From 0b75b0351dd97c4c12f0c11c01ed6fc9a89b1f93 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Wed, 24 Jun 2026 15:10:18 -0700 Subject: [PATCH 1/6] docs: align release notes with Helm OCR v2 default After #2239, Nemotron OCR v2 is the Helm NIM default; update the 26.05 release notes so they no longer state v1 is the chart default. --- docs/docs/extraction/releasenotes.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/docs/extraction/releasenotes.md b/docs/docs/extraction/releasenotes.md index dd1a2b483..bc3423fdb 100644 --- a/docs/docs/extraction/releasenotes.md +++ b/docs/docs/extraction/releasenotes.md @@ -36,7 +36,7 @@ Highlights for the 26.05 release include: ### Models, OCR, and captioning -- Nemotron OCR v2 is the default OCR engine for HuggingFace, with CLI language selectors and unified OCR actors. For Helm NIM deployments, Nemotron OCR v1 is the default. +- Nemotron OCR v2 is the default OCR engine for HuggingFace, Helm NIM, and hosted remote deployments, with CLI language selectors and unified OCR actors. - Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference via vLLM where configured) - VLM image captioning via vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 - vLLM-backed text and vision-language embedders, multimodal VL reranker, and torch 2.11 for local GPU installs From faeced32ef2ca0ea70859735647f5fc8c39fa341 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Wed, 24 Jun 2026 15:10:27 -0700 Subject: [PATCH 2/6] docs: NRL OCR section polish on multimodal-extraction Link Helm defaults to the chart README instead of pinning nvcr.io inline, and use refer-to CTAs per NRL style rules. --- docs/docs/extraction/multimodal-extraction.md | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/docs/docs/extraction/multimodal-extraction.md b/docs/docs/extraction/multimodal-extraction.md index f44aee36c..153597923 100644 --- a/docs/docs/extraction/multimodal-extraction.md +++ b/docs/docs/extraction/multimodal-extraction.md @@ -60,7 +60,7 @@ Charts and infographic regions are classified with other page layout elements (t For chart and infographic detection and modality-specific retrieval, use the default **pdfium** layout path (page-elements detection and OCR), not `extract_method="nemotron_parse"`. -For how chart-labeled PDF regions interact with captioning, refer to [Image captioning](#image-captioning). +Chart-labeled PDF regions are **not** routed through the Omni caption stage; they remain on the layout-and-OCR path. For scope and validation guidance, refer to [Image captioning](#image-captioning). For natural-language infographic descriptions, optionally enable [image captioning](#image-captioning) and set `caption_infographics=True` when you need VLM captions on infographic regions. @@ -74,7 +74,13 @@ For natural-language infographic descriptions, optionally enable [image captioni Scanned PDFs and image-only pages rely on OCR and hybrid paths that combine native text extraction with OCR when needed. For extract methods such as `ocr` and `pdfium_hybrid`, refer to the [Python API reference](nemo-retriever-api-reference.md). -When you run extraction locally with Hugging Face weights, the default OCR engine is **Nemotron OCR v2**, which operates in **multilingual** mode by default. For CLI flags and API parameters, refer to [Nemotron OCR v2 — language mode](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/docs/cli/README.md#nemotron-ocr-v2-language-mode). For Kubernetes deployment, refer to [OCR NIM configuration](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/helm/README.md#ocr-nim-configuration) in the Helm chart README. +OCR artifacts depend on how you deploy. + +**Helm / NIM:** The production chart deploys **Nemotron OCR v2**. For image defaults and upgrade notes, refer to [OCR NIM configuration](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/helm/README.md#ocr-nim-configuration) in the Helm chart README. + +**Local Hugging Face inference:** The default engine is **Nemotron OCR v2**, which operates in **multilingual** mode by default. For CLI flags and API parameters, refer to [Nemotron OCR v2 — language mode](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/docs/cli/README.md#nemotron-ocr-v2-language-mode). + +For Kubernetes defaults and the Helm-vs-local split, refer to [OCR artifacts (Helm vs local Hugging Face)](prerequisites-support-matrix.md#nemotron-ocr-v2-language-mode) in the support matrix. **Related** From 9dfaf05a73813a38579c12ca98a07213ac9a11c7 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Wed, 24 Jun 2026 15:13:40 -0700 Subject: [PATCH 3/6] docs: apply audit, polish, and style fixes to OCR sections Split multimodal-extraction OCR deployment guidance for scannability, add hosted-remote default note, and fix Hugging Face spelling in release notes. --- docs/docs/extraction/releasenotes.