bit-ttt-engine 0.6.2__tar.gz → 0.7.0__tar.gz

bit_ttt_engine-0.7.0/PKG-INFO

@@ -0,0 +1,136 @@
+Metadata-Version: 2.4
+Name: bit-ttt-engine
+Version: 0.7.0
+Classifier: Development Status :: 4 - Beta
+Classifier: Programming Language :: Rust
+Classifier: Programming Language :: Python :: Implementation :: CPython
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.8
+Classifier: Programming Language :: Python :: 3.9
+Classifier: Programming Language :: Python :: 3.10
+Classifier: Programming Language :: Python :: 3.11
+Classifier: Programming Language :: Python :: 3.12
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Operating System :: Microsoft :: Windows
+Requires-Dist: tokenizers>=0.19
+Requires-Dist: huggingface-hub>=0.20
+Requires-Dist: bit-ttt-engine[server] ; extra == 'all'
+Requires-Dist: fastapi>=0.100 ; extra == 'server'
+Requires-Dist: uvicorn>=0.20 ; extra == 'server'
+Requires-Dist: sse-starlette>=1.6 ; extra == 'server'
+Provides-Extra: all
+Provides-Extra: server
+License-File: LICENSE
+Summary: Fast local LLM inference with TTT (Test-Time Training) and LoRA — the model that learns while it runs
+Keywords: llm,inference,ttt,lora,gguf,quantization,cuda
+Home-Page: https://github.com/imonoonoko/Bit-TTT-Engine
+Author: imonoonoko
+License: MIT
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown; charset=UTF-8; variant=GFM
+
+# 🧠 Bit-TTT-Engine
+
+[![PyPI](https://img.shields.io/pypi/v/bit-ttt-engine.svg)](https://pypi.org/project/bit-ttt-engine/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
+[![Rust](https://img.shields.io/badge/rust-1.70+-orange.svg)](https://www.rust-lang.org/)
+
+**Fast local LLM inference that learns while it runs.**
+
+- 🏎️ **47+ tok/s** on RTX 4060 Ti (7B Q4_K_M)
+- 🧠 **TTT** (Test-Time Training) — adapts during inference (world's first!)
+- 🎨 **LoRA** — fine-tune with one flag
+- 📦 **5 models** — Llama-2/3, Gemma-2, Qwen2.5, Mistral
+- 🔌 **OpenAI-compatible API** — drop-in replacement
+
+## 🚀 Quick Start
+
+```bash
+pip install bit-ttt-engine
+```
+
+```python
+import cortex_rust
+
+# Load any GGUF model (auto-downloads from HuggingFace!)
+model = cortex_rust.load("user/model-GGUF")
+
+# Chat
+response = model.chat([
+    {"role": "user", "content": "Hello!"}
+])
+print(response)
+
+# Stream
+for token in model.chat_stream([
+    {"role": "user", "content": "Tell me a story"}
+]):
+    print(token, end="", flush=True)
+```
+
+## 🖥️ CLI
+
+```bash
+# Interactive chat
+bit-ttt chat model.gguf
+
+# Generate text
+bit-ttt generate model.gguf -p "Once upon a time"
+
+# OpenAI-compatible API server
+bit-ttt serve model.gguf --port 8000
+
+# With LoRA + Q8 KV cache
+bit-ttt chat model.gguf --lora adapter.bin --q8-cache
+```
+
+## 🧠 TTT — Test-Time Training
+
+**The model learns while it generates.** No other local LLM does this.
+
+```python
+model = cortex_rust.load("model.gguf")
+model.enable_ttt(True)
+
+# Each conversation makes the model smarter
+response = model.chat([{"role": "user", "content": "My name is Alice"}])
+# Next time, it remembers context better!
+```
+
+## ⚡ Performance
+
+| Model | Speed | VRAM |
+|-------|-------|------|
+| Llama-2 7B (Q4_K_M) | 47.8 tok/s | ~5 GB |
+| Llama-3 8B (Q4_K_M) | 36.8 tok/s | ~6 GB |
+| Mistral 7B (Q4_K_M) | 40.8 tok/s | ~5 GB |
+| Qwen2.5 1.5B (Q4_K_M) | 70.4 tok/s | ~2 GB |
+
+With `--q8-cache`: **82% VRAM reduction** for KV cache.
+
+## 🔌 OpenAI-Compatible API
+
+```bash
+bit-ttt serve model.gguf --port 8000
+```
+
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")
+response = client.chat.completions.create(
+    model="default",
+    messages=[{"role": "user", "content": "Hi!"}],
+    stream=True,
+)
+```
+
+## 📖 Links
+
+- [GitHub](https://github.com/imonoonoko/Bit-TTT-Engine)
+- [Documentation](https://github.com/imonoonoko/Bit-TTT-Engine#readme)
+
+## 💖 License
+
+MIT License
+

