From c7a08ec05420db295e17862929ac5db17d26636e Mon Sep 17 00:00:00 2001
From: 0xrushi <6279035+0xrushi@users.noreply.github.com>
Date: Sun, 12 Oct 2025 12:53:22 -0400
Subject: [PATCH 1/4] pyproject update

---
 extras/asr-services/pyproject.toml | 126 +++++++++++++++++++++++------
 1 file changed, 103 insertions(+), 23 deletions(-)

diff --git a/extras/asr-services/pyproject.toml b/extras/asr-services/pyproject.toml
index 2951167e..d076ddd7 100644
--- a/extras/asr-services/pyproject.toml
+++ b/extras/asr-services/pyproject.toml
@@ -1,12 +1,15 @@
+[build-system]
+requires = ["setuptools>=69", "wheel"]
+build-backend = "setuptools.build_meta"
+
 [project]
 name = "asr-services"
 version = "0.1.0"
 description = "Wyoming compatible WebSocket ASR server for friend-lite"
 requires-python = ">=3.10"
 
+# Core deps (NO torch/torchaudio here)
 dependencies = [
-  "torch>=2.3",
-  "torchaudio>=2.3",
   "websockets>=12.0",
   "numpy>=1.26",
   "soundfile>=0.12",
@@ -19,35 +22,112 @@ dependencies = [
   "httpx>=0.28.1",
 ]
 
-[tool.uv]
-compile-bytecode = true
-conflicts = [
-  [
-    { group = "demo" },
-    { group = "parakeet" }
-  ]
-]
-
-[tool.uv.sources]
-useful-moonshine-onnx = { git = "https://github.com/usefulsensors/moonshine.git", subdirectory = "moonshine-onnx" }
+# Choose ONE of these extras per machine at install time
+[project.optional-dependencies]
+cpu   = ["torch", "torchaudio"]
+cu118 = ["torch", "torchaudio"]
+cu121 = ["torch", "torchaudio"]
+cu124 = ["torch", "torchaudio"]
+cu126 = ["torch", "torchaudio"]
+cu128 = ["torch", "torchaudio"]  # CUDA 12.8 / Blackwell
 