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/docs/docs/extraction/releasenotes.md b/docs/docs/extraction/releasenotes.md index bc3423fdb..322cf84da 100644 --- a/docs/docs/extraction/releasenotes.md +++ b/docs/docs/extraction/releasenotes.md @@ -36,7 +36,7 @@ Highlights for the 26.05 release include: ### Models, OCR, and captioning -- Nemotron OCR v2 is the default OCR engine for HuggingFace, Helm NIM, and hosted remote deployments, with CLI language selectors and unified OCR actors. +- Nemotron OCR v2 is the default OCR engine for Hugging Face and Helm NIM deployments, with CLI language selectors and unified OCR actors. - Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference via vLLM where configured) - VLM image captioning via vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 - vLLM-backed text and vision-language embedders, multimodal VL reranker, and torch 2.11 for local GPU installs From ef10bfc69ea36df8d9b4fd31cb51b4a26f193179 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Wed, 24 Jun 2026 15:14:32 -0700 Subject: [PATCH 4/6] docs: style fixes in release notes (via, spacing) Replace via with with/through per NVIDIA style guide and fix double space before reference answer generation. --- docs/docs/extraction/releasenotes.md | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/docs/docs/extraction/releasenotes.md b/docs/docs/extraction/releasenotes.md index 322cf84da..5932d8c03 100644 --- a/docs/docs/extraction/releasenotes.md +++ b/docs/docs/extraction/releasenotes.md @@ -37,8 +37,8 @@ Highlights for the 26.05 release include: ### Models, OCR, and captioning - Nemotron OCR v2 is the default OCR engine for Hugging Face and Helm NIM deployments, with CLI language selectors and unified OCR actors. -- Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference via vLLM where configured) -- VLM image captioning via vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 +- Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference with vLLM where configured) +- VLM image captioning with vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 - vLLM-backed text and vision-language embedders, multimodal VL reranker, and torch 2.11 for local GPU installs ### Multimodal extraction @@ -48,7 +48,7 @@ Highlights for the 26.05 release include: ### Retrieval and RAG -- Live RAG SDK with `Retriever.retrieve()`, reference answer generation `Retriever.answer()`, and optional batch operator graphs via LiteLLM (`[llm]` extra) +- Live RAG SDK with `Retriever.retrieve()`, reference answer generation `Retriever.answer()`, and optional batch operator graphs through LiteLLM (`[llm]` extra) ### Vector database From d2e71030677551cb05473640802c2ca2a8397258 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Thu, 2 Jul 2026 10:33:45 -0700 Subject: [PATCH 5/6] docs: revert releasenotes OCR edit; 26.5.0 shipped Helm v1 #2239 merged after 26.5.0, so the 26.05 release notes must keep the historical HF v2 / Helm v1 split. Helm v2 current-state docs stay in multimodal-extraction.md only. --- docs/docs/extraction/releasenotes.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/docs/docs/extraction/releasenotes.md b/docs/docs/extraction/releasenotes.md index 5932d8c03..dd1a2b483 100644 --- a/docs/docs/extraction/releasenotes.md +++ b/docs/docs/extraction/releasenotes.md @@ -36,9 +36,9 @@ Highlights for the 26.05 release include: ### Models, OCR, and captioning -- Nemotron OCR v2 is the default OCR engine for Hugging Face and Helm NIM deployments, with CLI language selectors and unified OCR actors. -- Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference with vLLM where configured) -- VLM image captioning with vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 +- Nemotron OCR v2 is the default OCR engine for HuggingFace, with CLI language selectors and unified OCR actors. For Helm NIM deployments, Nemotron OCR v1 is the default. +- Nemotron Parse is available as an alternate PDF extraction method (v1.2 HTTP interface; optional Helm NIM; local inference via vLLM where configured) +- VLM image captioning via vLLM (including Omni caption model profiles) addresses the capability deferred in 26.03 - vLLM-backed text and vision-language embedders, multimodal VL reranker, and torch 2.11 for local GPU installs ### Multimodal extraction @@ -48,7 +48,7 @@ Highlights for the 26.05 release include: ### Retrieval and RAG -- Live RAG SDK with `Retriever.retrieve()`, reference answer generation `Retriever.answer()`, and optional batch operator graphs through LiteLLM (`[llm]` extra) +- Live RAG SDK with `Retriever.retrieve()`, reference answer generation `Retriever.answer()`, and optional batch operator graphs via LiteLLM (`[llm]` extra) ### Vector database From c09805da499b260a42f238ca8f5c36202f9d3723 Mon Sep 17 00:00:00 2001 From: Kurt Heiss Date: Thu, 2 Jul 2026 10:41:12 -0700 Subject: [PATCH 6/6] docs: add NVCF hosted OCR v2 default to multimodal-extraction Documents hosted CPU actor defaults after #2295 lands. Merge #2295 before this PR per eng review. --- docs/docs/extraction/multimodal-extraction.md | 2 ++ 1 file changed, 2 insertions(+) diff --git a/docs/docs/extraction/multimodal-extraction.md b/docs/docs/extraction/multimodal-extraction.md index 153597923..dbe879b97 100644 --- a/docs/docs/extraction/multimodal-extraction.md +++ b/docs/docs/extraction/multimodal-extraction.md @@ -78,6 +78,8 @@ OCR artifacts depend on how you deploy. **Helm / NIM:** The production chart deploys **Nemotron OCR v2**. For image defaults and upgrade notes, refer to [OCR NIM configuration](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/helm/README.md#ocr-nim-configuration) in the Helm chart README. +**Hosted remote inference (NVCF):** CPU OCR actors default to the hosted **Nemotron OCR v2** endpoint on build.nvidia.com when you do not set `ocr_invoke_url`. + **Local Hugging Face inference:** The default engine is **Nemotron OCR v2**, which operates in **multilingual** mode by default. For CLI flags and API parameters, refer to [Nemotron OCR v2 — language mode](https://github.com/NVIDIA/NeMo-Retriever/blob/main/nemo_retriever/docs/cli/README.md#nemotron-ocr-v2-language-mode). For Kubernetes defaults and the Helm-vs-local split, refer to [OCR artifacts (Helm vs local Hugging Face)](prerequisites-support-matrix.md#nemotron-ocr-v2-language-mode) in the support matrix.