Add embed model name prefix for LiteLLM proxy compatability

[ChrisJar]

Description

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[greptile-apps]

Greptile Summary

This PR adds an embed_model_name_prefix feature that prepends a LiteLLM route/org prefix to the embedding model name when using remote endpoints. It is a targeted fix to allow deployment with namespaced model IDs (e.g. "team/model/provider") required by LiteLLM proxies.

• New prepend_model_name_prefix utility in string_processing/__init__.py applies the prefix once, respecting None/empty-string guards and stripping leading slashes from the model name.
• normalize_embed_kwargs is now the single normalization point: it eagerly applies and removes embed_model_name_prefix from the task kwargs dict, ensuring the prefix propagates correctly through the EmbedParams → build_embed_kwargs path without double-application.
• Helm, CLI, service config, vectordb app, runtime, and agentic paths all receive the new parameter with consistent forwarding and correct remote-only semantics (prefix is silently dropped when no remote endpoint is configured).

Confidence Score: 5/5

Safe to merge — the prefix feature follows a clear remote-only contract and the three distinct application points (normalize_embed_kwargs, infer_microservice, _http_embed_openai_compat) are non-overlapping, so no double-prefix can occur.

The change is additive: all new parameters default to None/'' and produce identical behavior when unset. The normalization logic correctly pops embed_model_name_prefix before it can be applied a second time in any downstream path. Tests cover the key normalization scenarios, CLI paths, and the vectordb flow. The one untested path (agentic) is plumbing-only and has no novel logic.

The agentic embedding path in nemo_retriever/src/nemo_retriever/query/agentic.py has no test verifying that the new prefix reaches the actual HTTP embedding call.

Important Files Changed

Filename	Overview
nemo_retriever/src/nemo_retriever/common/api/util/string_processing/init.py	Adds prepend_model_name_prefix — correctly handles None/empty prefix and strips leading slashes from the model name.
nemo_retriever/src/nemo_retriever/common/params/utils.py	Extends normalize_embed_kwargs to pop and eagerly apply embed_model_name_prefix when a remote endpoint is present; build_embed_option_kwargs now plumbs through the new parameter.
nemo_retriever/src/nemo_retriever/models/inference/main_text_embed.py	Threads model_name_prefix through _async_runner → _async_request_handler → _make_async_request → _http_embed_openai_compat; prefix applied once just before the HTTP call.
nemo_retriever/src/nemo_retriever/models/nim/util/init.py	Adds model_name_prefix parameter to infer_microservice; prefix is applied via prepend_model_name_prefix on the HTTP branch only — correct, single-application point.
nemo_retriever/src/nemo_retriever/service/vectordb_app.py	Adds embed_model_name_prefix to _embed_queries_remote, VectorDBState, create_vectordb_app, and the argparse CLI entry point; --embed-model-name-prefix '' is correctly normalized to None.
nemo_retriever/src/nemo_retriever/query/agentic.py	Adds query_embedder_prefix to AgenticRetrievalConfig and forwards it as embed_model_name_prefix in embed_kwargs; no dedicated test for this path with the prefix feature.
nemo_retriever/tests/test_embed_params.py	Five new unit tests cover remote-only, bare-name, other-vendor-namespace, no-prefix, and prefix-applied scenarios for build_embed_option_kwargs.
nemo_retriever/tests/test_service_local_models.py	Extended test_build_embed_params_from_nim_config to assert embed_model_name_prefix is stored in EmbedParams — correct for lazy normalization, but downstream prefix application is not verified in this file.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A["User / Helm config\n(embed_model_name_prefix)"] --> B{Code path}

    B -->|"CLI / service / query"| C["build_embed_option_kwargs\nor build_embed_params"]
    C --> D["normalize_embed_kwargs\n(pops embed_model_name_prefix,\neagerly prefixes model_name\nif remote endpoint present)"]
    D --> E["EmbedParams / task_config\n(model_name already prefixed;\nno embed_model_name_prefix key)"]
    E --> F["Pipeline / HTTP call\n(model_name used as-is,\nno further prefix applied)"]

    B -->|"Library runtime\n(_embed_group)"| G["embed_text_main_text_embed\n(embed_model_name_prefix passed through)"]
    G --> H["_embed_group\n(sets task_config embed_model_name_prefix\nand TextEmbeddingConfig.embedding_model_prefix)"]
    H --> I["create_text_embeddings_for_df\n(resolves model_name_prefix from task_config)"]
    I --> J["_http_embed_openai_compat\n(calls prepend_model_name_prefix once,\nmodel_name still un-prefixed at this point)"]

    B -->|"VectorDB query"| K["_embed_queries_remote\n(embed_model_name_prefix passed)"]
    K --> L["infer_microservice\n(calls prepend_model_name_prefix once\non HTTP branch only)"]

    B -->|"Agentic query"| M["AgenticRetriever\n(embed_kwargs includes\nembed_model_name_prefix — untested)"]

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A["User / Helm config\n(embed_model_name_prefix)"] --> B{Code path}

    B -->|"CLI / service / query"| C["build_embed_option_kwargs\nor build_embed_params"]
    C --> D["normalize_embed_kwargs\n(pops embed_model_name_prefix,\neagerly prefixes model_name\nif remote endpoint present)"]
    D --> E["EmbedParams / task_config\n(model_name already prefixed;\nno embed_model_name_prefix key)"]
    E --> F["Pipeline / HTTP call\n(model_name used as-is,\nno further prefix applied)"]

    B -->|"Library runtime\n(_embed_group)"| G["embed_text_main_text_embed\n(embed_model_name_prefix passed through)"]
    G --> H["_embed_group\n(sets task_config embed_model_name_prefix\nand TextEmbeddingConfig.embedding_model_prefix)"]
    H --> I["create_text_embeddings_for_df\n(resolves model_name_prefix from task_config)"]
    I --> J["_http_embed_openai_compat\n(calls prepend_model_name_prefix once,\nmodel_name still un-prefixed at this point)"]

    B -->|"VectorDB query"| K["_embed_queries_remote\n(embed_model_name_prefix passed)"]
    K --> L["infer_microservice\n(calls prepend_model_name_prefix once\non HTTP branch only)"]

    B -->|"Agentic query"| M["AgenticRetriever\n(embed_kwargs includes\nembed_model_name_prefix — untested)"]

Loading

_{Reviews (3): Last reviewed commit: "Add embed model name prefix for LiteLLM ..." | Re-trigger Greptile}