Single source of truth for versions of NIMs and dependencies between …

docker-compose.yaml

@@ -9,7 +9,6 @@ services:
       - "6379:6379"
 
   page-elements:
-    image: ${YOLOX_IMAGE:-nvcr.io/nim/nvidia/nemoretriever-page-elements-v3}:${YOLOX_TAG:-1.7.0}
     shm_size: 16gb
     ports:
       - "8000:8000"
@@ -46,7 +45,6 @@ services:
     runtime: nvidia
 
   graphic-elements:
-    image: ${YOLOX_GRAPHIC_ELEMENTS_IMAGE:-nvcr.io/nim/nvidia/nemoretriever-graphic-elements-v1}:${YOLOX_GRAPHIC_ELEMENTS_TAG:-1.6.0}
     shm_size: 16gb
     ports:
       - "8003:8000"
@@ -71,7 +69,6 @@ services:
     runtime: nvidia
 
   table-structure:
-    image: ${YOLOX_TABLE_STRUCTURE_IMAGE:-nvcr.io/nim/nvidia/nemoretriever-table-structure-v1}:${YOLOX_TABLE_STRUCTURE_TAG:-1.6.0}
     shm_size: 16gb
     ports:
       - "8006:8000"
@@ -96,7 +93,6 @@ services:
     runtime: nvidia
 
   ocr:
-    image: ${OCR_IMAGE:-nvcr.io/nim/nvidia/nemoretriever-ocr-v1}:${OCR_TAG:-1.2.1}
     shm_size: 16gb
     ports:
       - "8009:8000"
@@ -121,7 +117,6 @@ services:
 
   embedding:
     # NIM ON
-    image: ${EMBEDDING_IMAGE:-nvcr.io/nim/nvidia/llama-3.2-nv-embedqa-1b-v2}:${EMBEDDING_TAG:-1.10.1}
     shm_size: 16gb
     ports:
       - "8012:8000"
@@ -146,7 +141,6 @@ services:
 
   reranker:
     # NIM ON
-    image: ${RERANKER_IMAGE:-nvcr.io/nim/nvidia/llama-3.2-nv-rerankqa-1b-v2}:${RERANKER_TAG:-1.8.0}
     shm_size: 16gb
     ports:
       - "8020:8000"
@@ -167,7 +161,6 @@ services:
       - reranker
 
   nemotron-parse:
-    image: ${NEMOTRON_PARSE_IMAGE:-nvcr.io/nim/nvidia/nemotron-parse}:${NEMOTRON_PARSE_TAG:-1.5.0}
     shm_size: 16gb
     ports:
       - "8015:8000"
@@ -190,7 +183,6 @@ services:
       - nemotron-parse
 
   vlm:
-    image: ${VLM_IMAGE:-nvcr.io/nim/nvidia/nemotron-nano-12b-v2-vl}:${VLM_TAG:-1.5.0}
     shm_size: 16gb
     ports:
       - "8018:8000"
@@ -215,7 +207,6 @@ services:
       - vlm
 
   audio:
-    image: ${AUDIO_IMAGE:-nvcr.io/nim/nvidia/parakeet-1-1b-ctc-en-us}:${AUDIO_TAG:-1.4.0}
     shm_size: 2gb
     ports:
       - "8021:50051"  # grpc

docs/docs/extraction/audio.md

@@ -65,7 +65,7 @@ Use the following procedure to run the NIM locally.
 3. Start the nv-ingest services with the `audio` profile. This profile includes the necessary components for audio processing. Use the following command. The `--profile audio` flag ensures that speech-specific services are launched. For more information, refer to [Profile Information](quickstart-guide.md#profile-information).
 
     ```shell
-    docker compose --profile retrieval --profile audio up
+    docker compose -f docker-compose.yaml -f release.yaml --profile retrieval --profile audio up
     ```
 
 4. After the services are running, you can interact with nv-ingest by using Python.

docs/docs/extraction/nemoretriever-parse.md

@@ -37,7 +37,7 @@ Use the following procedure to run the NIM locally.
     - The --profile nemotron-parse flag ensures that vision-language retrieval services are launched.  For more information, refer to [Profile Information](quickstart-guide.md#profile-information).
 
