mop-tom/semantic-search
1
0
Fork 0
generated from dopt-python/py311
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

prototyping

Browse Source
This commit is contained in:
Florian Förster 2026-04-10 15:26:30 +02:00
parent 64b0216c62
commit 7904a6ab2d
1 changed files with 17 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
ONNX_quantize.ps1 → model_management.ps1
View File

@ -1,8 +1,14 @@
$TARGET_PTH = "B:\projects\MOP-TOM\models\models--BAAI--bge-reranker-v2-m3\snapshots\953dc6f6f85a1b2dbfca4c34a2796e7dde08d41e"
$MODEL_NAME = 'BAAI/bge-reranker-v2-m3'
$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_test"
$MODEL_EXPORT = Join-Path $model_path "onnx_tmp"
$MODEL_EXPORT_QUANT = Join-Path $MODEL_EXPORT "quant"
Write-Host "Path to base: $base"
@ -17,13 +23,18 @@ if (Test-Path -Path $model_path) {
$Env:SENTENCE_TRANSFORMERS_HOME = $model_path
$Env:TRANSFORMERS_CACHE = $model_path
$Env:HF_HOME = $model_path
$Env:HF_HUB_DISABLE_SYMLINKS = 1
if ($true) {
pdm run optimum-cli export onnx -m $TARGET_PTH --task text-classification --optimize O1 $MODEL_EXPORT
if ($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 $MODEL_EXPORT
}
if ($true) {
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
}