feat: add static embedding metadata to dynamic model listings for providers using OpenAIMixin (#3547)

# What does this PR do?

- remove auto-download of ollama embedding models
- add embedding model metadata to dynamic listing w/ unit test
- add support and tests for allowed_models
- removed inference provider models.py files where dynamic listing is
enabled
- store embedding metadata in embedding_model_metadata field on
inference providers
- make model_entries optional on ModelRegistryHelper and
LiteLLMOpenAIMixin
- make OpenAIMixin a ModelRegistryHelper
- skip base64 embedding test for remote::ollama, always returns floats
- only use OpenAI client for ollama model listing
- remove unused build_model_entry function
- remove unused get_huggingface_repo function


## Test Plan

ci w/ new tests

llama_stack/providers/remote/inference/nvidia/models.py

@@ -1,109 +0,0 @@
-# Copyright (c) Meta Platforms, Inc. and affiliates.
-# All rights reserved.
-#
-# This source code is licensed under the terms described in the LICENSE file in
-# the root directory of this source tree.
-
-from llama_stack.apis.models import ModelType
-from llama_stack.models.llama.sku_types import CoreModelId
-from llama_stack.providers.utils.inference.model_registry import (
-    ProviderModelEntry,
-    build_hf_repo_model_entry,
-)
-
-SAFETY_MODELS_ENTRIES = []
-
-# https://docs.nvidia.com/nim/large-language-models/latest/supported-llm-agnostic-architectures.html
-MODEL_ENTRIES = [
-    build_hf_repo_model_entry(
-        "meta/llama3-8b-instruct",
-        CoreModelId.llama3_8b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama3-70b-instruct",
-        CoreModelId.llama3_70b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.1-8b-instruct",
-        CoreModelId.llama3_1_8b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.1-70b-instruct",
-        CoreModelId.llama3_1_70b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.1-405b-instruct",
-        CoreModelId.llama3_1_405b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.2-1b-instruct",
-        CoreModelId.llama3_2_1b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.2-3b-instruct",
-        CoreModelId.llama3_2_3b_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.2-11b-vision-instruct",
-        CoreModelId.llama3_2_11b_vision_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.2-90b-vision-instruct",
-        CoreModelId.llama3_2_90b_vision_instruct.value,
-    ),
-    build_hf_repo_model_entry(
-        "meta/llama-3.3-70b-instruct",
-        CoreModelId.llama3_3_70b_instruct.value,
-    ),
-    ProviderModelEntry(
-        provider_model_id="nvidia/vila",
-        model_type=ModelType.llm,
-    ),
-    # NeMo Retriever Text Embedding models -
-    #
-    # https://docs.nvidia.com/nim/nemo-retriever/text-embedding/latest/support-matrix.html
-    #
-    # +-----------------------------------+--------+-----------+-----------+------------+
-    # | Model ID                          | Max    | Publisher | Embedding | Dynamic    |
-    # |                                   | Tokens |           | Dimension | Embeddings |
-    # +-----------------------------------+--------+-----------+-----------+------------+
-    # | nvidia/llama-3.2-nv-embedqa-1b-v2 | 8192   | NVIDIA    | 2048      | Yes        |
-    # | nvidia/nv-embedqa-e5-v5           |  512   | NVIDIA    | 1024      |  No        |
-    # | nvidia/nv-embedqa-mistral-7b-v2   |  512   | NVIDIA    | 4096      |  No        |
-    # | snowflake/arctic-embed-l          |  512   | Snowflake | 1024      |  No        |
-    # +-----------------------------------+--------+-----------+-----------+------------+
-    ProviderModelEntry(
-        provider_model_id="nvidia/llama-3.2-nv-embedqa-1b-v2",
-        model_type=ModelType.embedding,
-        metadata={
-            "embedding_dimension": 2048,
-            "context_length": 8192,
-        },
-    ),
-    ProviderModelEntry(
-        provider_model_id="nvidia/nv-embedqa-e5-v5",
-        model_type=ModelType.embedding,
-        metadata={
-            "embedding_dimension": 1024,
-            "context_length": 512,
-        },
-    ),
-    ProviderModelEntry(
-        provider_model_id="nvidia/nv-embedqa-mistral-7b-v2",
-        model_type=ModelType.embedding,
-        metadata={
-            "embedding_dimension": 4096,
-            "context_length": 512,
-        },
-    ),
-    ProviderModelEntry(
-        provider_model_id="snowflake/arctic-embed-l",
-        model_type=ModelType.embedding,
-        metadata={
-            "embedding_dimension": 1024,
-            "context_length": 512,
-        },
-    ),
-    # TODO(mf): how do we handle Nemotron models?
-    # "Llama3.1-Nemotron-51B-Instruct" -> "meta/llama-3.1-nemotron-51b-instruct",
-] + SAFETY_MODELS_ENTRIES

llama_stack/providers/remote/inference/nvidia/nvidia.py

@@ -37,9 +37,6 @@ from llama_stack.apis.inference import (
 )
 from llama_stack.log import get_logger
 from llama_stack.models.llama.datatypes import ToolDefinition, ToolPromptFormat
-from llama_stack.providers.utils.inference.model_registry import (
-    ModelRegistryHelper,
-)
 from llama_stack.providers.utils.inference.openai_compat import (
     convert_openai_chat_completion_choice,
     convert_openai_chat_completion_stream,
@@ -48,7 +45,6 @@ from llama_stack.providers.utils.inference.openai_mixin import OpenAIMixin
 from llama_stack.providers.utils.inference.prompt_adapter import content_has_media
 
 from . import NVIDIAConfig
-from .models import MODEL_ENTRIES
 from .openai_utils import (
     convert_chat_completion_request,
     convert_completion_request,
@@ -60,7 +56,7 @@ from .utils import _is_nvidia_hosted
 logger = get_logger(name=__name__, category="inference::nvidia")
 
 
-class NVIDIAInferenceAdapter(OpenAIMixin, Inference, ModelRegistryHelper):
+class NVIDIAInferenceAdapter(OpenAIMixin, Inference):
     """
     NVIDIA Inference Adapter for Llama Stack.
 
@@ -74,10 +70,15 @@ class NVIDIAInferenceAdapter(OpenAIMixin, Inference, ModelRegistryHelper):
     - ModelRegistryHelper.check_model_availability() just returns False and shows a warning
     """
 
-    def __init__(self, config: NVIDIAConfig) -> None:
-        # TODO(mf): filter by available models
-        ModelRegistryHelper.__init__(self, model_entries=MODEL_ENTRIES)
+    # source: https://docs.nvidia.com/nim/nemo-retriever/text-embedding/latest/support-matrix.html
+    embedding_model_metadata = {
+        "nvidia/llama-3.2-nv-embedqa-1b-v2": {"embedding_dimension": 2048, "context_length": 8192},
+        "nvidia/nv-embedqa-e5-v5": {"embedding_dimension": 512, "context_length": 1024},
+        "nvidia/nv-embedqa-mistral-7b-v2": {"embedding_dimension": 512, "context_length": 4096},
+        "snowflake/arctic-embed-l": {"embedding_dimension": 512, "context_length": 1024},
+    }
 
+    def __init__(self, config: NVIDIAConfig) -> None:
         logger.info(f"Initializing NVIDIAInferenceAdapter({config.url})...")
 
         if _is_nvidia_hosted(config):