     ```shell
-    docker compose --profile nemotron-parse up
+    docker compose -f docker-compose.yaml -f release.yaml --profile nemotron-parse up
     ```
 
 2. After the services are running, you can interact with nv-ingest by using Python.

docs/docs/extraction/nv-ingest-python-api.md

@@ -510,7 +510,7 @@ ingestor = ingestor.store(
 ```bash
 # Set DATASET_ROOT before starting services
 export DATASET_ROOT=/raid/my-project/nv-ingest-data
-docker compose up -d
+docker compose -f docker-compose.yaml -f release.yaml up -d
 ```
 
 ```python

docs/docs/extraction/quickstart-guide.md

@@ -48,9 +48,14 @@ If you prefer, you can run on Kubernetes by using [our Helm chart](https://githu
 
     `sudo nvidia-ctk runtime configure --runtime=docker --set-as-default`
 
-6. Start core services. This example uses the retrieval profile.  For more information about other profiles, see [Profile Information](#profile-information).
+6. Start core services. This example uses the retrieval profile. For more information about other profiles, see [Profile Information](#profile-information).
 
-    `docker compose --profile retrieval up`
+    ```shell
+    docker compose \
+      -f docker-compose.yaml \
+      -f release.yaml \
+      --profile retrieval up
+    ```
 
     !!! tip
 
@@ -61,12 +66,13 @@ If you prefer, you can run on Kubernetes by using [our Helm chart](https://githu
         The default configuration may not fit on a single GPU for some hardware targets. If you are running on any of the following GPUs, use a `docker compose` override file to reduce VRAM usage:
         - A100-SXM4-40GB
         - A10G
-        Override files typically lower per-service memory allocation, batch sizes, or concurrency. This trades peak throughput for making the full pipeline runnable on the available GPU. To use an override file, include it in your `docker compose up` command by using a second `-f` flag after the base `docker-compose.yaml` file. The settings in the second file override the values that are set in the first file.
+        Override files typically lower per-service memory allocation, batch sizes, or concurrency. This trades peak throughput for making the full pipeline runnable on the available GPU. To use an override file, include it in your `docker compose up` command by adding an additional `-f` flag after `release.yaml`. The settings in later files override the values that are set in earlier files.
 
     The following example uses an override file that contains settings that are optimized for an NVIDIA A100 GPU with 40GB of VRAM.
     ```shell
     docker compose \
       -f docker-compose.yaml \
+      -f release.yaml \
       -f docker-compose.a100-40gb.yaml \
       --profile retrieval up
     ```

docs/docs/extraction/vlm-embed.md

@@ -17,18 +17,30 @@ The model supports images that contain text, tables, charts, and infographics.
 
 Use the following procedure to configure and run the multimodal embedding NIM locally.
 
-1. Set the embedding model in your .env file. This tells NeMo Retriever extraction to use the Llama 3.2 Multimodal model instead of the default text-only embedding model.
+1. Set the embedding model name in your `.env` file. This tells NeMo Retriever extraction to use the Llama 3.2 Multimodal model instead of the default text-only embedding model.
 
     ```
-    EMBEDDING_IMAGE=nvcr.io/nvidia/nemo-microservices/llama-3.2-nemoretriever-1b-vlm-embed-v1
-    EMBEDDING_TAG=1.7.0
     EMBEDDING_NIM_MODEL_NAME=nvidia/llama-3.2-nemoretriever-1b-vlm-embed-v1
     ```
 
-2. Start the NeMo Retriever extraction services. The multimodal embedding service is included by default.
+2. Create a Docker Compose override file to swap the embedding NIM image.
+
+    Create `docker-compose.vlm-embed.yaml` with the following contents:
+
+    ```yaml
+    services:
+      embedding:
+        image: nvcr.io/nvidia/nemo-microservices/llama-3.2-nemoretriever-1b-vlm-embed-v1:1.7.0
+    ```
+
+3. Start the NeMo Retriever extraction services.
 
     ```
-    docker compose --profile retrieval up
+    docker compose \
+      -f docker-compose.yaml \
+      -f release.yaml \
+      -f docker-compose.vlm-embed.yaml \
+      --profile retrieval up
     ```
 
 

helm/templates/_helpers.tpl

@@ -74,3 +74,34 @@ Create secret to access docker registry
   {{- define "nv-ingest.ngcApiSecret" }}
   {{- printf "%s" .Values.ngcApiSecret.password  }}
   {{- end }}
+
+{{/*
+Resolve NIM image ref/repo/tag from release.yaml.
+
+release.yaml is intended to be passed as a Helm values file and contains:
+  x-nv-ingest-release:
+    nim:
+      <nimOperatorKey>:
+        ref: "repo:tag"
+*/}}
+
+{{- define "nv-ingest.release.nimRef" -}}
