simajilord 0.1.0__tar.gz

simajilord-0.1.0/PKG-INFO

@@ -0,0 +1,138 @@
+Metadata-Version: 2.4
+Name: simajilord
+Version: 0.1.0
+Summary: Local OpenAI API runtime and tool-search foundation for multimodal Cappuccino-class models on Apple Silicon.
+Requires-Python: >=3.12
+Description-Content-Type: text/markdown
+Requires-Dist: fastapi>=0.135.1
+Requires-Dist: httpx>=0.28.1
+Requires-Dist: mlx>=0.31.1
+Requires-Dist: mlx-lm>=0.31.1
+Requires-Dist: mlx-vlm>=0.4.0
+Requires-Dist: pdfplumber>=0.11.9
+Requires-Dist: pydantic>=2.12.5
+Requires-Dist: pypdf>=6.9.2
+Requires-Dist: uvicorn>=0.42.0
+Provides-Extra: dev
+Requires-Dist: mypy>=1.18.2; extra == "dev"
+Requires-Dist: pytest>=9.0.2; extra == "dev"
+
+# Cappuccino
+
+`simajilord` is the package and CLI that exposes a local OpenAI-compatible runtime for Cappuccino-class models on Apple Silicon.
+
+## Install
+
+The model name stays `cappuccino`, but the package name is:
+
+```bash
+pip install simajilord openai
+```
+
+Public import and CLI:
+
+```bash
+python -c "import simajilord; print(simajilord.__file__)"
+simajilord --help
+```
+
+## Serve a local OpenAI-compatible endpoint
+
+```bash
+simajilord serve \
+  --default-model-path /path/to/Cappuccino-27B \
+  --public-model-id cappuccino \
+  --host 127.0.0.1 \
+  --port 8020
+```
+
+## Quickstart with the OpenAI Python SDK
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.chat.completions.create(
+    model="cappuccino",
+    messages=[
+        {"role": "user", "content": "日本語で一文だけ自己紹介して。"},
+    ],
+)
+
+print(response.choices[0].message.content)
+```
+
+## Quickstart with skill selection
+
+Skill selection is a Cappuccino extension. The request stays OpenAI-compatible because skill metadata is converted into ordinary upstream instructions before the model call.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.responses.create(
+    model="cappuccino",
+    input="OpenAI API の公式情報だけで確認して要点をまとめて。",
+    extra_body={
+        "skill_choice": "auto",
+    },
+)
+
+print(response.output_text)
+```
+
+Explicit skill selection:
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.responses.create(
+    model="cappuccino",
+    input="最新の API 仕様を確認して。",
+    extra_body={
+        "skills": ["openai-docs"],
+    },
+)
+
+print(response.output_text)
+```
+
+## Skill registry layout
+
+Each skill lives in its own directory and uses `SKILL.md` frontmatter:
+
+```md
+---
+name: openai-docs
+description: Use when a task needs current OpenAI API documentation and official references.
+---
+
+Read the official docs first.
+```
+
+Load the registry at serve time:
+
+```bash
+simajilord serve \
+  --default-model-path /path/to/Cappuccino-27B \
+  --skill-registry /path/to/skills
+```
+
+Compatibility CLI alias:
+
+```bash
+cappuccino --help
+```

simajilord-0.1.0/PYPI.md

@@ -0,0 +1,119 @@
+# Cappuccino
+
+`simajilord` is the package and CLI that exposes a local OpenAI-compatible runtime for Cappuccino-class models on Apple Silicon.
+
+## Install
+
+The model name stays `cappuccino`, but the package name is:
+
+```bash
+pip install simajilord openai
+```
+
+Public import and CLI:
+
+```bash
+python -c "import simajilord; print(simajilord.__file__)"
+simajilord --help
+```
+
+## Serve a local OpenAI-compatible endpoint
+
+```bash
+simajilord serve \
+  --default-model-path /path/to/Cappuccino-27B \
+  --public-model-id cappuccino \
+  --host 127.0.0.1 \
+  --port 8020
+```
+
+## Quickstart with the OpenAI Python SDK
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.chat.completions.create(
+    model="cappuccino",
+    messages=[
+        {"role": "user", "content": "日本語で一文だけ自己紹介して。"},
+    ],
+)
+
+print(response.choices[0].message.content)
+```
+
+## Quickstart with skill selection
+
+Skill selection is a Cappuccino extension. The request stays OpenAI-compatible because skill metadata is converted into ordinary upstream instructions before the model call.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.responses.create(
+    model="cappuccino",
+    input="OpenAI API の公式情報だけで確認して要点をまとめて。",
+    extra_body={
+        "skill_choice": "auto",
+    },
+)
+
+print(response.output_text)
+```
+
+Explicit skill selection:
+
+```python
+from openai import OpenAI
+
+client = OpenAI(
+    base_url="http://127.0.0.1:8020/v1",
+    api_key="dummy",
+)
+
+response = client.responses.create(
+    model="cappuccino",
+    input="最新の API 仕様を確認して。",
+    extra_body={
+        "skills": ["openai-docs"],
+    },
+)
+
+print(response.output_text)
+```
+
+## Skill registry layout
+
+Each skill lives in its own directory and uses `SKILL.md` frontmatter:
+
+```md
+---
+name: openai-docs
+description: Use when a task needs current OpenAI API documentation and official references.
+---
+
+Read the official docs first.
+```
+
+Load the registry at serve time:
+
+```bash
+simajilord serve \
+  --default-model-path /path/to/Cappuccino-27B \
+  --skill-registry /path/to/skills
+```
+
+Compatibility CLI alias:
+
+```bash
+cappuccino --help
+```

