Add Python bindings: from quantcpp import Model

bindings/python/quantcpp/ — ctypes wrapper for libturboquant:
- Model("model.gguf") — load GGUF model
- model.ask("question") — one-shot generation
- model.chat("question") — auto chat template
- model.generate("prompt") — streaming iterator
- KV compression support

Install: cd bindings/python && pip install -e .

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: unamedkr

bindings/python/README.md

@@ -0,0 +1,109 @@
+# quantcpp
+
+Python bindings for [quant.cpp](https://github.com/hunscompany/quant.cpp) -- a minimal C inference engine for local LLMs with KV cache compression.
+
+## Installation
+
+```bash
+# Build the shared library first
+cd /path/to/quant.cpp
+cmake -B build -DCMAKE_BUILD_TYPE=Release -DBUILD_SHARED_LIBS=ON
+cmake --build build -j$(nproc)
+
+# Install the Python package
+cd bindings/python
+pip install .
+```
+
+Or set the library path explicitly:
+
+```bash
+export TURBOQUANT_LIB=/path/to/build/libturboquant.dylib
+pip install .
+```
+
+## Usage
+
+### Basic generation
+
+```python
+from quantcpp import Model
+
+m = Model("model.gguf")
+text = m.ask("What is 2+2?")
+print(text)
+```
+
+### KV cache compression
+
+```python
+m = Model("model.gguf", kv_compress="4bit")
+text = m.ask("Explain quantum computing")
+```
+
+Available compression modes: `"1bit"`, `"2bit"`, `"3bit"` (default), `"4bit"`, `"polar3"`, `"polar4"`, `"qjl"`, `"turbo3"`, `"turbo4"`, `"uniform2"`, `"uniform3"`, `"uniform4"`, `"none"`.
+
+### Streaming
+
+```python
+for token in m.generate("Once upon a time"):
+    print(token, end="", flush=True)
+```
+
+### Chat mode
+
+```python
+text = m.chat("What is the capital of France?")
+print(text)
+```
+
+### Raw completion (no chat template)
+
+```python
+text = m.complete("The quick brown fox", max_tokens=64)
+print(text)
+```
+
+### Context manager
+
+```python
+with Model("model.gguf") as m:
+    print(m.ask("Hello!"))
+```
+
+## API Reference
+
+### `Model(path, kv_compress=None, n_threads=0)`
+
+Load a GGUF or TQM model file.
+
+- `path` -- Path to model file.
+- `kv_compress` -- KV cache compression mode (see above).
+- `n_threads` -- CPU thread count (0 = auto).
+
+### `Model.ask(prompt, *, max_tokens=512, temperature=0.6, top_p=0.9)`
+
+Chat-formatted generation. Returns the full response string.
+
+### `Model.chat(message, **kwargs)`
+
+Alias for `ask()`.
+
+### `Model.generate(prompt, *, max_tokens=512, temperature=0.6, top_p=0.9, chat=False)`
+
+Streaming generation. Yields token strings. Set `chat=True` to apply chat template.
+
+### `Model.complete(prompt, **kwargs)`
+
+Raw text completion without chat template.
+
+### `Model.close()`
+
+Release model resources. Called automatically on garbage collection or context exit.
+
+## Library search order
+
+1. `TURBOQUANT_LIB` environment variable
+2. Same directory as the Python package
+3. `build/` relative to the project root
+4. System library path