 [dependency-groups]
 demo = [
-    "fastrtc>=0.0.23",
-    "gradio>=5.29.0",
-    "sounddevice>=0.5.1",
+  "fastrtc>=0.0.23",
+  "gradio>=5.29.0",
+  "sounddevice>=0.5.1",
 ]
 dev = [
-    "black>=25.1.0",
-    "requests>=2.31.0",
-    "pytest>=8.0.0",
+  "black>=25.1.0",
+  "requests>=2.31.0",
+  "pytest>=8.0.0",
 ]
-
 moonshine = [
   "useful-moonshine-onnx",
 ]
 parakeet = [
-    "nemo-toolkit[asr]>=2.2.0",
-    "cuda-python>=12.3",
-    "numpy>=1.26,<2.0",
+  "nemo-toolkit[asr]>=2.2.0",
+  "cuda-python>=12.3",
+  "numpy>=1.26,<2.0",
+]
+
+# --- Setuptools: restrict what gets packaged (fixes multi top-level dirs) ---
+[tool.setuptools]
+include-package-data = true
+
+[tool.setuptools.packages.find]
+# If your package is a directory named `asr_services/` with __init__.py:
+include = ["asr_services*"]
+exclude = ["debug*", "charts*", "results*", "model_cache*"]
+
+# If instead you have a single module file `asr_services.py`,
+# comment the block above and use:
+# [tool.setuptools]
+# py-modules = ["asr_services"]
+
+# --- uv config ---
+
+[tool.uv]
+compile-bytecode = true
+conflicts = [
+  # prevent installing both demo and parakeet groups
+  [{ group = "demo" }, { group = "parakeet" }],
+  # prevent mixing CPU and CUDA extras
+  [{ extra = "cpu" }, { extra = "cu118" }],
+  [{ extra = "cpu" }, { extra = "cu121" }],
+  [{ extra = "cpu" }, { extra = "cu124" }],
+  [{ extra = "cpu" }, { extra = "cu126" }],
+  [{ extra = "cpu" }, { extra = "cu128" }],
+]
+
+[tool.uv.sources]
+# your git source
+useful-moonshine-onnx = { git = "https://github.com/usefulsensors/moonshine.git", subdirectory = "moonshine-onnx" }
+
+# map extras → PyTorch wheel indexes
+torch = [
+  { index = "pytorch-cpu",   extra = "cpu"   },
+  { index = "pytorch-cu118", extra = "cu118" },
+  { index = "pytorch-cu121", extra = "cu121" },
+  { index = "pytorch-cu124", extra = "cu124" },
+  { index = "pytorch-cu126", extra = "cu126" },
+  { index = "pytorch-cu128", extra = "cu128" },
 ]
+torchaudio = [
+  { index = "pytorch-cpu",   extra = "cpu"   },
+  { index = "pytorch-cu118", extra = "cu118" },
+  { index = "pytorch-cu121", extra = "cu121" },
+  { index = "pytorch-cu124", extra = "cu124" },
+  { index = "pytorch-cu126", extra = "cu126" },
+  { index = "pytorch-cu128", extra = "cu128" },
+]
+
+[[tool.uv.index]]
+name = "pytorch-cpu"
+url  = "https://download.pytorch.org/whl/cpu"
+explicit = true
+
+[[tool.uv.index]]
+name = "pytorch-cu118"
+url  = "https://download.pytorch.org/whl/cu118"
+explicit = true
+
+[[tool.uv.index]]
+name = "pytorch-cu121"
+url  = "https://download.pytorch.org/whl/cu121"
+explicit = true
+
+[[tool.uv.index]]
+name = "pytorch-cu124"
+url  = "https://download.pytorch.org/whl/cu124"
+explicit = true
+
+[[tool.uv.index]]
+name = "pytorch-cu126"
+url  = "https://download.pytorch.org/whl/cu126"
+explicit = true
+
+[[tool.uv.index]]
+name = "pytorch-cu128"
+url  = "https://download.pytorch.org/whl/cu128"
+explicit = true

From 22502e88a816c47732193e7b17b171e95ea2fae6 Mon Sep 17 00:00:00 2001
From: 0xrushi <6279035+0xrushi@users.noreply.github.com>
Date: Mon, 13 Oct 2025 19:52:28 -0400
Subject: [PATCH 2/4] Update Docker configuration to support optional
 installation of extras and remove deprecated files

- Added INSTALL_EXTRAS argument in docker-compose.yml for customizable package installation.
- Updated Dockerfile_Parakeet to utilize INSTALL_EXTRAS for pip installation.
- Removed obsolete Dockerfile_Parakeet.blackwell and pyproject.blackwell.toml files.
---
 extras/asr-services/Dockerfile_Parakeet       |  8 ++-
 .../Dockerfile_Parakeet.blackwell             | 25 --------
 extras/asr-services/docker-compose.yml        |  2 +
 extras/asr-services/env.template              |  6 ++
 extras/asr-services/pyproject.blackwell.toml  | 62 -------------------
 5 files changed, 14 insertions(+), 89 deletions(-)
 delete mode 100644 extras/asr-services/Dockerfile_Parakeet.blackwell
 create mode 100644 extras/asr-services/env.template
 delete mode 100644 extras/asr-services/pyproject.blackwell.toml

diff --git a/extras/asr-services/Dockerfile_Parakeet b/extras/asr-services/Dockerfile_Parakeet
index a67bd494..e6290254 100644
--- a/extras/asr-services/Dockerfile_Parakeet
+++ b/extras/asr-services/Dockerfile_Parakeet
@@ -12,8 +12,12 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 
 # Dependency manifest first for cache‑friendly installs
 COPY pyproject.toml uv.lock ./
+ARG INSTALL_EXTRAS
+
 RUN --mount=type=cache,target=/root/.cache/uv \
-    uv sync --no-install-project --group parakeet
+    uv venv && \
+    uv pip install -e "${INSTALL_EXTRAS}" --group demo --group dev --group parakeet
+
 
 # Should prepare the .venv for use :)
 
@@ -22,7 +26,7 @@ FROM python:3.12-slim-bookworm AS runtime
 ENV PYTHONUNBUFFERED=1
 WORKDIR /app
 
