feat(cli): expose env-only infrastructure settings as CLI flags

docs/user-guide/docker.md

@@ -244,7 +244,7 @@ docker run --gpus all --shm-size=1g \
   nss-gpu:latest run --config /workspace/data/config.yaml --data-source /workspace/data/input.csv
 ```
 
-See [Environment Variables -- Hugging Face Cache](environment.md#hugging-face-cache)
+See [Environment Variables -- Hugging Face cache and offline](environment.md#hugging-face-cache-and-offline)
 for details on `HF_HOME`, `HF_HUB_OFFLINE`, and `VLLM_CACHE_ROOT`.
 
 ---

docs/user-guide/running.md

@@ -274,7 +274,7 @@ execute in order (`config` → `dataframe` → `metadata` → `advisory`).
 | Check name | Stage | What it validates |
 |-------|-------|-------------------|
 | `gpu.cuda` | config | PyTorch is importable and a CUDA GPU is visible |
-| `env.inference_key` | config | `NSS_INFERENCE_KEY` is set when PII classification is enabled (warning only) |
+| `env.inference` | config | Inference config for PII classification: `NSS_INFERENCE_KEY` is set, `NSS_INFERENCE_MODEL` is non-empty, and `NSS_INFERENCE_ENDPOINT` is a valid http(s) URL (warnings only) |
 | `env.hf_model_availability` | config | The pretrained model reference is usable locally or can be fetched from Hugging Face; warns about a missing HF token only when online HF access may be needed |
 | `dataset.size` | dataframe | Training split meets the hard minimum row count |
 | `columns.groupby` | dataframe | `group_training_examples_by` column is present and has no nulls |
@@ -1230,10 +1230,12 @@ See [`artifacts clean`](#artifacts-clean) in the CLI Commands section for option
 ## Running in Offline Environments
 
 Pre-cache models by running once with internet access, then set
-`HF_HUB_OFFLINE=1` in your target environment. For detailed cache setup
-and environment variables (`HF_HOME`, `HF_HUB_OFFLINE`, `LOCAL_FILES_ONLY`,
-`VLLM_CACHE_ROOT`), see
-[Environment Variables -- Hugging Face Cache](environment.md#hugging-face-cache).
+`HF_HUB_OFFLINE=1` in your target environment. Export it before launching
+`safe-synthesizer` (or pass `--disable-huggingface-remote`) -- huggingface_hub
+reads the value once at import time, so setting it after the process starts has
+no effect. For detailed cache setup and environment variables (`HF_HOME`,
+`HF_HUB_OFFLINE`, `VLLM_CACHE_ROOT`), see
+[Environment Variables -- Hugging Face cache and offline](environment.md#hugging-face-cache-and-offline).
 
 For offline-specific errors, see [Program Runtime](troubleshooting.md).
 

docs/user-guide/troubleshooting.md

@@ -513,7 +513,9 @@ check of its own.
 | `torch_missing` | error | `gpu.cuda` | PyTorch not installed; cannot verify GPU availability |
 | `no_gpu` | error | `gpu.cuda` | No CUDA GPU detected (required for training or generation) |
 | `low_vram` | warning | `gpu.vram` | Free GPU VRAM may be insufficient |
-| `inference_key_missing` | warning | `env.inference_key` | `NSS_INFERENCE_KEY` not set; PII classification degraded |
+| `inference_key_missing` | warning | `env.inference` | `NSS_INFERENCE_KEY` not set; PII classification degraded |
+| `inference_model_blank` | warning | `env.inference` | `NSS_INFERENCE_MODEL` set but empty; the blank value is ignored and the default model id is used |
+| `inference_endpoint_invalid` | error | `env.inference` | `NSS_INFERENCE_ENDPOINT` set but not a valid http(s) URL; classification requests will fail |
 | `hf_token_missing` | warning | `env.hf_model_availability` | Neither `HF_TOKEN` nor `HUGGING_FACE_HUB_TOKEN` set, and model loading may need online Hugging Face access |
 | `hf_model_not_cached` | warning/error | `env.hf_model_availability` | Hugging Face model is not present in the local cache; severity is error when HF offline mode is enabled |
 | `hf_model_cache_incomplete` | warning/error | `env.hf_model_availability` | Cached Hugging Face model snapshot is missing required config, tokenizer, weights, or shards; severity is error when HF offline mode is enabled |
@@ -545,7 +547,29 @@ The PII replacer downloads the GLiNER NER model on first use. If the download
 fails, it raises an exception immediately.