bit_ttt_engine-0.7.0/README_PYPI.md

@@ -0,0 +1,103 @@
+# 🧠 Bit-TTT-Engine
+
+[![PyPI](https://img.shields.io/pypi/v/bit-ttt-engine.svg)](https://pypi.org/project/bit-ttt-engine/)
+[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE)
+[![Rust](https://img.shields.io/badge/rust-1.70+-orange.svg)](https://www.rust-lang.org/)
+
+**Fast local LLM inference that learns while it runs.**
+
+- 🏎️ **47+ tok/s** on RTX 4060 Ti (7B Q4_K_M)
+- 🧠 **TTT** (Test-Time Training) — adapts during inference (world's first!)
+- 🎨 **LoRA** — fine-tune with one flag
+- 📦 **5 models** — Llama-2/3, Gemma-2, Qwen2.5, Mistral
+- 🔌 **OpenAI-compatible API** — drop-in replacement
+
+## 🚀 Quick Start
+
+```bash
+pip install bit-ttt-engine
+```
+
+```python
+import cortex_rust
+
+# Load any GGUF model (auto-downloads from HuggingFace!)
+model = cortex_rust.load("user/model-GGUF")
+
+# Chat
+response = model.chat([
+    {"role": "user", "content": "Hello!"}
+])
+print(response)
+
+# Stream
+for token in model.chat_stream([
+    {"role": "user", "content": "Tell me a story"}
+]):
+    print(token, end="", flush=True)
+```
+
+## 🖥️ CLI
+
+```bash
+# Interactive chat
+bit-ttt chat model.gguf
+
+# Generate text
+bit-ttt generate model.gguf -p "Once upon a time"
+
+# OpenAI-compatible API server
+bit-ttt serve model.gguf --port 8000
+
+# With LoRA + Q8 KV cache
+bit-ttt chat model.gguf --lora adapter.bin --q8-cache
+```
+
+## 🧠 TTT — Test-Time Training
+
+**The model learns while it generates.** No other local LLM does this.
+
+```python
+model = cortex_rust.load("model.gguf")
+model.enable_ttt(True)
+
+# Each conversation makes the model smarter
+response = model.chat([{"role": "user", "content": "My name is Alice"}])
+# Next time, it remembers context better!
+```
+
+## ⚡ Performance
+
+| Model | Speed | VRAM |
+|-------|-------|------|
+| Llama-2 7B (Q4_K_M) | 47.8 tok/s | ~5 GB |
+| Llama-3 8B (Q4_K_M) | 36.8 tok/s | ~6 GB |
+| Mistral 7B (Q4_K_M) | 40.8 tok/s | ~5 GB |
+| Qwen2.5 1.5B (Q4_K_M) | 70.4 tok/s | ~2 GB |
+
+With `--q8-cache`: **82% VRAM reduction** for KV cache.
+
+## 🔌 OpenAI-Compatible API
+
+```bash
+bit-ttt serve model.gguf --port 8000
+```
+
+```python
+from openai import OpenAI
+client = OpenAI(base_url="http://localhost:8000/v1", api_key="none")
+response = client.chat.completions.create(
+    model="default",
+    messages=[{"role": "user", "content": "Hi!"}],
+    stream=True,
+)
+```
+
+## 📖 Links
+
+- [GitHub](https://github.com/imonoonoko/Bit-TTT-Engine)
+- [Documentation](https://github.com/imonoonoko/Bit-TTT-Engine#readme)
+
+## 💖 License
+
+MIT License

