0}, "step_ratio_down must be > 0"),
+            ({"min_step": 0}, "min_step must be > 0"),
+            ({"max_step": 0}, "max_step must be > 0"),
             ({"min_duration": 0}, "min_duration and max_duration must be > 0"),
             ({"max_duration": 0}, "min_duration and max_duration must be > 0"),
             ({"min_duration": 60, "max_duration": 30}, "min_duration must be < max_duration"),
@@ -190,7 +207,8 @@ class TestBatchSizeSuggester:
             min_batch_size=min_batch_size,
             min_duration=0.5,  # 0.5 seconds (test scale)
             max_duration=1.0,  # 1.0 seconds (test scale)
-            step_ratio=0.1,
+            step_ratio_up=0.1,
+            step_ratio_down=0.1,
             sample_size=3,
         )
 
@@ -214,7 +232,7 @@ class TestBatchSizeSuggester:
             min_batch_size=5,
             min_duration=0.5,  # 0.5 seconds (test scale)
             max_duration=1.0,  # 1.0 seconds (test scale)
-            step_ratio=0.5,  # Large step to force below minimum
+            step_ratio_down=0.5,  # Large step to force below minimum
             sample_size=3,
         )
 
@@ -227,6 +245,60 @@ class TestBatchSizeSuggester:
         assert new_size == 5  # Should not go below min_batch_size
         assert suggester.current_batch_size == 5
 
+    def test_min_step_enforces_minimum_delta_on_increase(self):
+        suggester = BatchSizeSuggester(
+            current_batch_size=10,
+            min_batch_size=1,
+            min_duration=0.5,
+            max_duration=1.0,
+            step_ratio_up=0.1,  # 10 * 0.1 = 1
+            min_step=5,  # force delta to 5
+            sample_size=3,
+        )
+
+        for i in range(3):
+            with suggester.record(batch_size=10):
+                time.sleep(0.1)  # fast -> increase
+
+        new_size = suggester.suggest_batch_size()
+        assert new_size == 15
+
+    def test_min_step_enforces_minimum_delta_on_decrease(self):
+        suggester = BatchSizeSuggester(
+            current_batch_size=20,
+            min_batch_size=1,
+            min_duration=0.5,
+            max_duration=1.0,
+            step_ratio_down=0.1,  # 20 * 0.1 = 2
+            min_step=5,  # force delta to 5
+            sample_size=3,
+        )
+
+        for i in range(3):
+            with suggester.record(batch_size=20):
+                time.sleep(1.5)  # slow -> decrease
+
+        new_size = suggester.suggest_batch_size()
+        assert new_size == 15
+
+    def test_max_step_caps_delta(self):
+        suggester = BatchSizeSuggester(
+            current_batch_size=50,
+            min_batch_size=1,
+            min_duration=0.5,
+            max_duration=1.0,
+            step_ratio_up=0.5,  # 50 * 0.5 = 25
+            max_step=10,  # cap delta at 10
+            sample_size=3,
+        )
+
+        for i in range(3):
+            with suggester.record(batch_size=50):
+                time.sleep(0.1)  # fast -> increase
+
+        new_size = suggester.suggest_batch_size()
+        assert new_size == 60
+
     def test_thread_safety(self):
         suggester = BatchSizeSuggester(sample_size=10)
         results = []

{openaivec-1.0.10 → openaivec-1.0.12}/tests/_cache/test_proxy_suggester.py

@@ -42,7 +42,8 @@ def test_sync_proxy_suggester_adapts_batch_size():
     proxy.suggester.min_duration = 0.001  # 1ms
     proxy.suggester.max_duration = 0.002  # 2ms
     proxy.suggester.sample_size = 2
-    proxy.suggester.step_ratio = 0.5
+    proxy.suggester.step_ratio_up = 0.5
+    proxy.suggester.step_ratio_down = 0.5
 
     import time