mlx-stack 0.1.0__py3-none-any.whl

mlx_stack-0.1.0.dist-info/METADATA

@@ -0,0 +1,397 @@
+Metadata-Version: 2.4
+Name: mlx-stack
+Version: 0.1.0
+Summary: CLI control plane for local LLM infrastructure on Apple Silicon
+Project-URL: Homepage, https://github.com/weklund/mlx-stack
+Project-URL: Repository, https://github.com/weklund/mlx-stack
+Project-URL: Issues, https://github.com/weklund/mlx-stack/issues
+Project-URL: Documentation, https://github.com/weklund/mlx-stack#readme
+Author: mlx-stack contributors
+License-Expression: MIT
+License-File: LICENSE
+Keywords: apple-silicon,cli,llm,mlx,vllm
+Classifier: Development Status :: 3 - Alpha
+Classifier: Environment :: Console
+Classifier: Intended Audience :: Developers
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: MacOS
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.13
+Requires-Dist: click>=8.1
+Requires-Dist: httpx>=0.27
+Requires-Dist: huggingface-hub[cli]>=1.8.0
+Requires-Dist: psutil>=5.9
+Requires-Dist: pyyaml>=6.0
+Requires-Dist: rich>=13.0
+Description-Content-Type: text/markdown
+
+# mlx-stack
+
+**CLI control plane for local LLM inference infrastructure on Apple Silicon.**
+
+[![CI](https://github.com/weklund/mlx-stack/actions/workflows/ci.yml/badge.svg)](https://github.com/weklund/mlx-stack/actions/workflows/ci.yml)
+[![PyPI](https://img.shields.io/pypi/v/mlx-stack.svg)](https://pypi.org/project/mlx-stack/)
+[![Python 3.13+](https://img.shields.io/badge/python-≥3.13-blue.svg)](https://www.python.org/downloads/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-green.svg)](LICENSE)
+[![Platform: macOS Apple Silicon](https://img.shields.io/badge/platform-macOS%20Apple%20Silicon-lightgrey.svg)](https://support.apple.com/en-us/116943)
+
+---
+
+## Table of Contents
+
+- [Architecture](#architecture)
+- [Feature Highlights](#feature-highlights)
+- [Installation](#installation)
+- [Quick Start](#quick-start)
+- [CLI Reference](#cli-reference)
+- [Configuration](#configuration)
+- [24/7 Operation](#247-operation)
+- [Model Catalog](#model-catalog)
+- [Architecture Details](#architecture-details)
+- [Development](#development)
+- [Contributing](#contributing)
+- [License](#license)
+
+## Architecture
+
+```
+                        ┌──────────────────────────────────────────────────┐
+                        │                  mlx-stack CLI                   │
+                        │  hardware detection · recommendation · lifecycle │
+                        └──────────────┬───────────────────────────────────┘
+                                       │
+              ┌────────────────────────┼────────────────────────┐
+              │                        │                        │
+              ▼                        ▼                        ▼
+   ┌───────────────────┐  ┌───────────────────┐  ┌───────────────────┐
+   │   vllm-mlx :8000  │  │   vllm-mlx :8001  │  │   vllm-mlx :8002  │
+   │  ── standard ──   │  │    ── fast ──      │  │  ── longctx ──    │
+   │  Qwen 3.5 14B     │  │  Qwen 3.5 3B      │  │  DeepSeek R1 8B   │
+   └────────┬──────────┘  └────────┬──────────┘  └────────┬──────────┘
+            │                      │                       │
+            └──────────────────────┼───────────────────────┘
+                                   │
+                        ┌──────────▼──────────┐
+                        │  LiteLLM Proxy :4000│
+                        │  routing · fallback  │
+                        │  load balancing      │
+                        └──────────┬──────────┘
+                                   │
+                        ┌──────────▼──────────┐
+                        │  OpenAI-compatible   │
+                        │  /v1 endpoint        │
+                        │                      │
+                        │  ← Your app / agent  │
+                        └─────────────────────┘
+```
+
+mlx-stack orchestrates [vllm-mlx](https://github.com/vllm-project/vllm) model servers and a [LiteLLM](https://github.com/BerriAI/litellm) API gateway to serve large language models locally on Apple Silicon Macs. Each tier runs a dedicated model optimized for a specific workload — quality, speed, or long-context — and LiteLLM routes requests through a single OpenAI-compatible endpoint with automatic fallback.
