From f085204c5e2f09a09fbfd167bea4940219871fbd Mon Sep 17 00:00:00 2001
From: David Corvoysier <david@huggingface.co>
Date: Tue, 11 Feb 2025 15:51:09 +0000
Subject: [PATCH] feat(neuron): add server standalone installation

---
 backends/neuron/Makefile              |  6 ++++--
 backends/neuron/server/Makefile       | 21 +++++++++++++++++++--
 backends/neuron/server/pyproject.toml |  3 ++-
 3 files changed, 25 insertions(+), 5 deletions(-)

diff --git a/backends/neuron/Makefile b/backends/neuron/Makefile
index f7a33393..6da77f79 100644
--- a/backends/neuron/Makefile
+++ b/backends/neuron/Makefile
@@ -16,13 +16,15 @@ mkfile_path := $(abspath $(lastword $(MAKEFILE_LIST)))
 mkfile_dir := $(dir $(mkfile_path))
 root_dir := "${mkfile_dir}/../.."
 
-.PHONY:	image
+.PHONY:	image install_server
 
 VERSION := $(shell gawk 'match($$0, /^version = "(.*)"/, a) {print a[1]}' ${root_dir}/Cargo.toml)
 
 image:
 	docker build --rm -f ${root_dir}/Dockerfile.neuron \
 	             --build-arg VERSION=$(VERSION) \
-				 --ulimit nofile=100000:100000 \
 				 -t text-generation-inference:$(VERSION)-neuron ${root_dir}
 	docker tag text-generation-inference:$(VERSION)-neuron text-generation-inference:latest-neuron
+
+install_server:
+	make -C ${mkfile_dir}/server install VERSION:=${VERSION}
diff --git a/backends/neuron/server/Makefile b/backends/neuron/server/Makefile
index 24d5e0d1..efe34bd0 100644
--- a/backends/neuron/server/Makefile
+++ b/backends/neuron/server/Makefile
@@ -1,15 +1,20 @@
 # Initialize base variables
+SHELL := /bin/bash
 pkg_name := text_generation_server
 BUILDDIR ?= $(CURDIR)/build
 VERSION ?= 0.0.1
 mkfile_path := $(abspath $(lastword $(MAKEFILE_LIST)))
 mkfile_dir := $(dir $(mkfile_path))
 pkg_dir := $(BUILDDIR)/$(pkg_name)
-pkg_dist := ${BUILDDIR}/dist/${pkg_name}-${VERSION}.tar.gz
+py_version := $(subst -,.,${VERSION})
+pkg_dist := ${BUILDDIR}/dist/${pkg_name}-$(py_version).tar.gz
 
 clean:
 	rm -rf $(BUILDDIR)/*
 
+${BUILDDIR}:
+	install -d $@
+
 # List static sources to be deployed in the package
 src_dir := $(mkfile_dir)/$(pkg_name)
 sources := $(wildcard $(src_dir)/*.py)
@@ -21,7 +26,15 @@ define COPY
 	cp -f $< $@
 endef
 
-$(BUILDDIR)/pyproject.toml: $(mkfile_dir)/pyproject.toml
+# We use a PHONY target to represent the VERSION
+.PHONY: VERSION
+
+VERSION: ${BUILDDIR}
+	# The trick is to compare the value of the variable with the content of a file in the build directory
+	@if [[ `cat ${BUILDDIR}/VERSION 2>&1` != '$(VERSION)' ]]; then echo -n $(VERSION) >${BUILDDIR}/VERSION; fi
+
+# Depending on the PHONY VERSION target makes sure the pyproject.toml is regenerated if the version changes
+$(BUILDDIR)/pyproject.toml: $(mkfile_dir)/pyproject.toml VERSION
 	mkdir -p $(BUILDDIR)
 	$(COPY)
 	sed -i -e 's/version = "VERSION"/version = \"${VERSION}\"/' $@
@@ -55,3 +68,7 @@ ${pkg_dist}: $(BUILDDIR)/pyproject.toml $(deployed_sources) $(generated_sources)
 	python -m build $(BUILDDIR)
 
 package: ${pkg_dist}
+
+install: ${pkg_dist}
+	python3 -m pip uninstall -y ${pkg_name}
+	python3 -m pip install ${pkg_dist}
diff --git a/backends/neuron/server/pyproject.toml b/backends/neuron/server/pyproject.toml
index f37dc91f..ed790fdb 100644
--- a/backends/neuron/server/pyproject.toml
+++ b/backends/neuron/server/pyproject.toml
@@ -15,7 +15,8 @@ dependencies = [
     'grpc-interceptor == 0.15.2',
     'typer == 0.6.1',
     'safetensors',
-    'loguru == 0.6.0'
+    'loguru == 0.6.0',
+    'optimum-neuron[neuronx] >= 0.0.28',
 ]
 
 [tool.setuptools]