-RUN apt-get update && apt-get install -y --no-install-recommends libsndfile1 build-essential portaudio19-dev curl && rm -rf /var/lib/apt/lists/*
+RUN apt-get update && apt-get install -y --no-install-recommends libsndfile1 build-essential ffmpeg portaudio19-dev curl && rm -rf /var/lib/apt/lists/*
 
 COPY --from=builder /app /app
 COPY . .
diff --git a/extras/asr-services/Dockerfile_Parakeet.blackwell b/extras/asr-services/Dockerfile_Parakeet.blackwell
deleted file mode 100644
index 2870eb27..00000000
--- a/extras/asr-services/Dockerfile_Parakeet.blackwell
+++ /dev/null
@@ -1,25 +0,0 @@
-# syntax=docker/dockerfile:1
-
-#########################  builder #################################
-FROM ghcr.io/astral-sh/uv:python3.12-bookworm-slim AS builder
-WORKDIR /app
-
-# NeMo and texterrors need libs and C++ compiler
-RUN apt-get update && apt-get install -y --no-install-recommends \
-        libsndfile1 \
-        build-essential git portaudio19-dev ffmpeg \
-    && rm -rf /var/lib/apt/lists/*
-
-# Dependency manifest first for cache‑friendly installs
-COPY pyproject.blackwell.toml ./pyproject.toml
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv sync --no-install-project --group parakeet
-
-RUN uv pip install --upgrade --no-deps --force-reinstall "numpy<2.0"
-
-COPY . .
-
-ENV PATH="/app/.venv/bin:$PATH"
-
-EXPOSE 8765
-CMD ["python", "parakeet-offline.py", "--port", "8765"]
diff --git a/extras/asr-services/docker-compose.yml b/extras/asr-services/docker-compose.yml
index 6ea39a77..935a6989 100644
--- a/extras/asr-services/docker-compose.yml
+++ b/extras/asr-services/docker-compose.yml
@@ -3,6 +3,8 @@ services:
     build:
       context: .
       dockerfile: Dockerfile_Parakeet
+      args:
+        - INSTALL_EXTRAS=${INSTALL_EXTRAS:-.[cu121]}
     image: parakeet-asr:latest
     ports:
       - "${PARAKEET_HOST_PORT:-8767}:${PARAKEET_CONTAINER_PORT:-8765}"
diff --git a/extras/asr-services/env.template b/extras/asr-services/env.template
new file mode 100644
index 00000000..15ab3e2e
--- /dev/null
+++ b/extras/asr-services/env.template
@@ -0,0 +1,6 @@
+# ASR Services Environment Configuration Template
+# Copy this file to .env and configure the values for your environment
+
+# Install extras for PyTorch (CUDA version)
+# Options: .[cu118], .[cu121], .[cu128], .[cpu], or just . for CPU-only
+# INSTALL_EXTRAS=.[cu121]
diff --git a/extras/asr-services/pyproject.blackwell.toml b/extras/asr-services/pyproject.blackwell.toml
deleted file mode 100644
index 75435a19..00000000
--- a/extras/asr-services/pyproject.blackwell.toml
+++ /dev/null
@@ -1,62 +0,0 @@
-[project]
-name = "asr-services"
-version = "0.1.0"
-description = "Wyoming compatible WebSocket ASR server for friend-lite"
-requires-python = ">=3.10"
-
-dependencies = [
-  "torch>=2.3",
-  "torchaudio>=2.3",
-  "websockets>=12.0",
-  "numpy>=1.26",
-  "soundfile>=0.12",
-  "friend-lite-sdk>=0.1.0",
-  "fastapi>=0.115.12",
-  "uvicorn>=0.34.2",
-  "silero-vad>=5.1.2",
-  "wyoming>=1.6.1",
-  "easy-audio-interfaces>=0.4.2",
-]
-
-[dependency-groups]
-demo = [
-  "fastrtc>=0.0.23",
-  "gradio>=4.0",
-  "sounddevice>=0.5.1",
-]
-dev = [
-  "black>=25.1.0",
-]
-#moonshine = [
-#  "useful-moonshine-onnx",
-#]
-parakeet = [
-  "nemo-toolkit[asr]>=2.2.0",
-  "cuda-python>=12.3",
-  "numpy>=1.26,<2.0",
-]
-
-[tool.uv]
-compile-bytecode = false
-conflicts = [
-  [
-    { group = "demo" },
-    { group = "parakeet" }
-  ]
-]
-
-[tool.uv.sources]
-#useful-moonshine-onnx = { git = "https://github.com/usefulsensors/moonshine.git", subdirectory = "moonshine-onnx" }
-torchaudio  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torchvision  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torch = [
-  { index  = "pytorch-cu128", marker = "sys_platform != 'darwin'" },
-]
-
-[[tool.uv.index]]
-name = "pypi"
-url  = "https://pypi.org/simple"
-
-[[tool.uv.index]]
-name = "pytorch-cu128"
-url  = "https://download.pytorch.org/whl/cu128"