+
+## Feature Highlights
+
+- **Hardware-Aware Recommendations** — Detects your Apple Silicon chip (M1–M5, Pro/Max/Ultra), measures memory bandwidth, and recommends an optimal model stack tuned to your exact hardware.
+- **Tiered Model Serving** — Assigns models to `standard`, `fast`, and `longctx` tiers so agents and apps can target the right balance of quality and speed per request.
+- **24/7 Unattended Operation** — Built-in watchdog with auto-restart, flap detection, exponential backoff, and macOS LaunchAgent integration for always-on inference on headless Mac Minis.
+- **One-Command Setup** — `mlx-stack init --accept-defaults` profiles your hardware, picks models, generates configs, and gets you from zero to a running OpenAI-compatible endpoint in minutes.
+- **15-Model Curated Catalog** — Ships with benchmark data for Qwen 3.5, Gemma 3, DeepSeek R1, Nemotron, and Llama 3.3 families — with quality scores, tool-calling metadata, and per-hardware performance data.
+
+## Installation
+
+The recommended way to install mlx-stack is with [uv](https://docs.astral.sh/uv/):
+
+```bash
+uv tool install mlx-stack
+```
+
+This installs `mlx-stack` globally as an isolated tool — no need to manage virtual environments.
+
+Alternatively, you can use [pipx](https://pipx.pypa.io/):
+
+```bash
+pipx install mlx-stack
+```
+
+Or try it without installing:
+
+```bash
+uvx mlx-stack profile
+```
+
+> **Note:** `uvx` runs in an ephemeral environment, which works great for one-off commands. For the watchdog and LaunchAgent features (`mlx-stack watch`, `mlx-stack install`), use `uv tool install` so the binary has a stable path.
+
+## Quick Start
+
+```bash
+# 1. Detect your hardware
+mlx-stack profile
+
+# 2. Generate stack configuration
+mlx-stack init --accept-defaults
+
+# 3. Download required models
+mlx-stack pull qwen3.5-8b
+
+# 4. Start all services
+mlx-stack up
+
+# 5. Verify
+mlx-stack status
+```
+
+The OpenAI-compatible API is now available at `http://localhost:4000/v1`.
+
+```bash
+# Stop everything when done
+mlx-stack down
+```
+
+## CLI Reference
+
+### Setup & Configuration
+
+| Command | Description |
+|---------|-------------|
+| `mlx-stack profile` | Detect Apple Silicon hardware and save profile to `~/.mlx-stack/profile.json` |
+| `mlx-stack config set <key> <value>` | Set a configuration value |
+| `mlx-stack config get <key>` | Get a configuration value |
+| `mlx-stack config list` | List all configuration values with defaults and sources |
+| `mlx-stack config reset --yes` | Reset all configuration to defaults |
+
+### Model Management
+
+**`mlx-stack recommend`** — Recommend an optimal model stack based on your hardware profile.
+
+| Option | Description |
+|--------|-------------|
+| `--budget <value>` | Memory budget override (e.g., `30gb`). Defaults to 40% of unified memory |
+| `--intent <balanced\|agent-fleet>` | Optimization strategy |
+| `--show-all` | Show all budget-fitting models ranked by score |
+
+**`mlx-stack models`** — List locally downloaded models with disk size, quantization, and active stack status.
+
+| Option | Description |
+|--------|-------------|
+| `--catalog` | Show all catalog models with hardware-specific benchmark data |
+| `--family <name>` | Filter by model family (e.g., `qwen3.5`) |
+| `--tag <name>` | Filter by tag (e.g., `agent-ready`) |
+| `--tool-calling` | Filter to tool-calling-capable models only |
+
+**`mlx-stack pull <model>`** — Download a model from the catalog.
