diff --git a/other/non-avx tgi docker/Dockerfile b/other/non-avx tgi docker/Dockerfile
index 9855f9c..378c9e3 100644
--- a/other/non-avx tgi docker/Dockerfile	
+++ b/other/non-avx tgi docker/Dockerfile	
@@ -1,3 +1,7 @@
+# syntax = docker/dockerfile:experimental
+
+# DOCKER_BUILDKIT=1 DOCKER_CLI_EXPERIMENTAL=enabled
+
 # Rust builder
 FROM lukemathwalker/cargo-chef:latest-rust-1.71 AS chef
 WORKDIR /usr/src
@@ -40,16 +44,13 @@ RUN cargo build --release
 
 # Python builder
 # Adapted from: https://github.com/pytorch/pytorch/blob/master/Dockerfile
-# FROM debian:bullseye-slim as pytorch-install
-FROM nvidia/cuda:11.8.0-devel-ubuntu20.04 as pytorch-install
+FROM ubuntu:20.04 as dev-base
+# FROM nvidia/cuda:11.8.0-devel-ubuntu20.04 as pytorch-build
 
 ARG PYTORCH_VERSION=2.0.1
 ARG PYTHON_VERSION=3.9
 # Keep in sync with `server/pyproject.toml
 ARG CUDA_VERSION=11.8
-ARG MAMBA_VERSION=23.1.0-1
-ARG CUDA_CHANNEL=nvidia
-ARG INSTALL_CHANNEL=pytorch
 
 # Automatically set by buildx
 ARG TARGETPLATFORM
@@ -58,24 +59,26 @@ ENV PATH /opt/conda/bin:$PATH
 
 RUN apt-get update
 
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends gnupg2
+
 # Add new repo to install gcc 11 on Ubuntu 20.04
 RUN echo "deb http://ppa.launchpad.net/ubuntu-toolchain-r/test/ubuntu focal main" >> /etc/apt/sources.list && \
     apt-key adv --keyserver keyserver.ubuntu.com --recv-keys 60C317803A41BA51845E371A1E9377A2BA9EF27F && \
     apt-get update
 
-RUN DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
-      gcc-11 \
-      g++-11 \
-      build-essential \
-      ca-certificates \
-      ccache \
-      curl \
-      git \
-      cmake \
-      libjpeg-dev \
-      libpng-dev \
-      ninja-build \
-    && rm -rf /var/lib/apt/lists/*
+RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
+        gcc-11 \
+        g++-11 \
+        build-essential \
+        ca-certificates \
+        ccache \
+        cmake \
+        curl \
+        git \
+        ninja-build \
+        libjpeg-dev \
+        libpng-dev && \
+    rm -rf /var/lib/apt/lists/*
 RUN /usr/sbin/update-ccache-symlinks
 RUN mkdir /opt/ccache && ccache --set-config=cache_dir=/opt/ccache
 ENV PATH /opt/conda/bin:$PATH
@@ -84,38 +87,43 @@ ENV PATH /opt/conda/bin:$PATH
 RUN update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-11 60 --slave /usr/bin/g++ g++ /usr/bin/g++-11
 
 # Install conda
-# translating Docker's TARGETPLATFORM into mamba arches
+FROM dev-base as pytorch-build
+# Automatically set by buildx
+ARG TARGETPLATFORM
+# translating Docker's TARGETPLATFORM into miniconda arches
 RUN case ${TARGETPLATFORM} in \
-         "linux/arm64")  MAMBA_ARCH=aarch64  ;; \
-         *)              MAMBA_ARCH=x86_64   ;; \
+         "linux/arm64")  MINICONDA_ARCH=aarch64  ;; \
+         *)              MINICONDA_ARCH=x86_64   ;; \
     esac && \
-    curl -fsSL -v -o ~/mambaforge.sh -O  "https://github.com/conda-forge/miniforge/releases/download/${MAMBA_VERSION}/Mambaforge-${MAMBA_VERSION}-Linux-${MAMBA_ARCH}.sh"
-RUN chmod +x ~/mambaforge.sh && \
-    bash ~/mambaforge.sh -b -p /opt/conda && \
-    rm ~/mambaforge.sh
+    curl -fsSL -v -o ~/miniconda.sh -O  "https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-${MINICONDA_ARCH}.sh"
+
+# Manually invoke bash on miniconda script per https://github.com/conda/conda/issues/10431
+RUN chmod +x ~/miniconda.sh && \
+    bash ~/miniconda.sh -b -p /opt/conda && \
+    rm ~/miniconda.sh
 
 RUN git clone --recursive https://github.com/pytorch/pytorch && \
     cd pytorch && \
-    git checkout v${PYTORCH_VERSION} && \
-    git submodule update --init --recursive
+    git checkout v${PYTORCH_VERSION}
 
