NVIDIA-NeMo · mckornfield · Aug 18, 2026 · Aug 18, 2026 · Aug 21, 2026 · mikeknep
@@ -24,7 +24,7 @@ evaluation result before promotion.
 | ------------------- | ------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------- |
 | Model optimization  | An agent uses a single frontier model where a smaller model or route split may preserve quality at lower cost | Suggests a model swap or Switchyard random-routing virtual model                                            |
 | Skill optimization  | The agent uses skills and has an evaluation suite                                                             | Suggests running `nemo agents optimize-skills` to improve skill files and keep changes that pass evaluation |
-| Prompt optimization | The agent has an optimization config and baseline dataset                                                     | Suggests `nemo agents optimize run` for Fabric-backed tuning                                                |
+| Prompt optimization | The agent has an optimization config and baseline dataset                                                     | Suggests `nemo agents optimize submit` for Fabric-backed tuning                                                |
 | New model scan      | Difference between the current model list and the previous optimizer snapshot                                 | Suggests evaluating or auditing newly available models                                                      |
 
 Optimizer state is stored in the `nemo-agent-optimizer` fileset:
@@ -195,7 +195,7 @@ and keeps the change only when the evaluation result improves.
 <Tab title="CLI">
 
 ```bash
-nemo agents optimize-skills run --spec-file .agent-improver.yml
+nemo agents optimize-skills submit --spec-file .agent-improver.yml
 ```
 
 Set `open_pr: true` in the YAML when you want the loop to prepare a
@@ -230,6 +230,7 @@ What it does under the hood:
 <Tab title="Python SDK">
 
 ```python
+import os
 import yaml
 from pathlib import Path
 
@@ -241,6 +242,7 @@ NemoJobScheduler().submit_remote(
     OptimizeSkillsJob,
     spec,
     workspace="default",
+    base_url=os.environ.get("NMP_BASE_URL", "http://localhost:8080"),
 )
 ```
 
@@ -273,7 +275,7 @@ nemo files list nemo-agent-telemetry
 
 ## Run Prompt and Parameter Tuning
 
-The `nemo agents optimize run` command runs Fabric-backed numeric
+The `nemo agents optimize submit` command runs Fabric-backed numeric
 optimization through `agents.optimize` (implementation in
 `nemo-optimization`). Input must be a Fabric-native agent package
 (`schema_version: fabric.agent/v1alpha1`). The golden-path harness is
@@ -294,7 +296,7 @@ After `uv sync --package nemo-agents-plugin` (and activating `.venv`), invoke
 <Tab title="CLI">
 
 ```bash
-nemo agents optimize run \
+nemo agents optimize submit \
   --optimize-config "$(pwd)/plugins/nemo-optimization/examples/hermes-optimize/optimize-chatonly.yaml" \
   --workspace default
 ```
@@ -306,7 +308,7 @@ Ask your coding agent:
 
 > Run prompt tuning on my deployed agent against this optimization config.
 
-The `agents-optimize` skill suggests `nemo agents optimize run` when the
+The `agents-optimize` skill suggests `nemo agents optimize submit` when the
 agent has an optimization config and a baseline dataset. Verify it is
 installed:
 
@@ -317,7 +319,7 @@ nemo skills show agents-optimize
 What it does under the hood:
 
 - Confirms the agent has a Fabric-native optimization YAML.
-- Runs `nemo agents optimize run` (or `submit` for platform jobs).
+- Runs `nemo agents optimize submit` (or `submit` for platform jobs).
 - Compares results against the evaluation baseline and surfaces deltas
   for review.
 
@@ -349,7 +351,7 @@ result = NemoJobScheduler().submit_remote(
         "workspace": WORKSPACE,
     },
     workspace=WORKSPACE,
-    sdk=client,
+    base_url=os.environ.get("NMP_BASE_URL", "http://localhost:8080"),
 )
 print(result)
 ```
@@ -376,7 +378,7 @@ export PHISHING_MCP_BIN="$PHISHING_AGENT_ROOT/.venv/bin/email-phishing-analyzer-
 
 # These environment variables are templated into optimize-mcp.yaml.
 
-nemo agents optimize run \
+nemo agents optimize submit \
   --optimize-config "$(pwd)/plugins/nemo-optimization/examples/hermes-optimize/optimize-mcp.yaml" \
   --workspace default
 ```
@@ -421,7 +423,7 @@ result = NemoJobScheduler().submit_remote(
         "workspace": WORKSPACE,
     },
     workspace=WORKSPACE,
-    sdk=client,
+    base_url=os.environ.get("NMP_BASE_URL", "http://localhost:8080"),
 )
 print(result)
 ```

@@ -14,9 +14,7 @@ This reference covers the `nemo anonymizer` commands exposed by the Anonymizer p
 | Command                          | Source                          | Description                                                  |
 |----------------------------------|---------------------------------|--------------------------------------------------------------|
 | `nemo anonymizer validate`       | Manual Typer command            | Validate an `AnonymizerConfig` (and optional `model_configs`).|
-| `nemo anonymizer preview submit`    | Generated from `NemoFunction`   | Local streaming preview.                                     |
 | `nemo anonymizer preview submit` | Generated from `NemoFunction`   | Remote streaming preview against the plugin service.         |
-| `nemo anonymizer run submit`        | Generated from `NemoJob`        | Local job execution in the CLI process.                      |
 | `nemo anonymizer run submit`     | Generated from `NemoJob`        | Submit an `anonymizer.run` job to the NeMo Platform Jobs worker. |
 | `nemo anonymizer run explain`    | Generated from `NemoJob`        | Print the job key, submit endpoint, and JSON schemas.        |
 
@@ -39,13 +37,9 @@ The command does not accept `data.source`. Input-source validation happens durin
 
 ## `nemo anonymizer preview`
 
-Both `preview submit` and `preview submit` take a spec file matching `PreviewRequest`.
+`preview submit` takes a spec file matching `PreviewRequest`.
 
 ```bash
-nemo anonymizer preview submit \
-  --spec-file /tmp/anonymizer-preview.yaml \
-  --workspace "${NMP_WORKSPACE:-default}"
-
 nemo anonymizer preview submit \
   --spec-file /tmp/anonymizer-preview.yaml \
   --workspace "${NMP_WORKSPACE:-default}" \
@@ -60,11 +54,11 @@ nemo anonymizer preview submit \
 
 ### Preview source kinds
 
-| Form                                  | `preview submit` | `preview submit` |
-|---------------------------------------|---------------|------------------|
-| Local path (`/tmp/input.csv`)         | yes           | no               |
-| HTTP(S) URL (`https://.../input.csv`) | yes           | yes              |
-| Fileset reference (`fs#path`)         | yes           | yes              |
+| Form                                  | `preview submit` |
+|---------------------------------------|------------------|
+| Local path (`/tmp/input.csv`)         | no               |
+| HTTP(S) URL (`https://.../input.csv`) | yes              |
+| Fileset reference (`fs#path`)         | yes              |
 
 ### Preview output
 
@@ -152,7 +146,7 @@ Both preview and run specs use the shared `AnonymizerRequest` shape:
 | `data.text_column`| string                                          | no       | Defaults to `text`.                                                    |
 | `data.id_column`  | string                                          | no       | Optional record identifier column.                                     |
 | `data.data_summary` | string                                        | no       | Optional short description of the data.                                |
-| `model_configs`   | list of Data Designer `ModelConfig`             | depends  | Required for `preview submit` and `run submit`; optional for `preview submit` and `run submit`. |
+| `model_configs`   | list of Data Designer `ModelConfig`             | depends  | Required for `preview submit` and `run submit`. |
 | `selected_models` | object with `detection` / `replace` / `rewrite` | no       | Role overrides on top of bundled defaults. Requires `model_configs`.   |
 
 Preview-only:

@@ -7,7 +7,7 @@ description: ""
 ---
 <a id="data-designer-cli"></a>
 
-The NeMo Data Designer plugin adds the `nemo data-designer` command group. Use it to execute Data Designer workloads locally in the CLI process or submit them to NeMo Services.
+The NeMo Data Designer plugin adds the `nemo data-designer` command group. Use it to submit preview and create workloads to NeMo Services.
 
 ## Configuration Sources
 
@@ -16,6 +16,7 @@ The `preview` and `create` commands accept a configuration source path. The most
 ```python
 import data_designer.config as dd
 
+
 def load_config_builder() -> dd.DataDesignerConfigBuilder:
     model_configs = [
         dd.ModelConfig(
@@ -30,41 +31,17 @@ def load_config_builder() -> dd.DataDesignerConfigBuilder:
     return config_builder
 ```
 
-The same configuration source can usually be used with `run` or `submit`. Resource choices determine whether it is compatible with NeMo Services execution; see [Execution Modes](/documentation/design-synthetic-data/execution-modes).
-
-## Run Versus Submit
+The same configuration source is used with `submit`. Resource choices determine whether it is compatible with NeMo Services execution; see [Execution Modes](/documentation/design-synthetic-data/execution-modes).
 
-`run` executes the Data Designer workload locally, in the CLI process. This can be fully local, but it is not an offline-only mode. A local run can still use the Files API, Secrets API, and Inference Gateway API from a running NeMo Services cluster when the configuration references the corresponding resources.
+## Submit Versus Platform Execution
 
 `submit` sends the workload to NeMo Services. The Data Designer API and Jobs API coordinate execution, job lifecycle, logs, and artifact persistence. The NeMo Services deployment may itself be local or remote.
 
 | Command | Workload execution | NeMo Services required? |
 |---------|--------------------|-------------------------|
-| `preview submit` | Local CLI process | Optional |
-| `create submit` | Local CLI process | Optional |
 | `preview submit` | Data Designer API | Yes |
 | `create submit` | Jobs worker | Yes |
 
-## Preview Locally
-
-Use local preview for fast iteration:
-
-```bash
-nemo data-designer preview submit product_reviews.py --num-records 5
-```
-
-The workload runs in your current Python environment. It can use local-only resources, NeMo resources, or both.
-
-## Create Locally
-
-Use local create when you want to generate a larger dataset without submitting work to NeMo Services:
-
-```bash
-nemo data-designer create submit product_reviews.py --num-records 1000
-```
-
-This executes the plugin job locally. It is useful for development and for workloads that should stay in the local environment.
-
 ## Submit Preview to NeMo Services
 
 Submit preview when you want to exercise the Data Designer API path:

@@ -11,15 +11,16 @@ This tutorial demonstrates how to use external datasets as seed data for synthet
 
 For more detail about seed dataset behavior, see the [open-source library's version](https://docs.nvidia.com/nemo/datadesigner/v0.9.1/tutorials/seeding-with-an-external-dataset) of this tutorial.
 
-## Seed Sources by Execution Mode
+## Seed Sources on the Platform
 
-Seed source support depends on platform resource availability:
+This tutorial covers platform execution (`nemo data-designer … submit` and the SDK). Seed support there is narrower than the open-source Data Designer library:
 
-| Seed source | CLI `submit` / SDK today | Use case |
-|-------------|--------------------------|----------|
-| **Local files or DataFrames** | Not supported | Upload local files to Filesets first. |
-| **HuggingFace** | Supported | Publicly available datasets or private HuggingFace datasets. |
+| Seed source | Platform (`submit` / SDK) | Notes |
+|-------------|---------------------------|-------|
+| **Local files or DataFrames** | Not supported on the platform path | Still valid in the [open-source library](https://docs.nvidia.com/nemo/datadesigner/v0.9.1/tutorials/seeding-with-an-external-dataset). For platform jobs, upload files to a Fileset first (see Step 1 below). |
+| **HuggingFace** | Supported | Public or private HuggingFace datasets (private needs a Secrets API token). |
 | **Files API Filesets** | Supported | Shared seed data stored through the Files API. |
+
 ### HuggingFace Datasets
 
 Use `HuggingFaceSeedSource` to load data from HuggingFace:

@@ -17,11 +17,11 @@ LLM-as-a-Judge evaluation sends each dataset row to a judge LLM and parses the j
 - **Pre-generated data**: Score existing question-answer pairs or conversations.
 - **Custom criteria**: Define range scores, rubric scores, prompt templates, and parser behavior.
 
-NeMo Evaluator supports two execution modes through the Evaluator plugin SDK:
+Use the Evaluator plugin SDK to submit durable platform jobs:
 
 | Mode | Use Case | SDK Call |
 |------|----------|----------|
-| **Durable remote job** | Production workloads that should run as platform jobs | `evaluator.submit(metric=metric, dataset=dataset)` |
+| **Platform job** | Durable platform execution with polling and result retrieval | `evaluator.submit(metric=metric, dataset=dataset)` |
 
 ## Prerequisites
 
@@ -125,8 +125,6 @@ result = LocalEvaluator().run_sync(
         },
     ],
 )
-job.wait_until_done()
-result = job.get_result()
 
 for score in result.aggregate_scores.scores:
     print(f"{score.name}: mean={score.mean:.2f}, count={score.count}")
@@ -230,8 +228,6 @@ result = LocalEvaluator().run_sync(
     ],
     aggregate_fields=("rubric_distribution", "mode_category"),
 )
-job.wait_until_done()
-result = job.get_result()
 
 print(result.aggregate_scores.model_dump(exclude_none=True))
 ```
@@ -251,8 +247,6 @@ result = LocalEvaluator().run_sync(
     ],
     aggregate_fields=("std_dev", "variance"),
 )
-job.wait_until_done()
-result = job.get_result()
 
 for score in result.aggregate_scores.scores:
     print(f"{score.name}:")

@@ -16,7 +16,7 @@ An `EvaluationResult` contains:
 - **Aggregate scores**: `result.aggregate_scores.scores`, with statistics such as mean, min, max, count, NaN count, variance, standard deviation, percentiles, and rubric distributions when applicable.
 - **Row scores**: `result.row_scores`, with the original dataset row, generated sample payload, metric scores, request logs, and row-level errors.
 
-## Get Results from a Local Run
+## Get Results Inline
 
 ```python
 from nemo_evaluator_sdk import Evaluator as LocalEvaluator

@@ -62,7 +62,7 @@
     path: ../../safe-synthesizer/about/data-synthesis.mdx
   - page: Evaluation
     path: ../../safe-synthesizer/about/evaluation.mdx
-  - page: Host-Local Development
+  - page: Runtime Environment
     path: ../../safe-synthesizer/about/host-local-development.mdx
   - page: Overview
     path: ../../safe-synthesizer/about/index.mdx

@@ -310,7 +310,7 @@ navigation:
                 path: ../../safe-synthesizer/about/evaluation.mdx
               - page: Jobs
                 path: ../../safe-synthesizer/about/jobs.mdx
-              - page: Local and Subprocess Execution
+              - page: Runtime Environment
                 path: ../../safe-synthesizer/about/host-local-development.mdx
               - page: Parameters Reference
                 path: ../../safe-synthesizer/about/reference.mdx

@@ -8,7 +8,7 @@ description: ""
 
 <a id="host-local-development"></a>
 
-NeMo Safe Synthesizer no longer exposes a host-local workload command. Create synthesis jobs through the platform Jobs API or SDK so scheduling, storage, logs, and artifacts all flow through the supported job path.
+Create Safe Synthesizer workloads through the platform Jobs API or SDK so scheduling, storage, logs, and artifacts all flow through the supported job path.
 
 The `nemo safe-synthesizer` CLI still includes runtime utilities for inspecting or preparing the task runtime used by development and job debugging.
 

@@ -15,7 +15,7 @@ Before using NeMo Safe Synthesizer, complete [Setup](/documentation/get-started)
 
 NeMo Safe Synthesizer has the following additional requirements:
 
-- A GPU-capable Jobs backend with enough VRAM for the selected model and dataset.
+- A GPU-capable Jobs backend with **80GB+ VRAM** recommended for the selected model and dataset (check with `nvidia-smi` on GPU nodes).
 - Sufficient disk space for generated datasets (50GB+ recommended)
 
 For general platform troubleshooting (port conflicts, health checks, and so on), refer to [Setup](/documentation/get-started).