TACHY-Compiler 0.1.1__tar.gz

tachy_compiler-0.1.1/PKG-INFO

@@ -0,0 +1,93 @@
+Metadata-Version: 2.4
+Name: TACHY-Compiler
+Version: 0.1.1
+Summary: TACHY model compiler and platform conversion tools.
+Author: Deeper-I
+License: BSD-3-Clause
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: numpy
+Requires-Dist: networkx
+Requires-Dist: onnx
+Requires-Dist: commentjson
+Requires-Dist: ordered-set
+Requires-Dist: easydict
+Requires-Dist: DDesignerAPI
+Provides-Extra: yolov9
+Requires-Dist: torch; extra == "yolov9"
+
+# TACHY-Compiler
+
+TACHY-Compiler is a comprehensive deep learning compiler and platform conversion framework developed by Deeper-I. It is designed to bridge the gap between popular deep learning frameworks (like PyTorch) and the TACHY hardware accelerator runtime environment.
+
+This compiler handles model parsing, optimization, tensor alignment, and binary generation. It robustly supports large-scale models, complex operators, and dual-checkpoint pipelines out of the box.
+
+## Key Features
+
+- **PyTorch to ONNX to TACHY**: Seamless end-to-end compilation pipeline.
+- **Dual Checkpoint Compilation (`compile-dual`)**: Officially supports simultaneous parsing and compilation of Reference (`eval` / `pre`) and Target (`comp` / `opt`) checkpoints into a unified output structure.
+- **Large Model Support**: Built-in mechanisms to handle models larger than the Protobuf 2GB limit using `external_data` mapping.
+- **Robust ONNX Versioning**: Automatically utilizes ONNX Opset 18 for stability, equipped with automatic version conversion and fallback mechanisms to mitigate `onnxscript` conversion issues.
+- **MCP Server Integrated**: Exposes tools to LLMs and agentic frameworks via the Model Context Protocol (MCP) using `FastMCP`.
+- **Dynamic Binary Resolution**: Intelligently locates dependency binaries utilizing environment variables (`TACHY_UTIL_PATH`) and relative package paths.
+
+## Installation
+
+You can install TACHY-Compiler directly via PyPI:
+
+```bash
+pip install TACHY-Compiler
+```
+
+### Requirements
+- Python >= 3.8
+- `numpy`, `networkx`, `onnx`, `commentjson`, `ordered-set`, `easydict`, `DDesignerAPI`
+- (Optional) `torch` if utilizing the PyTorch to ONNX conversion pipeline.
+
+## CLI Usage
+
+TACHY-Compiler provides a simple, unified command-line interface under the `tachy` command.
+
+### 1. Dual Checkpoint Compilation
+Best for executing the full compilation pipeline for two checkpoints (e.g. `eval` and `comp`) simultaneously. Intermediate files are automatically managed and cleaned up if requested.
+
+```bash
+tachy compile-dual \
+    --eval_model ./eval_model.pth \
+    --comp_model ./comp_model.pth \
+    --input_shape "[1,3,224,224]" \
+    --output_dir ./dual_output \
+    --cleanup
+```
+Outputs will be cleanly structured into `./dual_output/final_binaries/eval/` and `./dual_output/final_binaries/comp/`.
+
+### 2. Single Model Pipeline Steps
+If you need granular control, you can execute the pipeline steps individually.
+
+**Convert PyTorch to ONNX**
+```bash
+tachy convert-pt2onnx ./model.pth ./model.onnx "[1,3,224,224]" False
+```
+
+**Convert ONNX to TACHY Format**
+```bash
+tachy convert-onnx ./model.onnx ./model.json
+```
+
+**Compile TACHY JSON to Runtime Binaries**
+```bash
+tachy compile ./model.json ./output_bin_dir
+```
+
+### 3. Model Context Protocol (MCP) Server
+You can spin up the built-in MCP server to allow LLMs and AI Agents to compile models on your behalf.
+```bash
+tachy mcp
+```
+
+## Environment Variables
+- `TACHY_UTIL_PATH` (Optional): An absolute path to the directory containing external utility binaries (`fc.out`, `block_4bit.out`). If not set, the compiler will intelligently look for the `optional` directory within the installed package.
+
+## License
+
+This project is licensed under the BSD-3-Clause License. See the LICENSE file for details.

tachy_compiler-0.1.1/README.md

@@ -0,0 +1,75 @@
+# TACHY-Compiler
+
+TACHY-Compiler is a comprehensive deep learning compiler and platform conversion framework developed by Deeper-I. It is designed to bridge the gap between popular deep learning frameworks (like PyTorch) and the TACHY hardware accelerator runtime environment.
+
+This compiler handles model parsing, optimization, tensor alignment, and binary generation. It robustly supports large-scale models, complex operators, and dual-checkpoint pipelines out of the box.
+
+## Key Features
+
+- **PyTorch to ONNX to TACHY**: Seamless end-to-end compilation pipeline.
+- **Dual Checkpoint Compilation (`compile-dual`)**: Officially supports simultaneous parsing and compilation of Reference (`eval` / `pre`) and Target (`comp` / `opt`) checkpoints into a unified output structure.
+- **Large Model Support**: Built-in mechanisms to handle models larger than the Protobuf 2GB limit using `external_data` mapping.
+- **Robust ONNX Versioning**: Automatically utilizes ONNX Opset 18 for stability, equipped with automatic version conversion and fallback mechanisms to mitigate `onnxscript` conversion issues.
+- **MCP Server Integrated**: Exposes tools to LLMs and agentic frameworks via the Model Context Protocol (MCP) using `FastMCP`.
+- **Dynamic Binary Resolution**: Intelligently locates dependency binaries utilizing environment variables (`TACHY_UTIL_PATH`) and relative package paths.
+
+## Installation
+
+You can install TACHY-Compiler directly via PyPI:
+
+```bash
+pip install TACHY-Compiler
+```
+
+### Requirements
+- Python >= 3.8
+- `numpy`, `networkx`, `onnx`, `commentjson`, `ordered-set`, `easydict`, `DDesignerAPI`
+- (Optional) `torch` if utilizing the PyTorch to ONNX conversion pipeline.
+
+## CLI Usage
+
+TACHY-Compiler provides a simple, unified command-line interface under the `tachy` command.
+
+### 1. Dual Checkpoint Compilation
+Best for executing the full compilation pipeline for two checkpoints (e.g. `eval` and `comp`) simultaneously. Intermediate files are automatically managed and cleaned up if requested.
+
+```bash
+tachy compile-dual \
+    --eval_model ./eval_model.pth \
+    --comp_model ./comp_model.pth \
+    --input_shape "[1,3,224,224]" \
+    --output_dir ./dual_output \
+    --cleanup
+```
+Outputs will be cleanly structured into `./dual_output/final_binaries/eval/` and `./dual_output/final_binaries/comp/`.
+
+### 2. Single Model Pipeline Steps
+If you need granular control, you can execute the pipeline steps individually.
+
+**Convert PyTorch to ONNX**
+```bash
+tachy convert-pt2onnx ./model.pth ./model.onnx "[1,3,224,224]" False
+```
+
+**Convert ONNX to TACHY Format**
+```bash
+tachy convert-onnx ./model.onnx ./model.json
+```
+
+**Compile TACHY JSON to Runtime Binaries**
+```bash
+tachy compile ./model.json ./output_bin_dir
+```
+
+### 3. Model Context Protocol (MCP) Server
+You can spin up the built-in MCP server to allow LLMs and AI Agents to compile models on your behalf.
+```bash
+tachy mcp
+```
+
+## Environment Variables
+- `TACHY_UTIL_PATH` (Optional): An absolute path to the directory containing external utility binaries (`fc.out`, `block_4bit.out`). If not set, the compiler will intelligently look for the `optional` directory within the installed package.
+
+## License
+
+This project is licensed under the BSD-3-Clause License. See the LICENSE file for details.

