prototyping

model_management.ps1

@@ -0,0 +1,40 @@
+$DOWNLOAD_MODEL = $false
+$OPTIMISE = $true
+
+# $TARGET_PTH = "B:\projects\MOP-TOM\models\models--BAAI--bge-reranker-v2-m3\snapshots\953dc6f6f85a1b2dbfca4c34a2796e7dde08d41e"
+# $MODEL_NAME = 'BAAI/bge-reranker-v2-m3'
+# $TARGET_PTH = "B:\projects\MOP-TOM\models\models--jinaai--jina-reranker-v3\snapshots\10fb694fc21f7a710a563ff1eb977a460f3868e4"
+$TARGET_PTH = "B:\projects\MOP-TOM\models\models--jinaai--jina-reranker-v2-base-multilingual\snapshots\9cfeff2df7d40d1b78e75e5e9cebec92a99813c9"
+$MODEL_NAME = 'jinaai/jina-reranker-v3'
+$base = Get-Location
+$model_path = Join-Path $base "..\models" -Resolve
+$MODEL_EXPORT = Join-Path $model_path "onnx_tmp"
+$MODEL_EXPORT_QUANT = Join-Path $MODEL_EXPORT "quant"
+
+Write-Host "Path to base: $base"
+Write-Host "Path to model folder: $model_path"
+
+if (Test-Path -Path $model_path) {
+    Write-Host "Model directory found"
+} else {
+    Write-Host "Model directory not found: $model_path"
+    exit
+}
+$Env:SENTENCE_TRANSFORMERS_HOME = $model_path
+$Env:TRANSFORMERS_CACHE = $model_path
+$Env:HF_HOME = $model_path
+$Env:HF_HUB_DISABLE_SYMLINKS = 1
+
+if ($DOWNLOAD_MODEL) {
+    pdm run hf download $MODEL_NAME
+}
+
+if ($OPTIMISE) {
+    pdm run optimum-cli export onnx -m $TARGET_PTH --task text-classification --trust-remote-code --optimize O2 $MODEL_EXPORT
+    # pdm run optimum-cli export onnx -m $TARGET_PTH --task text-classification $MODEL_EXPORT
+}
+
+if ($OPTIMISE) {
+    pdm run optimum-cli onnxruntime quantize --onnx_model $MODEL_EXPORT -o $MODEL_EXPORT_QUANT --avx2
+    # pdm run optimum-cli onnxruntime quantize --onnx_model $TARGET_PTH -o $MODEL_EXPORT_QUANT --avx2
+}

pyproject.toml

@@ -5,8 +5,8 @@ description = "use case for M.O.P. - T.O.M.: search semantically for problems an
 authors = [
     {name = "d-opt GmbH, resp. Florian Förster", email = "f.foerster@d-opt.com"},
 ]
-dependencies = []
-requires-python = ">=3.11"
+dependencies = ["torch>=2.10.0", "torchvision>=0.25.0", "dopt-basics>=0.2.4", "llama-cpp-python>=0.3.16", "lancedb>=0.30.0", "pydantic>=2.12.5", "sentence-transformers>=5.3.0", "onnx>=1.20.1", "onnxruntime>=1.24.4", "tantivy>=0.25.1"]
+requires-python = "<3.14,>=3.11"
 readme = "README.md"
 license = {text = "LicenseRef-Proprietary"}
 
@@ -122,14 +122,28 @@ package-dir = "src"
 [tool.pdm.resolution]
 respect-source-order = true
 
+[[tool.pdm.source]]
+name = "torch"
+url = "https://download.pytorch.org/whl/cu128"
+include_packages = ["torch*"]
+exclude_packages = ["llama-cpp*"]
+
+# [[tool.pdm.source]]
+# name = "llama-cpp-python-cuda"
+# url = "https://abetlen.github.io/llama-cpp-python/whl/cu124"
+# include_packages = ["llama-cpp-py*"]
+
 [[tool.pdm.source]]
 name = "private"
 url = "http://localhost:8001/simple"
 verify_ssl = false
+include_packages = ["llama-cpp-python*"]
 
 [[tool.pdm.source]]
 name = "pypi"
 url = "https://pypi.org/simple"
+exclude_packages = ["llama-cpp*"]
+
 [dependency-groups]
 tests = [
     "pytest>=8.3.4",
@@ -143,8 +157,20 @@ dev = [
     "pdoc3>=0.11.5",
     "bump-my-version>=1.1.1",
     "nox>=2025.2.9",
+    "huggingface-hub<1.0",
+    "tqdm>=4.66.5",
+    "peft>=0.18.1",
+    "einops>=0.8.2",
+    "onnx>=1.20.1",
+    "triton-windows>=3.6.0.post26",
+    "xformers>=0.0.34",
+    "optimum[onnxruntime]>=2.1.0",
 ]
 nb = [
     "jupyterlab>=4.3.5",
     "ipywidgets>=8.1.5",
 ]
+data = [
+    "polars>=1.38.1",
+    "striprtf>=0.0.29",
+]