From ce5f6a912b73e0d0cd47f8cc0c2e69bee690d015 Mon Sep 17 00:00:00 2001
From: 0xrushi <6279035+0xrushi@users.noreply.github.com>
Date: Mon, 13 Oct 2025 19:53:11 -0400
Subject: [PATCH 3/4] Remove obsolete ASR services environment template file

---
 extras/asr-services/.env.template | 27 ---------------------------
 1 file changed, 27 deletions(-)
 delete mode 100644 extras/asr-services/.env.template

diff --git a/extras/asr-services/.env.template b/extras/asr-services/.env.template
deleted file mode 100644
index 8f367ece..00000000
--- a/extras/asr-services/.env.template
+++ /dev/null
@@ -1,27 +0,0 @@
-# ASR Services Configuration
-# Copy this file to .env and configure as needed
-
-# Parakeet ASR Model Selection
-PARAKEET_MODEL=nvidia/parakeet-tdt-0.6b-v3
-
-# Service Port Configuration
-PARAKEET_HOST_PORT=8767
-PARAKEET_CONTAINER_PORT=8765
-
-# Enhanced Chunking Configuration for Long Audio
-# Enable/disable chunking for long audio processing
-CHUNKING_ENABLED=true
-
-# Duration of each audio chunk in seconds (recommended: 20-40s)
-CHUNK_DURATION_SECONDS=120.0
-
-# Overlap duration between chunks in seconds (recommended: 3-7s)
-OVERLAP_DURATION_SECONDS=10.0
-
-# Minimum audio duration to trigger chunking (in seconds)
-# Audio shorter than this will use single-pass processing
-MIN_AUDIO_FOR_CHUNKING=120.0
-
-# Confidence threshold for overlap reconciliation (0.0-1.0)
-# Higher values prefer higher confidence words during overlap resolution
-CONFIDENCE_THRESHOLD=0.8
\ No newline at end of file

From 3cc7ca00f548e791975e78f4dc042e0f547d7c7f Mon Sep 17 00:00:00 2001
From: 0xrushi <6279035+0xrushi@users.noreply.github.com>
Date: Mon, 13 Oct 2025 20:37:41 -0400
Subject: [PATCH 4/4] Refactor Dockerfiles and update dependency management

- Updated Dockerfile for advanced backend to use `uv pip install` with deepgram extra and removed legacy commands.
- Deleted obsolete Dockerfile.blackwell and pyproject.blackwell.toml files.
- Enhanced speaker recognition Dockerfile to support CPU/GPU installation modes with dynamic CUDA version handling.
- Updated pyproject.toml files to streamline dependency management and removed deprecated dependency groups.
---
 backends/advanced/Dockerfile                  | 11 ++--
 backends/advanced/Dockerfile.blackwell        | 33 -----------
 backends/advanced/pyproject.blackwell.toml    | 53 ------------------
 backends/advanced/pyproject.toml              | 44 ++++++++-------
 extras/speaker-recognition/Dockerfile         | 12 ++--
 .../speaker-recognition/Dockerfile.blackwell  | 40 -------------
 extras/speaker-recognition/docker-compose.yml |  2 +
 .../pyproject.blackwell.toml                  | 56 -------------------
 extras/speaker-recognition/pyproject.toml     | 36 +++++-------
 9 files changed, 50 insertions(+), 237 deletions(-)
 delete mode 100644 backends/advanced/Dockerfile.blackwell
 delete mode 100644 backends/advanced/pyproject.blackwell.toml
 delete mode 100644 extras/speaker-recognition/Dockerfile.blackwell
 delete mode 100644 extras/speaker-recognition/pyproject.blackwell.toml

diff --git a/backends/advanced/Dockerfile b/backends/advanced/Dockerfile
index be3e1019..10407367 100644
--- a/backends/advanced/Dockerfile
+++ b/backends/advanced/Dockerfile
@@ -19,16 +19,12 @@ WORKDIR /app
 
 # Copy package structure and dependency files first
 COPY pyproject.toml README.md ./
-COPY uv.lock .
 RUN mkdir -p src/advanced_omi_backend
 COPY src/advanced_omi_backend/__init__.py src/advanced_omi_backend/
 
-# Install dependencies using uv with deepgram extra
-# Use cache mount for BuildKit, fallback for legacy builds
-# RUN --mount=type=cache,target=/root/.cache/uv \
-#     uv sync --extra deepgram
+# Install dependencies using uv pip with deepgram extra
 # Fallback for legacy Docker builds (CI compatibility)
-RUN uv sync --extra deepgram
+RUN uv pip install --system ".[deepgram]" -f https://download.pytorch.org/whl/cpu
 
 # Copy all application code
 COPY . .
@@ -38,6 +34,7 @@ COPY . .
 COPY memory_config.yaml* ./
 COPY diarization_config.json* ./
 
+ENV PATH="/app/.venv/bin:$PATH"
 
 # Run the application
-CMD ["uv", "run", "--extra", "deepgram", "python3", "src/advanced_omi_backend/main.py"]
+CMD ["python", "src/advanced_omi_backend/main.py"]
diff --git a/backends/advanced/Dockerfile.blackwell b/backends/advanced/Dockerfile.blackwell
deleted file mode 100644
index 892541a9..00000000
--- a/backends/advanced/Dockerfile.blackwell
+++ /dev/null
@@ -1,33 +0,0 @@
-FROM python:3.12-slim-bookworm AS builder
-
-# Install system dependencies for building
-RUN apt-get update && \
-    apt-get install -y --no-install-recommends \
-    build-essential \
-    libsndfile1 \
-    git \
-    curl \
-    ffmpeg \
-    && rm -rf /var/lib/apt/lists/*
-
-# Install uv
-COPY --from=ghcr.io/astral-sh/uv:0.6.10 /uv /uvx /bin/
-
-# Set up the working directory
-WORKDIR /app
-
-# Copy dependency files
-COPY pyproject.blackwell.toml pyproject.toml
-COPY README.md .
-
-# Install dependencies using uv
-RUN --mount=type=cache,target=/root/.cache/uv \
-    uv sync
-
-
-# Copy application code
-COPY . .
-
-
-# Run the application
-CMD ["uv", "run", "python3", "src/advanced_omi_backend/main.py"]
diff --git a/backends/advanced/pyproject.blackwell.toml b/backends/advanced/pyproject.blackwell.toml
deleted file mode 100644
index 2f661dea..00000000
--- a/backends/advanced/pyproject.blackwell.toml
+++ /dev/null
@@ -1,53 +0,0 @@
-[project]
-name = "advanced-omi-backend"
-version = "0.1.0"
-description = "Add your description here"
-readme = "README.md"
-requires-python = ">=3.12"
-dependencies = [
-    "easy-audio-interfaces>=0.5.1",
-    "fastapi>=0.115.12",
-    "mem0ai>=0.1.111",
-    "motor>=3.7.1",
-    "ollama>=0.4.8",
-    "python-dotenv>=1.1.0",
-    "uvicorn>=0.34.2",
-    "wyoming>=1.6.1",
-    "aiohttp>=3.8.0",
-    "langfuse==3.3.0",
-    "spacy>=3.8.2",
-    "en-core-web-sm @ https://github.com/explosion/spacy-models/releases/download/en_core_web_sm-3.7.1/en_core_web_sm-3.7.1-py3-none-any.whl",
-]
-
-[dependency-groups]
-deepgram = [
-    "deepgram-sdk>=4.0.0",
-]
-dev = [
-    "black>=25.1.0",
-    "isort>=6.0.1",
-]
-tests = [
-    "pytest>=8.4.1",
-    "pytest-asyncio>=1.0.0",
-]
-
-
-[tool.isort]
-profile = "black"
-
-[tool.uv.sources]
-useful-moonshine-onnx = { git = "https://github.com/usefulsensors/moonshine.git", subdirectory = "moonshine-onnx" }
-torchaudio  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torchvision  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torch = [
-  { index  = "pytorch-cu128", marker = "sys_platform != 'darwin'" },
-]
-
-[[tool.uv.index]]
-name = "pypi"
-url  = "https://pypi.org/simple"
-
-[[tool.uv.index]]
-name = "pytorch-cu128"
-url  = "https://download.pytorch.org/whl/cu128"
diff --git a/backends/advanced/pyproject.toml b/backends/advanced/pyproject.toml
index c355509f..29debefb 100644
--- a/backends/advanced/pyproject.toml
+++ b/backends/advanced/pyproject.toml
@@ -1,3 +1,7 @@
+[build-system]
+requires = ["setuptools>=69", "wheel"]
+build-backend = "setuptools.build_meta"
+
 [project]
 name = "advanced-omi-backend"
 version = "0.1.0"
@@ -7,7 +11,7 @@ requires-python = ">=3.12"
 dependencies = [
     "easy-audio-interfaces>=0.7.1", # we need to add local-audio for scripts/local-audio.py | If we don't need that, we can remove this, and then remove portaudio19-dev from Dockerfile
     "fastapi>=0.115.12",
-    "mem0ai", # Using main branch with PR #3250 AsyncMemory fix
+    "mem0ai @ git+https://github.com/AnkushMalaker/mem0.git@main", # Using main branch with PR #3250 AsyncMemory fix
     "langchain_neo4j",
     "motor>=3.7.1",
     "ollama>=0.4.8",
@@ -32,9 +36,26 @@ local-audio = [
     "easy-audio-interfaces[local-audio]>=0.7.1",
 ]
 
