Add installation support for deepspeed

Signed-off-by: Daniel J Walsh <[email protected]>

training/Makefile

@@ -8,12 +8,16 @@ help:
 	@echo "  - make instruct-nvidia"
 	@echo "  - make instruct-vllm"
 	@echo
+	@echo "Once instruct images created, create advanced training containers, deepspeed only for nvidia"
+	@echo
+	@echo "  - make deepspeed"
+	@echo "  - make vllm"
+	@echo
 	@echo "Once instruct images are created, creat bootc container images"
 	@echo
-	@echo "  - make bootc-amd"
-	@echo "  - make bootc-intel"
-	@echo "  - make bootc-nvidia"
-	@echo "  - make bootc-vllm"
+	@echo "  - make amd"
+	@echo "  - make intel"
+	@echo "  - make nvidia"
 	@echo
 	@echo "If these images are going to be used on a cloud, you might want to add cloud-init."
 	@echo
@@ -36,6 +40,14 @@ instruct-nvidia:
 .PHONY:
 instruct: instruct-amd instruct-nvidia
 
+.PHONY: deepspeed
+deepspeed:
+	make -C deepspeed/ image
+
+.PHONY: vllm
+vllm:
+	make -C vllm/ image
+
 #
 # Create bootc container images prepared for AI
 #
@@ -46,8 +58,6 @@ intel:
 	make -C intel-bootc/ bootc
 nvidia:
 	make -C nvidia-bootc/ dtk bootc
-vllm:
-	make -C vllm/ image
 
 #
 # Make Bootc container images preinstalled with cloud-init

training/amd-bootc/Makefile

@@ -4,6 +4,8 @@ IMAGE_NAME ?= $(VENDOR)-bootc
 
 include ../common/Makefile.common
 
+default: bootc
+
 .PHONY: bootc
 bootc: prepare-files growfs
 	"${CONTAINER_TOOL}" build \

training/deepspeed/Containerfile

@@ -4,11 +4,7 @@ FROM nvcr.io/nvidia/cuda:12.1.1-cudnn8-devel-ubi9
 RUN dnf install -y python python-devel git
 RUN python -m ensurepip --upgrade
 RUN pip3 install torch==2.1.2 --index-url https://download.pytorch.org/whl/cu121
-RUN pip3 install packaging wheel
-RUN pip3 install flash-attn==2.5.7
-RUN pip3 install deepspeed==0.14.2
-RUN pip3 install transformers==4.40.1
-RUN pip3 install ipdb jupyterlab gpustat matplotlib hydra-core datasets rich numba
+RUN pip3 install packaging wheel flash-attn==2.5.7 deepspeed==0.14.2 transformers==4.40.1 ipdb jupyterlab gpustat matplotlib hydra-core datasets rich numba
 RUN git clone https://github.com/instructlab/training.git
 RUN mkdir -p /ilab-data/training_output
 

training/deepspeed/Makefile

@@ -1,15 +1,14 @@
-REGISTRY ?= quay.io
-REGISTRY_ORG ?= ai-lab
 IMAGE_NAME ?= deepspeed-trainer
-IMAGE_TAG ?= latest
-
 CONTAINER_TOOL ?= podman
 
 default: image
 
 .PHONY: image
 image:
+	@mkdir -p ../build
 	"${CONTAINER_TOOL}" build \
 			$(ARCH:%=--platform linux/%) \
 			--file Containerfile \
-			--tag "${REGISTRY}/${REGISTRY_ORG}/${IMAGE_NAME}:${IMAGE_TAG}" \
+			--layers=false \
+			--squash-all \
+			--tag oci:../build/deepspeed-trainer .

training/instructlab/Makefile

@@ -6,7 +6,7 @@ IMAGE_TAG ?= latest
 
 .PHONY: instructlab
 instructlab:
-	mkdir -p ../build
+	@mkdir -p ../build
 	git clone https://github.com/instructlab/instructlab.git 2> /dev/null || true
 	(cd instructlab; git pull origin main)
 

training/intel-bootc/Makefile

@@ -2,6 +2,8 @@ IMAGE_NAME ?= intel-bootc
 
 include ../common/Makefile.common
 
+default: bootc
+
 .PHONY: bootc
 bootc: growfs
 	"${CONTAINER_TOOL}" build \

training/nvidia-bootc/Makefile

@@ -10,6 +10,8 @@ ENABLE_RT ?=
 
 include ../common/Makefile.common
 
+default: bootc
+
 .PHONY: dtk
 dtk:
 	"${CONTAINER_TOOL}" build \

training/vllm/Makefile

@@ -4,6 +4,7 @@ default: image
 
 .PHONY: image
 image:
+	@mkdir -p ../build
 	"${CONTAINER_TOOL}" build \
 			$(ARCH:%=--platform linux/%) \
 			--file Containerfile \