 
 Pre-download the model by running PII replacement once in an environment
-with internet access, or set `LOCAL_FILES_ONLY=true` after the model is cached.
+with internet access. To force offline use after the model is cached, set
+`HF_HUB_OFFLINE=1` or pass `--disable-huggingface-remote`.
+
+### Offline Mode Not Taking Effect
+
+Symptom: `HF_HUB_OFFLINE=1` (or `--disable-huggingface-remote`) is set, yet the
+run still attempts a download, or `--enable-huggingface-remote` does not
+re-enable downloads.
+
+Cause: huggingface_hub reads `HF_HUB_OFFLINE` once, at import time, and caches
+it. If the variable is changed after huggingface_hub has been imported in the
+process, the change is ignored.
+
+Fixes:
+
+- CLI: export `HF_HUB_OFFLINE` before launching `safe-synthesizer`, or use
+  `--enable-huggingface-remote` / `--disable-huggingface-remote`. The CLI
+  applies the flag before huggingface_hub loads, so the flag always wins over
+  an inherited environment value.
+- Programmatic / SDK: set `HF_HUB_OFFLINE` before importing
+  `nemo_safe_synthesizer` (or any library that imports huggingface_hub, such as
+  `transformers` or `datasets`). Setting it afterward has no effect for that
+  process.
 
 ### NER Processing Timeouts
 

script/slurm/slurm_nss_matrix.sh

@@ -129,7 +129,7 @@ echo "[NSS SLURM] nemo-safe-synthesizer version: $(python -c 'from nemo_safe_syn
 
 # for column classification
 export NSS_INFERENCE_ENDPOINT=https://integrate.api.nvidia.com/v1
-export NIM_MODEL_ID=qwen/qwen3-next-80b-a3b-instruct
+export NSS_INFERENCE_MODEL=qwen/qwen3-next-80b-a3b-instruct
 
 # Extract dataset name for path construction (handles both full paths and simple names)
 # e.g., "/path/to/adult.csv" -> "adult", "/path/to/data.parquet" -> "data", "adult" -> "adult"

src/nemo_safe_synthesizer/cli/run.py

@@ -159,6 +159,60 @@ def common_run_options(f: Callable[..., object]) -> Callable[..., object]:
             "If both env var and CLI option are provided, the CLI option takes precedence.",
         )
     )
+    options.append(
+        click.option(
+            "--inference-endpoint-url",
+            type=str,
+            required=False,
+            default=None,
+            help="OpenAI-compatible inference endpoint URL for PII column classification. "
+            "Can also be set via NSS_INFERENCE_ENDPOINT env var.",
+        )
+    )
+    options.append(
+        click.option(
+            "--inference-api-key",
+            type=str,
+            required=False,
+            default=None,
+            help="API key for the inference endpoint used in PII column classification. "
+            "Can also be set via NSS_INFERENCE_KEY env var.",
+        )
+    )
+    options.append(
+        click.option(
+            "--inference-model-id",
+            type=str,
+            required=False,
+            default=None,
+            help="Model ID sent to the inference endpoint for PII column classification. "
+            "Can also be set via NSS_INFERENCE_MODEL env var. "
+            "[default: qwen/qwen3-next-80b-a3b-instruct]",
+        )
+    )
+    options.append(
+        click.option(
+            "--enable-huggingface-remote/--disable-huggingface-remote",
+            "huggingface_remote",
+            required=False,
+            default=None,
+            help="Allow or block Hugging Face remote downloads for both the base model "
+            "and GLiNER. --disable-huggingface-remote forces a fully offline run by "
+            "setting HF_HUB_OFFLINE and TRANSFORMERS_OFFLINE; both must already be "
+            "cached. Equivalent to setting HF_HUB_OFFLINE in the environment.",
+        )
+    )
+    options.append(
+        click.option(
+            "--cpu-count",
+            type=int,
+            required=False,
+            default=None,
+            help="Number of CPU worker processes used for NER (PII replacement). "
+            "Can also be set via NSS_PII_REPLACER_CPU_COUNT env var. "
+            "[default: max(1, cpu_count - 1)]",
+        )
+    )
     # Apply each option decorator in reverse order (decorators apply bottom-up)
     for option in reversed(options):
         f = option(f)
@@ -170,6 +224,31 @@ def _parse_run_overrides(kwargs: dict[str, Any]) -> dict[str, Any]:
     return parse_overrides(kwargs)
 
 
+# CLISettings fields populated from common_run_options flags. ``synthesis_overrides``
+# is excluded -- it is derived from the leftover pydantic_options kwargs, not bound
+# to a single flag. ``observability``/``wandb`` are nested sub-settings with no CLI
+# flag, so they never appear in command kwargs.