 WORKDIR /pytorch
 
+RUN git submodule update --init --recursive
+
 # Write the Pytorch version into the version.txt file because it isn't always the same as the tag we checked out
 RUN echo $PYTORCH_VERSION > version.txt
 
-RUN /opt/conda/bin/conda install -y python=${PYTHON_VERSION} cmake ninja conda-build pyyaml numpy ipython && \
-    /opt/conda/bin/python -mpip install -r requirements.txt
-
 # Install things for building PyTorch
 RUN /opt/conda/bin/conda install -y mkl mkl-include cudnn && \
-    /opt/conda/bin/conda install -c pytorch magma-cuda118
+    /opt/conda/bin/conda install libgcc && \
+    /opt/conda/bin/conda install -c conda-forge libstdcxx-ng=12 && \
+    /opt/conda/bin/conda install -c pytorch magma-cuda118 && \
+    /opt/conda/bin/conda install -c "nvidia/label/cuda-11.8.0"  cuda==11.8
 
 # https://github.com/cresset-template/cresset/blob/37c7b5df7236d3b9d96c4908efe5af8bc90066e3/reqs/train-conda-build.requirements.txt
 # TODO: remove what we don't need
 RUN /opt/conda/bin/conda install -y \
       jemalloc \
-      astunparse \
       ccache \
       cmake \
       expecttest \
@@ -124,7 +132,6 @@ RUN /opt/conda/bin/conda install -y \
       git \
       hypothesis \
       jinja2 \
-      libjpeg-turbo \
       libpng \
       networkx \
       ninja \
@@ -134,7 +141,6 @@ RUN /opt/conda/bin/conda install -y \
       requests \
       setuptools \
       sympy \
-      types-dataclasses \
       typing-extensions
 
 RUN /opt/conda/bin/conda clean -ya
@@ -151,13 +157,21 @@ ENV LD_PRELOAD=/opt/conda/lib/libiomp5.so:${LD_PRELOAD}
 ENV LD_PRELOAD=/opt/conda/lib/libjemalloc.so:${LD_PRELOAD}
 ENV MALLOC_CONF="background_thread:true,metadata_thp:auto,dirty_decay_ms:30000,muzzy_decay_ms:30000"
 
+RUN /opt/conda/bin/conda install -y python=${PYTHON_VERSION} cmake conda-build pyyaml numpy ipython && \
+    /opt/conda/bin/python -mpip install -r requirements.txt
+RUN /opt/conda/bin/conda clean -ya
+
+RUN cat version.txt
+
+RUN make triton
+
 # Install PyTorch without AVX2
 # https://github.com/cresset-template/cresset/blob/37c7b5df7236d3b9d96c4908efe5af8bc90066e3/docker-compose.yaml#L124
 # print(torch.__config__.show().split("\n"), sep="\n")
 RUN --mount=type=cache,target=/opt/ccache \
     python setup.py clean && \
-
-    BLAS_INFO=mklBUILD_TYPE=Release \
+    BLAS_INFO=mkl \
+    BUILD_TYPE=Release \
     CXX_FLAGS="-D_GLIBCXX_USE_CXX11_ABI=0 -fabi-version=11 -Wno-deprecated -fvisibility-inlines-hidden -DUSE_PTHREADPOOL -DNDEBUG -DUSE_KINETO -DLIBKINETO_NOROCTRACER -DUSE_FBGEMM -DUSE_QNNPACK -DUSE_PYTORCH_QNNPACK -DUSE_XNNPACK -DSYMBOLICATE_MOBILE_DEBUG_HANDLE -O2 -fPIC -Wall -Wextra -Werror=return-type -Werror=non-virtual-dtor -Werror=bool-operation -Wnarrowing -Wno-missing-field-initializers -Wno-type-limits -Wno-array-bounds -Wno-unknown-pragmas -Wunused-local-typedefs -Wno-unused-parameter -Wno-unused-function -Wno-unused-result -Wno-strict-overflow -Wno-strict-aliasing -Wno-error=deprecated-declarations -Wno-stringop-overflow -Wno-psabi -Wno-error=pedantic -Wno-error=redundant-decls -Wno-error=old-style-cast -fdiagnostics-color=always -faligned-new -Wno-unused-but-set-variable -Wno-maybe-uninitialized -fno-math-errno -fno-trapping-math -Werror=format -Werror=cast-function-type -Wno-stringop-overflow" \
     LAPACK_INFO=mkl \
     PERF_WITH_AVX=1 \
@@ -177,35 +191,24 @@ RUN --mount=type=cache,target=/opt/ccache \
     USE_NNPACK=ON \
     USE_OPENMP=ON \
     USE_ROCM=OFF \
-    BUILD_TEST=0 \
-    TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" TORCH_NVCC_FLAGS="-Xfatbin -compress-all" \
-    CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" \
-    CMAKE_ARGS='-DDISABLE_AVX2:BOOL=TRUE -DCXX_AVX2_FOUND:BOOL=FALSE -DC_AVX2_FOUND:BOOL=FALSE -DDISABLE_AVX512F:BOOL=TRUE' \
-    python setup.py install && \
-    cd .. && \
-    rm -rf pytorch
+     BUILD_TEST=0 \
+     CMAKE_ARGS='-DDISABLE_AVX2:BOOL=TRUE -DCXX_AVX2_FOUND:BOOL=FALSE -DC_AVX2_FOUND:BOOL=FALSE -DDISABLE_AVX512F:BOOL=TRUE' \
+     TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" TORCH_NVCC_FLAGS="-Xfatbin -compress-all" \
+     CMAKE_PREFIX_PATH="$(dirname $(which conda))/../" \
+     python setup.py install && \
+     cd .. && \
+     rm -rf pytorch
+
 