+
+| Option | Description |
+|--------|-------------|
+| `--quant <int4\|int8\|bf16>` | Quantization level (default: `int4`) |
+| `--bench` | Run a quick benchmark after download |
+| `--force` | Re-download even if the model already exists |
+
+**`mlx-stack init`** — Generate stack definition and LiteLLM proxy configuration.
+
+| Option | Description |
+|--------|-------------|
+| `--accept-defaults` | Use defaults without prompting |
+| `--intent <balanced\|agent-fleet>` | Optimization strategy |
+| `--add <model>` | Add a model to the stack (repeatable) |
+| `--remove <tier>` | Remove a tier from the stack (repeatable) |
+| `--force` | Overwrite existing stack configuration |
+
+### Stack Lifecycle
+
+**`mlx-stack up`** — Start all services: one vllm-mlx process per tier plus the LiteLLM proxy.
+
+| Option | Description |
+|--------|-------------|
+| `--dry-run` | Show exact commands without starting anything |
+| `--tier <name>` | Start only the specified tier |
+
+**`mlx-stack down`** — Stop all managed services (SIGTERM → 10s grace → SIGKILL).
+
+| Option | Description |
+|--------|-------------|
+| `--tier <name>` | Stop only the specified tier |
+
+**`mlx-stack status`** — Show health and status of all services (healthy, degraded, down, crashed, stopped).
+
+| Option | Description |
+|--------|-------------|
+| `--json` | Output in JSON format |
+
+### Diagnostics
+
+**`mlx-stack bench <target>`** — Benchmark a running tier or catalog model. Runs 3 iterations and compares against catalog thresholds (PASS/WARN/FAIL).
+
+| Option | Description |
+|--------|-------------|
+| `--save` | Persist results for use by `recommend` and `init` scoring |
+
+### Ops & Reliability
+
+**`mlx-stack logs [service]`** — View and manage service logs. Without arguments, lists all log files.
+
+| Option | Description |
+|--------|-------------|
+| `--follow` / `-f` | Follow log output in real-time |
+| `--tail <N>` | Show last N lines (default: 50) |
+| `--service <name>` | Filter to a specific service |
+| `--rotate` | Rotate eligible log files |
+| `--all` | Show archived and current logs chronologically |
+
+**`mlx-stack watch`** — Health monitor with auto-restart, flap detection, and log rotation.
+
+| Option | Description |
+|--------|-------------|
+| `--interval <seconds>` | Polling interval (default: 30) |
+| `--max-restarts <N>` | Restarts before marking as flapping (default: 5) |
+| `--restart-delay <seconds>` | Base restart delay with exponential backoff (default: 5) |
+| `--daemon` | Run in background as a daemon |
+
+**`mlx-stack install`** — Install the watchdog as a macOS LaunchAgent.
+
+| Option | Description |
+|--------|-------------|
+| `--status` | Show current LaunchAgent status |
+
+**`mlx-stack uninstall`** — Remove the watchdog LaunchAgent. Running services are not affected.
+
+## Configuration
+
+Configuration is stored in `~/.mlx-stack/config.yaml`. Available keys:
+
+| Key | Default | Description |
+|-----|---------|-------------|
+| `openrouter-key` | *(not set)* | OpenRouter API key for cloud fallback |
+| `default-quant` | `int4` | Default quantization level (`int4`, `int8`, `bf16`) |
+| `memory-budget-pct` | `40` | Percentage of unified memory to budget for models (1–100) |
+| `litellm-port` | `4000` | LiteLLM proxy port |
+| `model-dir` | `~/.mlx-stack/models` | Model storage directory |
+| `auto-health-check` | `true` | Run health checks automatically on startup |
+| `log-max-size-mb` | `50` | Maximum log file size in MB before rotation |
+| `log-max-files` | `3` | Number of rotated log files to retain |
+
+## 24/7 Operation
+
+mlx-stack is designed to run unattended on always-on hardware like a Mac Mini.