+_CLI_SETTINGS_FIELDS: frozenset[str] = frozenset(CLISettings.model_fields) - {"synthesis_overrides"}
+
+
+def _settings_from_run_kwargs(kwargs: dict[str, Any]) -> CLISettings:
+    """Build ``CLISettings`` from a run command's ``**kwargs``.
+
+    ``common_run_options`` binds each infrastructure flag to a kwarg whose name
+    matches a ``CLISettings`` field; those are pulled out here. Everything left
+    (the ``pydantic_options`` ``--section__field`` options) becomes synthesis
+    overrides. This keeps the three run commands from re-listing the shared flag
+    set in both their signature and their settings construction -- adding a flag
+    now means editing ``common_run_options`` and ``CLISettings`` only.
+
+    ``kwargs`` is mutated: matched settings keys are popped before the remainder
+    is parsed into overrides.
+    """
+    settings_kwargs = {name: kwargs.pop(name) for name in _CLI_SETTINGS_FIELDS if name in kwargs}
+    settings_kwargs["synthesis_overrides"] = _parse_run_overrides(kwargs)
+    return CLISettings.from_cli_kwargs(**settings_kwargs)
+
+
 def _set_cli_deployment_type_default() -> None:
     """Default telemetry deployment type for CLI commands without overriding Slurm or explicit settings."""
     os.environ.setdefault("NEMO_DEPLOYMENT_TYPE", DeploymentTypeEnum.CLI.value)
@@ -311,20 +390,8 @@ def _build_validate_render_context(
 )
 def run(
     ctx: click.Context,
-    config_path: PathT | None,
-    data_source: str | None,
-    artifact_path: PathT | None,
-    run_path: PathT | None,
-    output_file: PathT | None,
-    log_file: PathT | None,
-    log_color: bool | None,
-    log_format: str | None,
-    verbose: int = 0,
-    wandb_mode: str | None = None,
-    wandb_project: str | None = None,
-    dataset_registry: str | None = None,
     validate: bool = False,
-    **kwargs: object,
+    **kwargs: Any,
 ) -> None:
     """Run the Safe Synthesizer end-to-end pipeline.
 
@@ -337,21 +404,7 @@ def run(
 
     _set_cli_deployment_type_default()
 
-    settings = CLISettings.from_cli_kwargs(
-        data_source=data_source,
-        config_path=config_path,
-        artifact_path=artifact_path,
-        run_path=run_path,
-        output_file=output_file,
-        log_file=log_file,
-        log_color=log_color,
-        log_format=log_format,
-        verbose=verbose,
-        wandb_mode=wandb_mode,
-        wandb_project=wandb_project,
-        synthesis_overrides=_parse_run_overrides(kwargs),
-        dataset_registry=dataset_registry,
-    )
+    settings = _settings_from_run_kwargs(kwargs)
 
     if validate:
         os.environ["NSS_PHASE"] = "process_data"
@@ -407,20 +460,8 @@ def run(
     help="Run pre-flight validation only, then exit without training or generating.",
 )
 def run_train(
-    config_path: PathT,
-    data_source: str | None,
-    artifact_path: PathT | None,
-    run_path: PathT | None,
-    output_file: PathT | None,
-    log_format: str | None,
-    log_color: bool | None,
-    log_file: PathT | None,
-    verbose: int,
-    wandb_mode: str | None = None,
-    wandb_project: str | None = None,
-    dataset_registry: str | None = None,
     validate: bool = False,
-    **kwargs: object,
+    **kwargs: Any,
 ) -> None:
     """Run the training stage only.
 
@@ -429,21 +470,7 @@ def run_train(
     """
     _set_cli_deployment_type_default()
 
-    settings = CLISettings.from_cli_kwargs(
-        data_source=data_source,
-        config_path=config_path,
-        artifact_path=artifact_path,
-        run_path=run_path,
-        output_file=output_file,
-        log_file=log_file,
-        log_color=log_color,
-        log_format=log_format,
-        verbose=verbose,
-        wandb_mode=wandb_mode,
-        wandb_project=wandb_project,
-        synthesis_overrides=_parse_run_overrides(kwargs),
-        dataset_registry=dataset_registry,
-    )
+    settings = _settings_from_run_kwargs(kwargs)
 
     if validate:
         os.environ["NSS_PHASE"] = "process_data"
@@ -498,21 +525,9 @@ def run_train(
 )
 @pydantic_options(SafeSynthesizerParameters, field_separator=CLI_NESTED_FIELD_SEPARATOR)
 def run_generate(
-    config_path: PathT,
-    data_source: str | None,
-    run_path: PathT | None,
-    artifact_path: PathT | None,
-    output_file: PathT | None,
-    log_format: str | None,
-    log_color: bool | None,
-    log_file: PathT | None,
-    verbose: int,
-    wandb_mode: str | None = None,
-    wandb_project: str | None = None,
     auto_discover_adapter: bool = False,
     wandb_resume_job_id: str | None = None,
-    dataset_registry: str | None = None,
-    **kwargs: object,
+    **kwargs: Any,
 ) -> None:
     """Run the generation stage only.