-[build-system]
-requires = ["setuptools>=61.0", "wheel"]
-build-backend = "setuptools.build_meta"
+# ---- Optional compute backends (choose ONE when installing) ----
+# Note: with plain pip, point to the PyTorch wheel index for the chosen extra (examples below).
+cpu   = ["torch", "torchaudio"]
+cu118 = ["torch", "torchaudio"]
+cu121 = ["torch", "torchaudio"]
+cu124 = ["torch", "torchaudio"]
+cu126 = ["torch", "torchaudio"]
+cu128 = ["torch", "torchaudio"]  # CUDA 12.8 / Blackwell
+
+# ---- Dev/test extras (moved from `dependency-groups` to standard extras) ----
+dev = [
+    "black>=25.1.0",
+    "isort>=6.0.1",
+    "pre-commit>=4.2.0",
+    "pre-commit-uv>=4.1.4",
+]
+test = [
+    "pytest>=8.4.1",
+    "pytest-asyncio>=1.0.0",
+]
 
 [tool.setuptools.packages.find]
 where = ["src"]
@@ -45,22 +66,7 @@ profile = "black"
 [tool.black]
 line-length = 100
 
-[tool.uv.sources]
-mem0ai = { git = "https://github.com/AnkushMalaker/mem0.git", rev = "async-client-unbound-var-fix" }
-
 [tool.pytest.ini_options]
 markers = [
     "integration: marks tests as integration tests",
 ]
-
-[dependency-groups]
-dev = [
-    "black>=25.1.0",
-    "isort>=6.0.1",
-    "pre-commit>=4.2.0",
-    "pre-commit-uv>=4.1.4",
-]
-test = [
-    "pytest>=8.4.1",
-    "pytest-asyncio>=1.0.0",
-]
diff --git a/extras/speaker-recognition/Dockerfile b/extras/speaker-recognition/Dockerfile
index d4d7ec72..af76bb96 100644
--- a/extras/speaker-recognition/Dockerfile
+++ b/extras/speaker-recognition/Dockerfile
@@ -1,4 +1,4 @@
-FROM python:3.12-bookworm
+FROM python:3.12-bookworm AS builder
 
 # Install system dependencies including PortAudio for pyaudio
 RUN apt-get update && apt-get install -y \
@@ -17,7 +17,6 @@ COPY --from=ghcr.io/astral-sh/uv:latest /uv /bin/uv
 
 # Copy dependency files first (for better caching)
 COPY pyproject.toml ./
-COPY uv.lock ./
 
 # Create minimal package structure for dependency installation
 RUN mkdir -p src/simple_speaker_recognition
@@ -26,10 +25,11 @@ COPY src/simple_speaker_recognition/__init__.py src/simple_speaker_recognition/
 # Install dependencies and package  
 # Use build arg to control CPU vs GPU mode
 ARG COMPUTE_MODE=cpu
+ARG CUDA_VERSION
 RUN if [ "$COMPUTE_MODE" = "gpu" ]; then \
-        uv sync --no-default-groups --group gpu; \
+        uv pip install --system ".[${CUDA_VERSION}]" -f https://download.pytorch.org/whl/${CUDA_VERSION}; \
     else \
-        uv sync --no-default-groups --group cpu; \
+        uv pip install --system ".[cpu]" -f https://download.pytorch.org/whl/cpu; \
     fi
 
 # Create directories
@@ -37,10 +37,10 @@ RUN mkdir -p /app/audio_chunks /app/debug /app/data /models
 
 # Set environment variables
 ENV HF_HOME=/models
-ENV PYTHONPATH=/app
+ENV PATH="/app/.venv/bin:$PATH"
 
 # Expose port
 EXPOSE 8085
 
 # Run the service
-CMD ["uv", "run", "--no-dev", "simple-speaker-service"] 
\ No newline at end of file
+CMD ["python", "-m", "simple_speaker_recognition"]
diff --git a/extras/speaker-recognition/Dockerfile.blackwell b/extras/speaker-recognition/Dockerfile.blackwell
deleted file mode 100644
index 0ee7705c..00000000
--- a/extras/speaker-recognition/Dockerfile.blackwell
+++ /dev/null
@@ -1,40 +0,0 @@
-FROM python:3.10-slim
-
-# Install system dependencies
-RUN apt-get update && apt-get install -y \
-    build-essential \
-    git \
-    ffmpeg \
-    portaudio19-dev \
-    libasound2-dev \
-    && rm -rf /var/lib/apt/lists/*
-
-WORKDIR /app
-
-# Install uv
-COPY --from=ghcr.io/astral-sh/uv:latest /uv /bin/uv
-
-# Copy dependency files
-COPY pyproject.blackwell.toml pyproject.toml
-
-# Install dependencies
-RUN uv sync
-
-# Copy application code
-COPY speaker_service.py ./
-# COPY client.py ./
-
-# Create directories
-RUN mkdir -p /app/audio_chunks /app/debug /models
-
-# Set environment variables
-ENV HF_HOME=/models
-ENV PYTHONPATH=/app
-
-# Expose port
-EXPOSE 8001
-
-ENV PATH="/app/.venv/bin:$PATH"
-
-# Run the service
-CMD ["python", "speaker_service.py"] 
diff --git a/extras/speaker-recognition/docker-compose.yml b/extras/speaker-recognition/docker-compose.yml
index 768c00fe..eb735647 100644
--- a/extras/speaker-recognition/docker-compose.yml
+++ b/extras/speaker-recognition/docker-compose.yml
@@ -8,6 +8,7 @@ services:
       dockerfile: Dockerfile
       args:
         COMPUTE_MODE: cpu
+        CUDA_VERSION: cu121
     image: speaker-recognition:latest
     env_file:
       - .env
@@ -46,6 +47,7 @@ services:
       dockerfile: Dockerfile
       args:
         COMPUTE_MODE: gpu
+        CUDA_VERSION: cu121
     deploy:
       resources:
         reservations:
diff --git a/extras/speaker-recognition/pyproject.blackwell.toml b/extras/speaker-recognition/pyproject.blackwell.toml
deleted file mode 100644
index 11bcbbe2..00000000
--- a/extras/speaker-recognition/pyproject.blackwell.toml
+++ /dev/null
@@ -1,56 +0,0 @@
-[project]
-name = "speaker-recognition-service"
-version = "0.1.0"
-description = "Speaker recognition and diarization service for friend-lite"
-requires-python = ">=3.10"
-
-dependencies = [
-    "fastapi>=0.115.12",
-    "uvicorn>=0.34.2",
-    "numpy>=1.26",
-    "scipy>=1.10.0",
-    "torch>=2.7.0",
-    "torchaudio>=2.7.0",
-    "pyannote.audio>=3.3.2",
-    "faiss-cpu>=1.7.1",
-    "aiohttp>=3.8.0",
-    "python-multipart>=0.0.6",
-    "pydantic>=2.0.0",
-    "soundfile>=0.12",
-    "easy-audio-interfaces[local-audio]>=0.7.1",
-    "pydantic-settings>=2.10.1",
-    "deepgram-sdk>=4.7.0",
-    "pydub>=0.25.1",
-    "yt-dlp>=2025.7.21",
-]
-
-[dependency-groups]
-dev = [
-    "black>=25.1.0",
-    "isort>=6.0.1",
-]
-gpu = [
-    "faiss-gpu>=1.7.1",
-]
-
-[tool.isort]
-profile = "black" 
-
-
-[tool.uv]
-compile-bytecode = false
-
-[tool.uv.sources]
-torchaudio  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torchvision  = { index = "pytorch-cu128",   marker = "sys_platform != 'darwin'" }
-torch = [
-  { index  = "pytorch-cu128", marker = "sys_platform != 'darwin'" },
-]
-
-[[tool.uv.index]]
-name = "pypi"
-url  = "https://pypi.org/simple"
-
-[[tool.uv.index]]
-name = "pytorch-cu128"
-url  = "https://download.pytorch.org/whl/cu128"
diff --git a/extras/speaker-recognition/pyproject.toml b/extras/speaker-recognition/pyproject.toml
index e1fc1267..799f5ccc 100644
--- a/extras/speaker-recognition/pyproject.toml
+++ b/extras/speaker-recognition/pyproject.toml
@@ -45,41 +45,31 @@ local-audio = [
     "easy-audio-interfaces[local-audio]>=0.7.1",
 ]
 
-[project.scripts]
-simple-speaker-service = "simple_speaker_recognition.api.service:main"
-simple-speaker-web = "simple_speaker_recognition.web.app:main"
-
-[dependency-groups]
+# CPU / GPU install modes
 cpu = [
     "faiss-cpu>=1.9",
     "torch>=2.0.0",
     "torchaudio>=2.0.0",
     "numpy>=1.26,<2",
 ]
-gpu = [
-    "faiss-cpu>=1.9",  # Use CPU FAISS for compatibility, GPU PyTorch for performance
-    "torch>=2.0.0",
-    "torchaudio>=2.0.0",
-]
+cu118 = ["torch>=2.0.0", "torchaudio>=2.0.0"]
+cu121 = ["torch>=2.0.0", "torchaudio>=2.0.0"]
+cu124 = ["torch>=2.0.0", "torchaudio>=2.0.0"]
+cu126 = ["torch>=2.0.0", "torchaudio>=2.0.0"]
+cu128 = ["torch>=2.0.0", "torchaudio>=2.0.0"]
+
 dev = [
     "black>=25.1.0",
     "isort>=6.0.1",
 ]
-
 test = [
     "pytest",
-    "requests",  # For integration tests
+    "requests",
 ]
 
-[tool.uv]
-# dev & cpu are installed automatically
-default-groups = ["dev", "cpu"]
-# cpu and gpu can never coexist
-conflicts = [
-    [
-        { group = "cpu" },
-        { group = "gpu" },
-    ],
-]
+[project.scripts]
+simple-speaker-service = "simple_speaker_recognition.api.service:main"
+simple-speaker-web = "simple_speaker_recognition.web.app:main"
+
 [tool.isort]
-profile = "black" 
+profile = "black"