+
+### Quick setup
+
+```bash
+mlx-stack init --accept-defaults
+mlx-stack install
+```
+
+This installs a macOS LaunchAgent that starts the watchdog on login. The watchdog:
+
+- Monitors service health every 30 seconds
+- Auto-restarts crashed processes with exponential backoff
+- Detects flapping services and stops restart loops
+- Rotates logs automatically to prevent unbounded disk usage
+
+### Manual monitoring
+
+```bash
+mlx-stack watch                  # Foreground with Rich status table
+mlx-stack watch --interval 60   # Less frequent polling
+mlx-stack watch --daemon         # Background without LaunchAgent
+```
+
+### Log management
+
+```bash
+mlx-stack logs                   # List all log files
+mlx-stack logs fast              # Last 50 lines of fast tier
+mlx-stack logs fast --follow     # Stream in real-time
+mlx-stack logs --rotate          # Rotate all eligible logs now
+```
+
+### Removing the agent
+
+```bash
+mlx-stack uninstall
+```
+
+This stops the watchdog and removes the LaunchAgent plist. Running services are not affected.
+
+## Model Catalog
+
+The built-in catalog includes 15 models across 5 families:
+
+| Family | Models | Parameters |
+|--------|--------|------------|
+| Qwen 3.5 | 6 variants | 0.8B, 3B, 8B, 14B, 32B, 72B |
+| Gemma 3 | 3 variants | 4B, 12B, 27B |
+| DeepSeek R1 | 2 variants | 8B, 32B |
+| Nemotron | 2 variants | 8B, 49B |
+| Qwen 3 / Llama 3.3 | 2 variants | 8B each |
+
+Each entry includes benchmark data for common Apple Silicon configurations, quality scores, and capability metadata (tool calling, thinking/reasoning, vision).
+
+## Architecture Details
+
+mlx-stack manages a **tiered local inference stack** with three layers:
+
+### Model Servers (vllm-mlx)
+
+One [vllm-mlx](https://github.com/vllm-project/vllm) instance per tier, each serving a single model on a dedicated port:
+
+- **standard** (port 8000) — Highest-quality model that fits your memory budget. Optimized for accuracy-sensitive tasks.
+- **fast** (port 8001) — Fastest model for latency-sensitive workloads like autocomplete and quick tool calls.
+- **longctx** (port 8002) — Architecturally diverse model (e.g., Mamba2 hybrid) for extended context windows.
+
+Each server runs with continuous batching, paged KV cache, and automatic tool-call parsing enabled.
+
+### API Gateway (LiteLLM)
+
+[LiteLLM](https://github.com/BerriAI/litellm) acts as the unified entry point on port 4000, providing:
+
+- **OpenAI-compatible `/v1` API** — Drop-in replacement for `api.openai.com` in any client or agent framework.
+- **Tier-based routing** — Requests target specific tiers by model name, or fall through a configurable chain.
+- **Automatic fallback** — If the primary tier is unavailable, requests cascade to the next healthy tier.
+
+### Cloud Fallback (Optional)
+
+With an OpenRouter API key configured, a `premium` cloud tier is available as a last-resort fallback, giving you access to frontier models when local capacity is insufficient.
+
+### Recommendation Engine
+
+The recommendation engine scores all catalog models against your hardware profile:
+
+1. **Hardware profiling** — Detects chip variant, GPU cores, unified memory, and memory bandwidth.
+2. **Memory budgeting** — Filters models to those fitting within your configured memory budget (default: 40% of unified memory).
+3. **Composite scoring** — Weights speed, quality, tool-calling capability, and memory efficiency based on your chosen intent (`balanced` or `agent-fleet`).
+4. **Tier assignment** — Assigns top-scoring models to `standard`, `fast`, and `longctx` tiers.