bit_ttt_engine-0.7.0/cortex_rust/__init__.py

@@ -0,0 +1,25 @@
+# Auto-add CUDA DLL path on Windows
+import os
+import sys
+
+if sys.platform == "win32":
+    cuda_paths = [
+        os.environ.get("CUDA_PATH", ""),
+        r"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.4",
+        r"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.3",
+        r"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v12.0",
+        r"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8",
+    ]
+    for cuda_path in cuda_paths:
+        bin_path = os.path.join(cuda_path, "bin")
+        if os.path.isdir(bin_path):
+            os.add_dll_directory(bin_path)
+            break
+
+from .cortex_rust import *
+from .chat import format_chat, format_simple, detect_template, list_templates
+from .engine import load, Model
+
+__doc__ = cortex_rust.__doc__
+if hasattr(cortex_rust, "__all__"):
+    __all__ = cortex_rust.__all__

bit_ttt_engine-0.7.0/cortex_rust/__main__.py

@@ -0,0 +1,4 @@
+"""Enable `python -m cortex_rust` as CLI entry point."""
+from cortex_rust.cli import main
+
+main()

bit_ttt_engine-0.7.0/cortex_rust/chat.py

@@ -0,0 +1,196 @@
+"""Chat template support for various LLM architectures.
+
+Provides format_chat() to convert messages into model-specific prompt strings.
+Supports Llama-3, Llama-2, Gemma-2, Qwen/ChatML, and generic formats.
+
+Usage:
+    from cortex_rust.chat import format_chat, detect_template
+
+    messages = [
+        {"role": "system", "content": "You are a helpful assistant."},
+        {"role": "user", "content": "Hello!"},
+    ]
+    prompt = format_chat(messages, template="llama3")
+    # Or auto-detect from model path:
+    template = detect_template("Meta-Llama-3-8B-Instruct.Q4_K_M.gguf")
+    prompt = format_chat(messages, template=template)
+"""
+
+from typing import List, Dict, Optional
+
+# ============================================================================
+# Template Definitions
+# ============================================================================
+
+TEMPLATES = {
+    "llama3": {
+        "bos": "<|begin_of_text|>",
+        "system_start": "<|start_header_id|>system<|end_header_id|>\n\n",
+        "system_end": "<|eot_id|>",
+        "user_start": "<|start_header_id|>user<|end_header_id|>\n\n",
+        "user_end": "<|eot_id|>",
+        "assistant_start": "<|start_header_id|>assistant<|end_header_id|>\n\n",
+        "assistant_end": "<|eot_id|>",
+        "default_system": "You are a helpful assistant.",
+    },
+    "llama2": {
+        "bos": "<s>",
+        "system_start": "<<SYS>>\n",
+        "system_end": "\n<</SYS>>\n\n",
+        "user_start": "[INST] ",
+        "user_end": " [/INST]",
+        "assistant_start": " ",
+        "assistant_end": " </s>",
+        "default_system": "You are a helpful, respectful and honest assistant.",
+        # Llama-2 embeds system inside first [INST]
+        "system_inside_user": True,
+    },
+    "gemma2": {
+        "bos": "<bos>",
+        "system_start": "",  # Gemma-2 has no system role
+        "system_end": "",
+        "user_start": "<start_of_turn>user\n",
+        "user_end": "<end_of_turn>\n",
+        "assistant_start": "<start_of_turn>model\n",
+        "assistant_end": "<end_of_turn>\n",
+        "default_system": None,  # No system support
+    },
+    "chatml": {
+        # Used by Qwen, Mistral-Instruct, etc.
+        "bos": "",
+        "system_start": "<|im_start|>system\n",
+        "system_end": "<|im_end|>\n",
+        "user_start": "<|im_start|>user\n",
+        "user_end": "<|im_end|>\n",
+        "assistant_start": "<|im_start|>assistant\n",
+        "assistant_end": "<|im_end|>\n",
+        "default_system": "You are a helpful assistant.",
+    },
+}
+
+# Model name patterns → template mapping
+_DETECTION_PATTERNS = [
+    ("llama-3", "llama3"),
+    ("llama3", "llama3"),
+    ("meta-llama-3", "llama3"),
+    ("llama-2", "llama2"),
+    ("llama2", "llama2"),
+    ("gemma-2", "gemma2"),
+    ("gemma2", "gemma2"),
+    ("qwen", "chatml"),
+    ("mistral", "chatml"),
+    ("yi-", "chatml"),
+]
+
+
+def detect_template(model_path: str) -> str:
+    """Auto-detect chat template from model filename.
+    
+    Args:
+        model_path: Path to model file (e.g., "Meta-Llama-3-8B-Instruct.Q4_K_M.gguf")
+    
+    Returns:
+        Template name (e.g., "llama3", "chatml"). Falls back to "chatml" if unknown.
+    """
+    name = model_path.lower().replace("\\", "/").split("/")[-1]
+    for pattern, template in _DETECTION_PATTERNS:
+        if pattern in name:
+            return template
+    return "chatml"  # Safe default
+
+
+def list_templates() -> List[str]:
+    """List all available template names."""
+    return list(TEMPLATES.keys())
+
+
+def format_chat(
+    messages: List[Dict[str, str]],
+    template: str = "chatml",
+    add_generation_prompt: bool = True,
+) -> str:
+    """Format chat messages into a model-specific prompt string.
+    
+    Args:
+        messages: List of {"role": "system"|"user"|"assistant", "content": "..."}
+        template: Template name ("llama3", "llama2", "gemma2", "chatml")
+        add_generation_prompt: If True, append assistant start token at the end
+    
+    Returns:
+        Formatted prompt string ready for model.generate()
+    
+    Example:
+        >>> messages = [{"role": "user", "content": "Hello!"}]
+        >>> format_chat(messages, template="llama3")
+        '<|begin_of_text|><|start_header_id|>user<|end_header_id|>\\n\\nHello!<|eot_id|><|start_header_id|>assistant<|end_header_id|>\\n\\n'
+    """
+    if template not in TEMPLATES:
+        raise ValueError(f"Unknown template '{template}'. Available: {list_templates()}")
+    
+    tmpl = TEMPLATES[template]
+    parts = [tmpl["bos"]]
+    
+    system_inside_user = tmpl.get("system_inside_user", False)
+    system_content = None
+    
+    for msg in messages:
+        role = msg["role"]
+        content = msg["content"]
+        
+        if role == "system":
+            if system_inside_user:
+                # Llama-2 style: save system for embedding in first user message
+                system_content = content
+            elif tmpl["system_start"]:  # Skip if no system support (Gemma-2)
+                parts.append(tmpl["system_start"])
+                parts.append(content)
+                parts.append(tmpl["system_end"])
+        
+        elif role == "user":
+            parts.append(tmpl["user_start"])
+            if system_inside_user and system_content is not None:
+                # Llama-2: embed system before user content
+                parts.append(tmpl["system_start"])
+                parts.append(system_content)
+                parts.append(tmpl["system_end"])
+                system_content = None  # Only first user message
+            parts.append(content)
+            parts.append(tmpl["user_end"])
+        
+        elif role == "assistant":
+            parts.append(tmpl["assistant_start"])
+            parts.append(content)
+            parts.append(tmpl["assistant_end"])
+    
+    if add_generation_prompt:
+        parts.append(tmpl["assistant_start"])
+    
+    return "".join(parts)
+
+
+def format_simple(
+    user_message: str,
+    system_message: Optional[str] = None,
+    template: str = "chatml",
+) -> str:
+    """Convenience: format a single user message (with optional system prompt).
+    
+    Args:
+        user_message: The user's message
+        system_message: Optional system prompt (uses template default if None)
+        template: Template name
+    
+    Returns:
+        Formatted prompt string
+    """
+    messages = []
+    
+    tmpl = TEMPLATES.get(template, TEMPLATES["chatml"])
+    if system_message is not None:
+        messages.append({"role": "system", "content": system_message})
+    elif tmpl.get("default_system"):
+        messages.append({"role": "system", "content": tmpl["default_system"]})
+    
+    messages.append({"role": "user", "content": user_message})
+    
+    return format_chat(messages, template=template)