simajilord-0.1.0/README.md

@@ -0,0 +1,56 @@
+---
+license: apache-2.0
+base_model:
+- cappuccino-27b-base
+pipeline_tag: text-generation
+tags:
+- cappuccino
+- multimodal
+- tool-use
+- thinking
+- local-runtime
+library_name: transformers
+---
+
+# Cappuccino-27B
+
+Cappuccino-27B is the local Cappuccino foundation checkpoint used by this repository for runtime serving, continual pretraining, and supervised fine-tuning.
+
+## Identity
+
+- Public model name: `Cappuccino-27B`
+- Public base-model identity: `cappuccino-27b-base`
+- Repository root stores the original-precision MLX weights.
+- `artifacts/q6` is a separate 6-bit inference copy for serving and benchmarking.
+- Model and processor identifiers in the model directory are branded as `Cappuccino`.
+
+## Architecture
+
+- Type: causal language model with a vision encoder
+- Scale: 27B-class multimodal model
+- Decoder layers: 64
+- Native context window: `262144`
+- Context length: 262,144 natively and extensible up to 1,010,000 tokens.
+- Training and runtime surface: text, image, video placeholders, tool calls, tool responses, and thinking traces
+
+## Runtime
+
+Cappuccino exposes the OpenAI API through the local wrapper, with the Responses API as the default interface for tools and multimodal turns, and can be inspected, trained, quantized, and served with the `cappuccino` CLI.
+
+```bash
+./.venv/bin/cappuccino inspect-model --model-path /Users/megantemeteo/Desktop/cappuccino
+```
+
+```bash
+./.venv/bin/cappuccino train \
+  --model-path /Users/megantemeteo/Desktop/cappuccino \
+  --dataset-path /path/to/train.jsonl \
+  --output-dir /path/to/out \
+  --dry-run
+```
+
+## Storage
+
+- `inspect-model` reports `weight_bytes` for the base checkpoint already stored in the model directory.
+- `train --dry-run` reports `saved_trainable_tensor_bytes` separately so the saved trainable tensors can be accounted for without conflating them with the base checkpoint.
+- If adapters are merged back into a full checkpoint later, the merged checkpoint size stays close to the base model size because the architecture does not grow.

simajilord-0.1.0/pyproject.toml

@@ -0,0 +1,64 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "simajilord"
+version = "0.1.0"
+description = "Local OpenAI API runtime and tool-search foundation for multimodal Cappuccino-class models on Apple Silicon."
+readme = "PYPI.md"
+requires-python = ">=3.12"
+dependencies = [
+  "fastapi>=0.135.1",
+  "httpx>=0.28.1",
+  "mlx>=0.31.1",
+  "mlx-lm>=0.31.1",
+  "mlx-vlm>=0.4.0",
+  "pdfplumber>=0.11.9",
+  "pydantic>=2.12.5",
+  "pypdf>=6.9.2",
+  "uvicorn>=0.42.0",
+]
+
+[project.optional-dependencies]
+dev = [
+  "mypy>=1.18.2",
+  "pytest>=9.0.2",
+]
+
+[project.scripts]
+simajilord = "simajilord.cli:main"
+cappuccino = "cappuccino.cli:main"
+
+[tool.pytest.ini_options]
+pythonpath = ["src"]
+testpaths = ["tests"]
+
+[tool.mypy]
+python_version = "3.12"
+files = ["src", "tests"]
+check_untyped_defs = true
+disallow_incomplete_defs = true
+disallow_untyped_defs = true
+no_implicit_optional = true
+warn_redundant_casts = true
+warn_return_any = true
+warn_unreachable = true
+warn_unused_ignores = true
+
+[[tool.mypy.overrides]]
+module = [
+  "mlx.*",
+  "mlx_lm.*",
+  "mlx_vlm.*",
+  "pdfplumber.*",
+  "pypdf.*",
+  "uvicorn.*",
+]
+ignore_missing_imports = true
+
+[tool.setuptools]
+package-dir = {"" = "src"}
+
+[tool.setuptools.packages.find]
+where = ["src"]

simajilord-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

simajilord-0.1.0/src/cappuccino/__init__.py

@@ -0,0 +1,3 @@
+__all__ = ["__version__"]
+
+__version__ = "0.1.0"

simajilord-0.1.0/src/cappuccino/adaptive_reasoning.py

@@ -0,0 +1,15 @@
+from __future__ import annotations
+
+from .auto_reasoning import (
+    run_adaptive_chat_completion,
+    run_auto_chat_completion,
+    run_on_chat_completion,
+    run_structured_reasoning_chat_completion,
+)
+
+__all__ = [
+    "run_adaptive_chat_completion",
+    "run_auto_chat_completion",
+    "run_on_chat_completion",
+    "run_structured_reasoning_chat_completion",
+]