+5. **Local calibration** — Saved benchmark data from `mlx-stack bench --save` overrides catalog estimates for precise scoring.
+
+### Process Management
+
+- **PID tracking** — Each service writes its PID to `~/.mlx-stack/pids/` for reliable lifecycle management.
+- **Lockfile** — Prevents concurrent `up`/`down` operations via `fcntl.flock`.
+- **Health checks** — HTTP polling with exponential backoff and 120-second timeout per service.
+- **5-state model** — Services are reported as `healthy`, `degraded`, `down`, `crashed`, or `stopped`.
+- **Graceful shutdown** — SIGTERM with 10-second grace period, escalating to SIGKILL.
+
+## Development
+
+See [DEVELOPING.md](DEVELOPING.md) for the full developer guide, including project architecture, testing strategy, and how to add new models or commands.
+
+```bash
+# Install dev dependencies
+uv sync
+
+# Run tests
+uv run pytest
+
+# Type checking
+uv run python -m pyright
+
+# Linting
+uv run ruff check src/ tests/
+```
+
+## Contributing
+
+See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines on reporting bugs, suggesting features, and submitting pull requests.
+
+## License
+
+[MIT](LICENSE)

mlx_stack-0.1.0.dist-info/RECORD

@@ -0,0 +1,61 @@
+mlx_stack/__init__.py,sha256=5TQNZPOGwNrNFwmir6ab9zHkTZNPR-cDmKavhzhDXEo,153
+mlx_stack/_version.py,sha256=n_5vdJsPNu7wZ57LGuRL585uvll-hiuvZUBWzdG0RQU,520
+mlx_stack/py.typed,sha256=AbpHGcgLb-kRsJGnwFEktk7uzpZOCcBY74-YBdrKVGs,1
+mlx_stack/cli/__init__.py,sha256=ACDxQhPiBLb9vltrJiDW4ZagB_w-tNFYh5yORrMDD0E,94
+mlx_stack/cli/bench.py,sha256=xqi3r9iZFtYyA6-3sG7rLJ2l3F9Hc0mSD5K5KHeJcpY,7615
+mlx_stack/cli/config.py,sha256=DbeOYqDaEO8OCvKbKTKo_96UillHol4MD9rruinE6kQ,5148
+mlx_stack/cli/down.py,sha256=-Cbgi-zIkJz09S8yY_nmOjO7wQBi4mpE_angvWPTgfw,3322
+mlx_stack/cli/init.py,sha256=z8VTvOE5UVlkBy2Kg0veKvoFsDlZy3gSku-2yxFM_YM,5249
+mlx_stack/cli/install.py,sha256=0Lx_KqpeGPGkwjHa04cdu7IqynoPCeszbedqLQklkYA,4727
+mlx_stack/cli/logs.py,sha256=iGmZC5_Ov7uVRRK3N2vPaconS7M2sHqbIL2-2VjuS_8,6671
+mlx_stack/cli/main.py,sha256=mnM05bRRHBF2vo_liQ0PxAQBY2IfCt_-Q58nQKuDHY8,6967
+mlx_stack/cli/models.py,sha256=ADI7WKkiIcohWHtaQrj-kRkvn-i1uEPLdETKTtJBoKQ,9836
+mlx_stack/cli/profile.py,sha256=DnF47wk6-vauG1QlKslr3blCnT64RpyFGoxqE2gOjqs,1924
+mlx_stack/cli/pull.py,sha256=-FNNPTAsb8-AxTkFFpxM3CE02HWexihnD5v1jjGOrOQ,4186
+mlx_stack/cli/recommend.py,sha256=zg5lGE-ME_qcm0yRRkKxwo1Io2QI7XWWBn1gvYWNlhk,12617
+mlx_stack/cli/status.py,sha256=M1ID6RSXu1ZD5EtTEyxJXnM0BCyaHdnSwhrIZR9XbuM,3033
+mlx_stack/cli/up.py,sha256=G65J5NN-9KLie3szD2habuFqvR09AMJMjLLGhT4ekgs,4879
+mlx_stack/cli/watch.py,sha256=-6NJc-etj1H1z-4URBkWjXQtdxwBStC_tt_Kle0OQCA,6922
+mlx_stack/core/__init__.py,sha256=vHZhwkjZ88hSb4Xb1ZgDhY7B0wd76QiCyRDgmaBuqHw,49
+mlx_stack/core/benchmark.py,sha256=KGXWExc3mV3ZtPBimxLH_3tqh1-UoAsiYs3FgEkbtmI,35224
+mlx_stack/core/catalog.py,sha256=E0IL3sYa4IY0MZFe6R234P093L4foHP4Eh0Nhfb-rwE,18044
+mlx_stack/core/config.py,sha256=tM0-8maElb1wcycmU7emBeYBl0DDbiqgWY5mtP79i7Y,13364
+mlx_stack/core/deps.py,sha256=s8ZUKo7aumap1hfuDt3WOrLSvHrhTLPtwjh3c4ROYP8,9886
+mlx_stack/core/hardware.py,sha256=Fyrac4f9P3OZnHyAm-DwW-l31wnCFvjSihPIELbZ0xg,9409
+mlx_stack/core/launchd.py,sha256=pqoiipd1k8Jt_RAaSXkD7jmL1HNGrf9ycWN50xzCow4,15523
+mlx_stack/core/litellm_gen.py,sha256=IBMm-_HZcan3Yq5T4XmSUv2M8z-htE3eTfcTMd1IZ1A,5312
+mlx_stack/core/log_rotation.py,sha256=zdCj1mv_Z9s1eCviCSr7xIWI3V_j4NGu40pTGZYgJc4,7182
+mlx_stack/core/log_viewer.py,sha256=wRsBgErn5X038Z6ej5Mxc8XM9J9gCZDFuwFEX-zh89M,11453
+mlx_stack/core/models.py,sha256=V8YSYitIop19ulTr8pO5LG3Rauq_AyJ3ACo8oaWPvEU,20843
+mlx_stack/core/paths.py,sha256=xxs2I_0-SRuWbDFjdp6GmYnCB3fLhPwcB8WPfWIDvko,2056
+mlx_stack/core/process.py,sha256=DjXMgobAyN1_3Z_TswVDvV85yNIzuQOpAq01X6UdNdg,25674
+mlx_stack/core/pull.py,sha256=r8T3NOIaRcUkBvWsUws6cyltJX5UhjayImgC1E_YzS4,24948
+mlx_stack/core/scoring.py,sha256=Y95MpCiwOVqEbfApwdHUUm6y33aBWkvnVBeokmfa5r0,19724
+mlx_stack/core/stack_down.py,sha256=Zpen8jebSd0IQI-gi2TGQsmBEny4l4tC4FSZZphcekg,9379
+mlx_stack/core/stack_init.py,sha256=sFvOLPYmvWUTIzEyEB_Dzm2m5BuaH40tsmDDH8zUkck,17106
+mlx_stack/core/stack_status.py,sha256=Ue-6qzPQWWIbJtpMcGnFTk_ailrbMMkSPF2JYfPBr6o,6757
+mlx_stack/core/stack_up.py,sha256=7WOk3Y_mqprbvUsQ6d81vrDq5IN-lfHFQ78lPLv1O8w,25981
+mlx_stack/core/watchdog.py,sha256=YCKcLZiezspHWYQwsXrsoy3e4iZNuhARZa6vi4aaZ0Y,21462
+mlx_stack/data/__init__.py,sha256=Ihuwvdj0niAehLr2Yiw_439sFdwZ0p8gqUpGq1ahN-g,39
+mlx_stack/data/catalog/__init__.py,sha256=9WLBBMgdXp6hnPdD3PldIl78RR8QjXiqHmxNyr5FH9g,37