 # Make sure we built everything properly. Build will fail if CUDA isn't available.
-RUN python -c "import torch; exit(1 if not torch.version.cuda else 0)"
+# RUN python -c "import torch; exit(1 if not torch.version.cuda else 0)"
 
 RUN nm -D /opt/conda/lib/python3.9/site-packages/torch/lib/libtorch.so
 
-# ==============================================================================
-# Set up the kernel-builder
-
-FROM pytorch-install as kernel-builder
-
-RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
-      ninja-build \
-    && rm -rf /var/lib/apt/lists/*
-
-RUN /opt/conda/bin/conda install -c "nvidia/label/cuda-11.8.0"  cuda==11.8 && \
-    /opt/conda/bin/conda clean -ya
-
 # ==============================================================================
 # Build Flash Attention CUDA kernels
 
-FROM kernel-builder as flash-att-builder
+FROM pytorch-build as flash-att-builder
 WORKDIR /usr/src
 COPY server/Makefile-flash-att Makefile
 
@@ -215,7 +218,7 @@ RUN MAX_JOBS=5 make build-flash-attention
 # ==============================================================================
 # Build Flash Attention v2 CUDA kernels
 
-FROM kernel-builder as flash-att-v2-builder
+FROM pytorch-build as flash-att-v2-builder
 WORKDIR /usr/src
 COPY server/Makefile-flash-att-v2 Makefile
 
@@ -225,7 +228,7 @@ RUN MAX_JOBS=10 make build-flash-attention-v2
 # ==============================================================================
 # Build Transformers exllama kernels
 
-FROM kernel-builder as exllama-kernels-builder
+FROM pytorch-build as exllama-kernels-builder
 WORKDIR /usr/src
 COPY server/exllama_kernels/ .
 
@@ -235,7 +238,7 @@ RUN TORCH_CUDA_ARCH_LIST="8.0;8.6+PTX" python setup.py build
 # ==============================================================================
 # Build Transformers CUDA kernels
 
-FROM kernel-builder as custom-kernels-builder
+FROM pytorch-build as custom-kernels-builder
 WORKDIR /usr/src
 COPY server/custom_kernels/ .
 
@@ -245,7 +248,7 @@ RUN MAX_JOBS=5 python setup.py build
 # ==============================================================================
 # Build vllm CUDA kernels
 
-FROM kernel-builder as vllm-builder
+FROM pytorch-build as vllm-builder
 WORKDIR /usr/src
 COPY server/Makefile-vllm Makefile
 
@@ -256,7 +259,7 @@ RUN MAX_JOBS=5 make build-vllm
 
 # Text Generation Inference base image
 # nvidia/cuda:11.8.0-base-ubuntu20.04
-FROM nvidia/cuda:11.8.0-devel-ubuntu20.04 as base
+FROM nvidia/cuda:11.8.0-base-ubuntu20.04 as base
 
 # Conda env
 ENV PATH=/opt/conda/bin:$PATH \
@@ -277,7 +280,7 @@ RUN apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y --no-ins
     && rm -rf /var/lib/apt/lists/*
 
 # Copy conda with PyTorch installed
-COPY --from=pytorch-install /opt/conda /opt/conda
+COPY --from=pytorch-build /opt/conda /opt/conda
 
 # Copy build artifacts from flash attention builder
 COPY --from=flash-att-builder /usr/src/flash-attention/build/lib.linux-x86_64-cpython-39 /opt/conda/lib/python3.9/site-packages
@@ -334,8 +337,8 @@ RUN python3 -c "import torch; import text_generation_server.utils.gptq.exllama"
 
 
 # Make sure our special dependencies were compiled and copied correctly
-RUN python -c "import torch; exit(1 if not torch.version.cuda else 0)"
-RUN python -c "import torch; torch.cuda.is_available()"
+# RUN python -c "import torch; exit(1 if not torch.version.cuda else 0)"
+# RUN python -c "import torch; torch.cuda.is_available()"
 RUN python -c "import torch; import flash_attn_2_cuda"
 RUN python -c "import torch; import flash_attn_cuda"
 RUN python -c "import torch; import vllm_cache_ops"