+{{- $root := .root -}}
+{{- $key := .key -}}
+{{- $release := index $root.Values "x-nv-ingest-release" -}}
+{{- if and $release (hasKey $release "nim") (hasKey (index $release "nim") $key) (hasKey (index (index $release "nim") $key) "ref") -}}
+{{- index (index (index $release "nim") $key) "ref" -}}
+{{- else -}}
+{{- fail (printf "Missing required release.yaml value: x-nv-ingest-release.nim.%s.ref (pass -f release.yaml)" $key) -}}
+{{- end -}}
+{{- end -}}
+
+{{- define "nv-ingest.release.nimRepository" -}}
+{{- $ref := include "nv-ingest.release.nimRef" . -}}
+{{- regexReplaceAll ":([^:]*)$" $ref "" -}}
+{{- end -}}
+
+{{- define "nv-ingest.release.nimTag" -}}
+{{- $ref := include "nv-ingest.release.nimRef" . -}}
+{{- regexReplaceAll "^.*:" $ref "" -}}
+{{- end -}}

helm/templates/audio.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.audio.image.repository }}:{{ .Values.nimOperator.audio.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "audio") }}"
       pullSecret: "{{ index .Values.nimOperator.audio.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.audio.authSecret }}
   storage:
@@ -24,16 +24,16 @@ metadata:
   name: audio
 spec:
   image:
-    repository: {{ .Values.nimOperator.audio.image.repository }}
-    tag: {{ .Values.nimOperator.audio.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "audio") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "audio") }}
     pullPolicy: IfNotPresent
     pullSecrets:
 {{ toYaml .Values.nimOperator.audio.image.pullSecrets | nindent 6 }}
   authSecret: {{ .Values.nimOperator.audio.authSecret }}
   storage:
     nimCache:
       name: audio
-  replicas: {{ .Values.nimOperator.nemotron_parse.replicas }}
+  replicas: {{ .Values.nimOperator.audio.replicas }}
   resources:
 {{ toYaml .Values.nimOperator.audio.resources | nindent 4 }}
   tolerations:

helm/templates/llama-3.2-nv-embedqa-1b-v2.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.embedqa.image.repository }}:{{ .Values.nimOperator.embedqa.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "embedqa") }}"
       pullSecret: "{{ index .Values.nimOperator.embedqa.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.embedqa.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: llama-32-nv-embedqa-1b-v2
 spec:
   image:
-    repository: {{ .Values.nimOperator.embedqa.image.repository }}
-    tag: {{ .Values.nimOperator.embedqa.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "embedqa") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "embedqa") }}
     pullPolicy: {{ .Values.nimOperator.embedqa.image.pullPolicy }}
     pullSecrets:
 {{ toYaml .Values.nimOperator.embedqa.image.pullSecrets | nindent 6 }}

helm/templates/llama-3.2-nv-rerankqa-1b-v2.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.repository }}:{{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "llama_3_2_nv_rerankqa_1b_v2") }}"
       pullSecret: "{{ index .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: llama-32-nv-rerankqa-1b-v2
 spec:
   image:
-    repository: {{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.repository }}
-    tag: {{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "llama_3_2_nv_rerankqa_1b_v2") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "llama_3_2_nv_rerankqa_1b_v2") }}
     pullPolicy: {{ .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.pullPolicy }}
     pullSecrets:
 {{ toYaml .Values.nimOperator.llama_3_2_nv_rerankqa_1b_v2.image.pullSecrets | nindent 6 }}

helm/templates/nemoretriever-graphic-elements-v1.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.graphic_elements.image.repository }}:{{ .Values.nimOperator.graphic_elements.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "graphic_elements") }}"
       pullSecret: "{{ index .Values.nimOperator.graphic_elements.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.graphic_elements.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: nemoretriever-graphic-elements-v1
 spec:
   image:
-    repository: {{ .Values.nimOperator.graphic_elements.image.repository }}
-    tag: {{ .Values.nimOperator.graphic_elements.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "graphic_elements") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "graphic_elements") }}
     pullPolicy: {{ .Values.nimOperator.graphic_elements.image.pullPolicy }}
     pullSecrets:
 {{ toYaml .Values.nimOperator.graphic_elements.image.pullSecrets | nindent 6 }}

helm/templates/nemoretriever-ocr-v1.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.nemoretriever_ocr_v1.image.repository }}:{{ .Values.nimOperator.nemoretriever_ocr_v1.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "nemoretriever_ocr_v1") }}"