+mlx_stack/data/catalog/deepseek-r1-32b.yaml,sha256=amhzGjxAbHJRv2zrlJwifHLpF6zsd5k8rdhFWnuyOXM,927
+mlx_stack/data/catalog/deepseek-r1-8b.yaml,sha256=pa6BihfFgI1S1yk9vTDk9nKQAlhUEasQd3Ffh3XZTZs,906
+mlx_stack/data/catalog/gemma3-12b.yaml,sha256=VtbNR2_dbClldrNCzhVHNJwwS-4xc88Zt4eiKqZjn2o,844
+mlx_stack/data/catalog/gemma3-27b.yaml,sha256=iHqVRlUQSUpzZ455XyUUb28qAjY7eWmImsAEJf3YDUg,846
+mlx_stack/data/catalog/gemma3-4b.yaml,sha256=ut3oyDhDKCOYYBwBHK952EIbtqB_ah3b9JOB_h-kQmo,846
+mlx_stack/data/catalog/llama3.3-8b.yaml,sha256=gmTSLn9ba52HXr0r8P6TS7pQCIQ1yIQXkSuT_0xQYZQ,870
+mlx_stack/data/catalog/nemotron-49b.yaml,sha256=nWoq7rE2MJi5hnsdgX4Yvn09uXIUZRsc3_I7-CNhqEI,799
+mlx_stack/data/catalog/nemotron-8b.yaml,sha256=68ZabSgwmIdoP7xqn70KNbLFLsN5A40g02NjueNoW9I,843
+mlx_stack/data/catalog/qwen3-8b.yaml,sha256=QPzslOUqENtEHxy1jMGioPzYMLbpjm28zFjhAVq8w_k,823
+mlx_stack/data/catalog/qwen3.5-0.8b.yaml,sha256=CQZsiqAsgjXgY7J6Vc9jaggiJZ7FBxubaef4STJWVS4,858
+mlx_stack/data/catalog/qwen3.5-14b.yaml,sha256=aNdoUAeqDnvc8Xbos2cZN95S3VaMuFUTWr_2lGPz7S4,852
+mlx_stack/data/catalog/qwen3.5-32b.yaml,sha256=PLnHe9gs7m1qAkGxgcvheIQaaXlUX1fwpD7xMcbA2Ak,843
+mlx_stack/data/catalog/qwen3.5-3b.yaml,sha256=AD0sfx0hiDclPABmd7L-pIvcy7DfCyr75m7nN9ApzO4,830
+mlx_stack/data/catalog/qwen3.5-72b.yaml,sha256=IUBvwZ-ORbOIfAzATuDz5Muk2zlF29zPlnaoAQBrUwc,784
+mlx_stack/data/catalog/qwen3.5-8b.yaml,sha256=_ws4BYLKkIb-PLP3Zk4ACORMPLvJDkt7rJUoxYdPcZY,835
+mlx_stack/utils/__init__.py,sha256=A83-xB78mFfkul0y3PLHIHyF61CcTbx4IEzepKabCzc,37
+mlx_stack-0.1.0.dist-info/METADATA,sha256=YvEdPzZVlDSx8eH1tYBZ8dVVoTwFpTT-uOQw_togLNw,16965
+mlx_stack-0.1.0.dist-info/WHEEL,sha256=QccIxa26bgl1E6uMy58deGWi-0aeIkkangHcxk2kWfw,87
+mlx_stack-0.1.0.dist-info/entry_points.txt,sha256=zZC93ftIoIZGWwMQMNIaL8adv6wvT6HjEAZefDj-iW0,48
+mlx_stack-0.1.0.dist-info/licenses/LICENSE,sha256=KqPG8dgvrddX0b0jUgE_GAlo2QVivpj2h_f-wlKdroI,1079
+mlx_stack-0.1.0.dist-info/RECORD,,

mlx_stack-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.29.0
+Root-Is-Purelib: true
+Tag: py3-none-any

mlx_stack-0.1.0.dist-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+mlx-stack = mlx_stack.cli:cli

mlx_stack-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 mlx-stack contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.