 
@@ -526,21 +541,7 @@ def run_generate(
     _set_cli_deployment_type_default()
 
     # Create unified settings from CLI kwargs
-    settings = CLISettings.from_cli_kwargs(
-        data_source=data_source,
-        config_path=config_path,
-        artifact_path=artifact_path,
-        run_path=run_path,
-        output_file=output_file,
-        log_file=log_file,
-        log_color=log_color,
-        log_format=log_format,
-        verbose=verbose,
-        wandb_mode=wandb_mode,
-        wandb_project=wandb_project,
-        synthesis_overrides=_parse_run_overrides(kwargs),
-        dataset_registry=dataset_registry,
-    )
+    settings = _settings_from_run_kwargs(kwargs)
 
     os.environ["NSS_PHASE"] = "generate"
     # Generation always resumes from an existing workdir with a trained model

src/nemo_safe_synthesizer/cli/settings.py

@@ -118,9 +118,10 @@ class CLISettings(BaseSettings):
 
     log_color: bool | None = Field(
         default=None,
+        validation_alias=AliasChoices("log_color", "NSS_LOG_COLOR"),
         description="Whether to colorize console output",
     )
-    """Whether to colorize console output."""
+    """Whether to colorize console output (env variable: ``NSS_LOG_COLOR``)."""
 
     log_file: str | None = Field(
         default=None,
@@ -164,6 +165,51 @@ class CLISettings(BaseSettings):
     )
     """URL or path to a dataset registry YAML file (env: ``NSS_DATASET_REGISTRY``)."""
 
+    inference_endpoint_url: str | None = Field(
+        default=None,
+        validation_alias=AliasChoices("inference_endpoint_url", "NSS_INFERENCE_ENDPOINT"),
+        description="OpenAI-compatible inference endpoint URL for PII column classification",
+    )
+    """OpenAI-compatible inference endpoint URL for PII column classification
+    (env: ``NSS_INFERENCE_ENDPOINT``)."""
+
+    inference_api_key: str | None = Field(
+        default=None,
+        validation_alias=AliasChoices("inference_api_key", "NSS_INFERENCE_KEY"),
+        description="API key for the inference endpoint used in PII column classification",
+    )
+    """API key for the inference endpoint used in PII column classification
+    (env: ``NSS_INFERENCE_KEY``)."""
+
+    inference_model_id: str | None = Field(
+        default=None,
+        validation_alias=AliasChoices("inference_model_id", "NSS_INFERENCE_MODEL"),
+        description="Model ID sent to the inference endpoint for PII column classification",
+    )
+    """Model ID sent to the inference endpoint for PII column classification
+    (env: ``NSS_INFERENCE_MODEL``)."""
+
+    huggingface_remote: bool | None = Field(
+        default=None,
+        validation_alias=AliasChoices("huggingface_remote"),
+        description="Whether to allow Hugging Face remote downloads (base model and GLiNER)",
+    )
+    """Whether to allow Hugging Face remote downloads for the base model and GLiNER.
+
+    ``None`` leaves the environment untouched. ``True`` / ``False`` is propagated
+    to the standard ``HF_HUB_OFFLINE`` and ``TRANSFORMERS_OFFLINE`` variables (the
+    canonical env switch) by ``_propagate_runtime_settings_to_env``; there is no
+    separate NSS env var."""
+
+    cpu_count: int | None = Field(
+        default=None,
+        ge=1,
+        validation_alias=AliasChoices("cpu_count", "NSS_PII_REPLACER_CPU_COUNT"),
+        description="Number of CPU worker processes used for NER (PII replacement)",
+    )
+    """Number of CPU worker processes used for NER (PII replacement)
+    (env: ``NSS_PII_REPLACER_CPU_COUNT``)."""
+
     @field_validator("wandb_mode", mode="before")
     @classmethod
     def validate_wandb_mode(cls, v: str | WandbMode | None) -> WandbMode | None: