Merge pull request #9 from joonsoome/rerank-model-use-update

feat: v1.5.0 — MLX Reranker, OpenAI rerank path, macOS service, 2560‑D embeddings

Author: joonsoome

.env.example

@@ -8,6 +8,35 @@ MODEL_NAME=mlx-community/Qwen3-Embedding-4B-4bit-DWQ
 MODEL_PATH=
 CROSS_ENCODER_MODEL=
 
+# Reranker (Cross-Encoder) — Optional
+# Choose one of the following ways to enable reranking:
+# 1) Torch CrossEncoder (sentence-transformers)
+#    RERANKER_BACKEND=torch
+#    CROSS_ENCODER_MODEL=cross-encoder/ms-marco-MiniLM-L-6-v2
+#
+# 2) MLX Reranker (experimental v1; pooled token embeddings + linear head)
+#    RERANKER_BACKEND=mlx
+#    RERANKER_MODEL_ID=vserifsaglam/Qwen3-Reranker-4B-4bit-MLX
+#
+# Auto selection prefers Torch for stability (set RERANKER_BACKEND=auto)
+RERANKER_BACKEND=auto
+RERANKER_MODEL_ID=
+# Alias for convenience; same as RERANKER_MODEL_ID
+RERANKER_MODEL_NAME=
+# Optional overrides
+RERANK_MAX_SEQ_LEN=512
+RERANK_BATCH_SIZE=16
+# MLX-only experimental options:
+# - RERANK_POOLING: mean | cls (default: mean)
+# - RERANK_SCORE_NORM: none | sigmoid | minmax (default: none)
+#   Use sigmoid to bound scores to [0,1] for schema-constrained clients.
+RERANK_POOLING=mean
+RERANK_SCORE_NORM=none
+
+# OpenAI compatibility (scores normalization on native path for OpenAI clients)
+# true | false (default true)
+OPENAI_RERANK_AUTO_SIGMOID=true
+
 # Model Cache & Storage
 # MODEL_PATH: Custom path for MLX models (overrides auto cache detection)
 # If empty, uses Hugging Face cache or environment variables below:

.github/workflows/publish.yml

@@ -0,0 +1,40 @@
+name: Publish to PyPI
+
+on:
+  push:
+    tags:
+      - "v*"
+  workflow_dispatch:
+
+jobs:
+  build-and-publish:
+    name: Build and publish Python package
+    runs-on: ubuntu-latest
+
+    permissions:
+      contents: read
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.13"
+
+      - name: Install build tools
+        run: |
+          python -m pip install --upgrade pip
+          python -m pip install build
+
+      - name: Build sdist and wheel
+        run: python -m build
+
+      - name: Publish to PyPI
+        if: ${{ github.event_name == 'push' || github.event_name == 'workflow_dispatch' }}
+        uses: pypa/gh-action-pypi-publish@v1.12.4
+        with:
+          user: __token__
+          password: ${{ secrets.PYPI_API_TOKEN }}
+          skip-existing: true