       pullSecret: "{{ index .Values.nimOperator.nemoretriever_ocr_v1.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.nemoretriever_ocr_v1.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: nemoretriever-ocr-v1
 spec:
   image:
-    repository: {{ .Values.nimOperator.nemoretriever_ocr_v1.image.repository }}
-    tag: {{ .Values.nimOperator.nemoretriever_ocr_v1.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "nemoretriever_ocr_v1") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "nemoretriever_ocr_v1") }}
     pullPolicy: {{ .Values.nimOperator.nemoretriever_ocr_v1.image.pullPolicy }}
     pullSecrets: {{ toYaml .Values.nimOperator.nemoretriever_ocr_v1.image.pullSecrets | nindent 6 }}
   authSecret: {{ .Values.nimOperator.nemoretriever_ocr_v1.authSecret }}

helm/templates/nemoretriever-page-elements-v3.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.page_elements.image.repository }}:{{ .Values.nimOperator.page_elements.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "page_elements") }}"
       pullSecret: "{{ index .Values.nimOperator.page_elements.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.page_elements.authSecret }}
   storage:
@@ -25,8 +25,8 @@ metadata:
   name: nemoretriever-page-elements-v3
 spec:
   image:
-    repository: {{ .Values.nimOperator.page_elements.image.repository }}
-    tag: {{ .Values.nimOperator.page_elements.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "page_elements") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "page_elements") }}
     pullPolicy: {{ .Values.nimOperator.page_elements.image.pullPolicy }}
     pullSecrets: {{ toYaml .Values.nimOperator.page_elements.image.pullSecrets | nindent 6 }}
   authSecret: {{ .Values.nimOperator.page_elements.authSecret }}

helm/templates/nemoretriever-table-structure-v1.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.table_structure.image.repository }}:{{ .Values.nimOperator.table_structure.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "table_structure") }}"
       pullSecret: "{{ index .Values.nimOperator.table_structure.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.table_structure.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: nemoretriever-table-structure-v1
 spec:
   image:
-    repository: {{ .Values.nimOperator.table_structure.image.repository }}
-    tag: {{ .Values.nimOperator.table_structure.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "table_structure") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "table_structure") }}
     pullPolicy: {{ .Values.nimOperator.table_structure.image.pullPolicy }}
     pullSecrets:
 {{ toYaml .Values.nimOperator.table_structure.image.pullSecrets | nindent 6 }}

helm/templates/nemotron-nano-12b-v2-vl.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.nemotron_nano_12b_v2_vl.image.repository }}:{{ .Values.nimOperator.nemotron_nano_12b_v2_vl.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "nemotron_nano_12b_v2_vl") }}"
       pullSecret: "{{ index .Values.nimOperator.nemotron_nano_12b_v2_vl.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.nemotron_nano_12b_v2_vl.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: nemotron-nano-12b-v2-vl
 spec:
   image:
-    repository: {{ .Values.nimOperator.nemotron_nano_12b_v2_vl.image.repository }}
-    tag: {{ .Values.nimOperator.nemotron_nano_12b_v2_vl.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "nemotron_nano_12b_v2_vl") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "nemotron_nano_12b_v2_vl") }}
     pullPolicy: IfNotPresent
     pullSecrets:
 {{ toYaml .Values.nimOperator.nemotron_nano_12b_v2_vl.image.pullSecrets | nindent 6 }}

helm/templates/nemotron-parse.yaml

@@ -8,7 +8,7 @@ metadata:
 spec:
   source:
     ngc:
-      modelPuller: "{{ .Values.nimOperator.nemotron_parse.image.repository }}:{{ .Values.nimOperator.nemotron_parse.image.tag }}"
+      modelPuller: "{{ include "nv-ingest.release.nimRef" (dict "root" . "key" "nemotron_parse") }}"
       pullSecret: "{{ index .Values.nimOperator.nemotron_parse.image.pullSecrets 0 }}"
       authSecret: {{ .Values.nimOperator.nemotron_parse.authSecret }}
   storage:
@@ -24,8 +24,8 @@ metadata:
   name: nemotron-parse
 spec:
   image:
-    repository: {{ .Values.nimOperator.nemotron_parse.image.repository }}
-    tag: {{ .Values.nimOperator.nemotron_parse.image.tag }}
+    repository: {{ include "nv-ingest.release.nimRepository" (dict "root" . "key" "nemotron_parse") }}
+    tag: {{ include "nv-ingest.release.nimTag" (dict "root" . "key" "nemotron_parse") }}
     pullPolicy: IfNotPresent
     pullSecrets:
 {{ toYaml .Values.nimOperator.nemotron_parse.image.pullSecrets | nindent 6 }}