tachy_compiler-0.1.1/pyproject.toml

@@ -0,0 +1,40 @@
+[build-system]
+requires = ["setuptools>=68", "wheel"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "TACHY-Compiler"
+version = "0.1.1"
+description = "TACHY model compiler and platform conversion tools."
+readme = "README.md"
+requires-python = ">=3.8"
+license = {text = "BSD-3-Clause"}
+authors = [{name = "Deeper-I"}]
+
+dependencies = [
+  "numpy",
+  "networkx",
+  "onnx",
+  "commentjson",
+  "ordered-set",
+  "easydict",
+  "DDesignerAPI",
+]
+
+[project.optional-dependencies]
+yolov9 = [
+  "torch",
+]
+
+[tool.setuptools]
+include-package-data = true
+
+[tool.setuptools.packages.find]
+where = ["src"]
+include = [
+  "tachy_compiler",
+  "tachy_compiler.*",
+]
+
+[project.scripts]
+tachy = "tachy_compiler.cli:main"

tachy_compiler-0.1.1/setup.cfg

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

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/PKG-INFO

@@ -0,0 +1,93 @@
+Metadata-Version: 2.4
+Name: TACHY-Compiler
+Version: 0.1.1
+Summary: TACHY model compiler and platform conversion tools.
+Author: Deeper-I
+License: BSD-3-Clause
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+Requires-Dist: numpy
+Requires-Dist: networkx
+Requires-Dist: onnx
+Requires-Dist: commentjson
+Requires-Dist: ordered-set
+Requires-Dist: easydict
+Requires-Dist: DDesignerAPI
+Provides-Extra: yolov9
+Requires-Dist: torch; extra == "yolov9"
+
+# TACHY-Compiler
+
+TACHY-Compiler is a comprehensive deep learning compiler and platform conversion framework developed by Deeper-I. It is designed to bridge the gap between popular deep learning frameworks (like PyTorch) and the TACHY hardware accelerator runtime environment.
+
+This compiler handles model parsing, optimization, tensor alignment, and binary generation. It robustly supports large-scale models, complex operators, and dual-checkpoint pipelines out of the box.
+
+## Key Features
+
+- **PyTorch to ONNX to TACHY**: Seamless end-to-end compilation pipeline.
+- **Dual Checkpoint Compilation (`compile-dual`)**: Officially supports simultaneous parsing and compilation of Reference (`eval` / `pre`) and Target (`comp` / `opt`) checkpoints into a unified output structure.
+- **Large Model Support**: Built-in mechanisms to handle models larger than the Protobuf 2GB limit using `external_data` mapping.
+- **Robust ONNX Versioning**: Automatically utilizes ONNX Opset 18 for stability, equipped with automatic version conversion and fallback mechanisms to mitigate `onnxscript` conversion issues.
+- **MCP Server Integrated**: Exposes tools to LLMs and agentic frameworks via the Model Context Protocol (MCP) using `FastMCP`.
+- **Dynamic Binary Resolution**: Intelligently locates dependency binaries utilizing environment variables (`TACHY_UTIL_PATH`) and relative package paths.
+
+## Installation
+
+You can install TACHY-Compiler directly via PyPI:
+
+```bash
+pip install TACHY-Compiler
+```
+
+### Requirements
+- Python >= 3.8
+- `numpy`, `networkx`, `onnx`, `commentjson`, `ordered-set`, `easydict`, `DDesignerAPI`
+- (Optional) `torch` if utilizing the PyTorch to ONNX conversion pipeline.
+
+## CLI Usage
+
+TACHY-Compiler provides a simple, unified command-line interface under the `tachy` command.
+
+### 1. Dual Checkpoint Compilation
+Best for executing the full compilation pipeline for two checkpoints (e.g. `eval` and `comp`) simultaneously. Intermediate files are automatically managed and cleaned up if requested.
+
+```bash
+tachy compile-dual \
+    --eval_model ./eval_model.pth \
+    --comp_model ./comp_model.pth \
+    --input_shape "[1,3,224,224]" \
+    --output_dir ./dual_output \
+    --cleanup
+```
+Outputs will be cleanly structured into `./dual_output/final_binaries/eval/` and `./dual_output/final_binaries/comp/`.
+
+### 2. Single Model Pipeline Steps
+If you need granular control, you can execute the pipeline steps individually.
+
+**Convert PyTorch to ONNX**
+```bash
+tachy convert-pt2onnx ./model.pth ./model.onnx "[1,3,224,224]" False
+```
+
+**Convert ONNX to TACHY Format**
+```bash
+tachy convert-onnx ./model.onnx ./model.json
+```
+
+**Compile TACHY JSON to Runtime Binaries**
+```bash
+tachy compile ./model.json ./output_bin_dir
+```
+
+### 3. Model Context Protocol (MCP) Server
+You can spin up the built-in MCP server to allow LLMs and AI Agents to compile models on your behalf.
+```bash
+tachy mcp
+```
+
+## Environment Variables
+- `TACHY_UTIL_PATH` (Optional): An absolute path to the directory containing external utility binaries (`fc.out`, `block_4bit.out`). If not set, the compiler will intelligently look for the `optional` directory within the installed package.
+
+## License
+
+This project is licensed under the BSD-3-Clause License. See the LICENSE file for details.

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/SOURCES.txt

@@ -0,0 +1,37 @@
+README.md
+pyproject.toml
+src/TACHY_Compiler.egg-info/PKG-INFO
+src/TACHY_Compiler.egg-info/SOURCES.txt
+src/TACHY_Compiler.egg-info/dependency_links.txt
+src/TACHY_Compiler.egg-info/entry_points.txt
+src/TACHY_Compiler.egg-info/requires.txt
+src/TACHY_Compiler.egg-info/top_level.txt
+src/tachy_compiler/__init__.py
+src/tachy_compiler/cli.py
+src/tachy_compiler/mcp_server.py
+src/tachy_compiler/compiler/__init__.py
+src/tachy_compiler/compiler/constraint.py
+src/tachy_compiler/compiler/convert_layer2block.py
+src/tachy_compiler/compiler/convert_tachy2tachyrt.py
+src/tachy_compiler/compiler/editor.py
+src/tachy_compiler/compiler/extract.py
+src/tachy_compiler/compiler/format_prototxt.py
+src/tachy_compiler/compiler/functions.py
+src/tachy_compiler/compiler/graph.py
+src/tachy_compiler/compiler/map_address.py
+src/tachy_compiler/compiler/onnx2tachy.py
+src/tachy_compiler/compiler/pipeline.py
+src/tachy_compiler/compiler/runtime.py
+src/tachy_compiler/compiler/tachy_block.py
+src/tachy_compiler/compiler/tachy_format.py
+src/tachy_compiler/compiler/tachy_inference.py
+src/tachy_compiler/compiler/tachy_layers.py
+src/tachy_compiler/compiler/tachy_model.py
+src/tachy_compiler/compiler/tachy_prototxt.py
+src/tachy_compiler/compiler/tachy_viewer.py
+src/tachy_compiler/compiler/tachyrt_model.py
+src/tachy_compiler/compiler/verify.py
+src/tachy_compiler/platform_converter/__init__.py
+src/tachy_compiler/platform_converter/deploy_pt_yolov9.py
+src/tachy_compiler/platform_converter/pt2onnx.py
+src/tachy_compiler/platform_converter/yolov9_utils/__init__.py

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[console_scripts]
+tachy = tachy_compiler.cli:main

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/requires.txt

@@ -0,0 +1,10 @@
+numpy
+networkx
+onnx
+commentjson
+ordered-set
+easydict
+DDesignerAPI
+
+[yolov9]
+torch

tachy_compiler-0.1.1/src/TACHY_Compiler.egg-info/top_level.txt

@@ -0,0 +1 @@
+tachy_compiler

tachy_compiler-0.1.1/src/tachy_compiler/__init__.py

@@ -0,0 +1,10 @@
+"""
+TACHY compiler core package.
+"""
+
+from __future__ import annotations
+
+__all__ = ["__version__"]
+
+__version__ = "0.1.0"
+

tachy_compiler-0.1.1/src/tachy_compiler/cli.py

@@ -0,0 +1,41 @@
+import sys
+import subprocess
+
+def main():
+    if len(sys.argv) < 2 or sys.argv[1] in ("-h", "--help"):
+        print("TACHY-Compiler CLI")
+        print("Usage: tachy <command> [args...]")
+        print("\nAvailable commands:")
+        print("  convert-pt2onnx   Convert PyTorch model to ONNX")
+        print("  convert-onnx      Convert ONNX model to TACHY format")
+        print("  compile           Compile TACHY model to runtime binaries")
+        print("  compile-dual      Compile dual models (eval & comp) in one pipeline")
+        print("  mcp               Start the MCP server")
+        sys.exit(0)
+
+    command = sys.argv[1]
+    forward_args = sys.argv[2:]
+
+    if command == "convert-pt2onnx":
+        module = "tachy_compiler.platform_converter.pt2onnx"
+    elif command == "convert-onnx":
+        module = "tachy_compiler.compiler.onnx2tachy"
+    elif command == "compile":
+        module = "tachy_compiler.compiler.runtime"
+    elif command == "compile-dual":
+        module = "tachy_compiler.compiler.pipeline"
+    elif command == "mcp":
+        module = "tachy_compiler.mcp_server"
+    else:
+        print(f"Unknown command: {command}")
+        sys.exit(1)
+
+    cmd = [sys.executable, "-m", module] + forward_args
+    try:
+        result = subprocess.run(cmd)
+        sys.exit(result.returncode)
+    except KeyboardInterrupt:
+        sys.exit(1)
+
+if __name__ == "__main__":
+    main()

tachy_compiler-0.1.1/src/tachy_compiler/compiler/__init__.py

@@ -0,0 +1,8 @@
+"""
+Utility entrypoints for TACHY compiler.
+"""
+
+from __future__ import annotations
+
+__all__ = []
+

tachy_compiler-0.1.1/src/tachy_compiler/compiler/constraint.py

@@ -0,0 +1,65 @@
+import numpy as np
+
+def align_odd_width(block_config):
+    _, in_w, _ = block_config["input_shape"]
+    ker_h, ker_w, _, _ = block_config["kernel_shape"]
+    str_h, str_w = block_config["stride_shape"]
+    p_t, p_l, p_b, p_r, _ = block_config["padding_shape"]
+
+    is_odd = (in_w % 2 == 1)
+    is_ker_1x1 = (ker_h == 1 and ker_w == 1)
+    is_str_2x2 = (str_h == 2 and str_w == 2)
+    is_no_pad = ((p_t == 0) and (p_l == 0) and (p_b == 0) and (p_r == 0))
+
+    if is_odd and is_ker_1x1 and is_str_2x2 and is_no_pad:
+        block_config["padding_shape"][2] += 1 # add bottom padding
+        block_config["padding_shape"][3] += 1 # add right padding
+
+    return block_config
+
+def add_dummy_padding(block_config):
+    in_h, in_w, _       = block_config["input_shape"]
+    out_h, out_w, _     = block_config["output_shape"]
+    ker_h, ker_w, _, __ = block_config["kernel_shape"]
+    str_h, str_w        = block_config["stride_shape"]
+    p_t, p_l, p_b, p_r, p_s = block_config["padding_shape"]
+
+    op = block_config["operation"]
+    if op == "Conv":
+        if out_w < 7:
+            for i in range(8):
+                if (1 + (in_w - 1 + p_l + p_r + i - (ker_w - 1)) / str_w >= 7):
+                    p_s = i
+                    break
+            p_r += p_s
+    else:
+        if op == "MaxPool": # Warning: Only for synergy ai
+            if ker_w+1 > in_w:
+                add_pad = max((ker_w + 1 - in_w), 0)
+                p_r += add_pad
+                p_s += ((in_w + p_r) - ker_w)
+
+    block_config["padding_shape"] = [p_t, p_l, p_b, p_r, p_s]
+    return block_config
+
+    # def add_dummy_padding(block_config):
+    #     in_h, in_w, _ = block_config["input_shape"]
+    #     ker_h, ker_w, _, __ = block_config["kernel_shape"]
+    # 
+    #     p_t, p_l, p_b, p_r, p_s = block_config["padding_shape"]
+    # 
+    #     op = block_config["operation"]
+    #     if op == "Conv":
+    #         if in_w < 7:
+    #             add_pad = (7 - in_w)
+    #             p_r += add_pad
+    #             p_s += add_pad
+    #     else:
+    #         if op == "MaxPool": # Warning: Only for synergy ai
+    #             if ker_w+1 > in_w:
+    #                 add_pad = max((ker_w + 1 - in_w), 0)
+    #                 p_r += add_pad
+    #                 p_s += ((in_w + p_r) - ker_w)
+    # 
+    #     block_config["padding_shape"] = [p_t, p_l, p_b, p_r, p_s]
+    #     return block_config