nvidia-haystack 1.4.0__tar.gz → 3.0.0__tar.gz

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/CHANGELOG.md

@@ -1,5 +1,27 @@
 # Changelog
 
+## [integrations/nvidia-v2.0.0] - 2026-09-08
+
+### 🚀 Features
+
+- [**breaking**] Nvidia - add `close` method; change the default models (#3930)
+
+### ⚙️ CI
+
+- Improve changelog generation; fix existing changelogs (#3883)
+
+### 🧹 Chores
+
+- [**breaking**] Remove deprecated `NvidiaGenerator` (use `NvidiaChatGenerator` instead) (#3891)
+
+
+## [integrations/nvidia-v1.4.0] - 2026-08-11
+
+### 🐛 Bug Fixes
+
+- NvidiaRanker reordered the top_k fallback to happen after validation (#3749)
+
+
 ## [integrations/nvidia-v1.3.0] - 2026-08-04
 
 ### 🐛 Bug Fixes
@@ -185,9 +207,14 @@
 
 ### 🚀 Features
 
-- Improvements to NvidiaRanker and adding user input timeout (#1193)
 - Add model `nvidia/llama-3.2-nv-rerankqa-1b-v2` to `_MODEL_ENDPOINT_MAP` (#1260)
 
+## [integrations/nvidia-v0.1.2] - 2024-11-25
+
+### 🚀 Features
+
+- Improvements to NvidiaRanker and adding user input timeout (#1193)
+
 ### 🧹 Chores
 
 - Fix linting/isort (#1215)
@@ -204,18 +231,34 @@
 
 ### 🚀 Features
 
-- Update default embedding model to nvidia/nv-embedqa-e5-v5 (#1015)
-- Add NVIDIA NIM ranker support (#1023)
 - Raise error when attempting to embed empty documents/strings with Nvidia embedders (#1118)
 
 ### 🐛 Bug Fixes
 
-- Lints in `nvidia-haystack` (#993)
 - Missing Nvidia embedding truncate mode (#1043)
 
-### 🚜 Refactor
+### ⚙️ CI
 
-- Remove deprecated Nvidia Cloud Functions backend and related code. (#803)
+- Adopt uv as installer (#1142)
+
+### 🧹 Chores
+
+- Update ruff linting scripts and settings (#1105)
+
+### 🌀 Miscellaneous
+
+- Feat: add nvidia/llama-3.2-nv-rerankqa-1b-v1 to set of known ranking models (#1183)
+
+## [integrations/nvidia-v0.0.5] - 2024-08-28
+
+### 🚀 Features
+
+- Update default embedding model to nvidia/nv-embedqa-e5-v5 (#1015)
+- Add NVIDIA NIM ranker support (#1023)
+
+### 🐛 Bug Fixes
+
+- Lints in `nvidia-haystack` (#993)
 
 ### 📚 Documentation
 
@@ -225,25 +268,31 @@
 
 - Do not retry tests in `hatch run test` command (#954)
 
+### 🌀 Miscellaneous
+
+- Raise warning for base_url ../embeddings .../completions .../rankings (#922)
+- Update NvidiaGenerator docstrings (#966)
+-  Add default model for NVIDIA HayStack local NIM endpoints (#915)
+
+## [integrations/nvidia-v0.0.4] - 2024-07-24
+
+### 🚜 Refactor
+
+- Remove deprecated Nvidia Cloud Functions backend and related code. (#803)
+
 ### ⚙️ CI
 
 - Retry tests to reduce flakyness (#836)
-- Adopt uv as installer (#1142)
 
 ### 🧹 Chores
 
 - Update ruff invocation to include check parameter (#853)
-- Update ruff linting scripts and settings (#1105)
 
 ### 🌀 Miscellaneous
 
 - Fix: make hosted nim default (#734)
 - Fix: align tests and docs on NVIDIA_API_KEY (instead of NVIDIA_CATALOG_API_KEY) (#731)
 - Ci: install `pytest-rerunfailures` where needed; add retry config to `test-cov` script (#845)
-- Raise warning for base_url ../embeddings .../completions .../rankings (#922)
-- Update NvidiaGenerator docstrings (#966)
--  Add default model for NVIDIA HayStack local NIM endpoints (#915)
-- Feat: add nvidia/llama-3.2-nv-rerankqa-1b-v1 to set of known ranking models (#1183)
 
 ## [integrations/nvidia-v0.0.3] - 2024-05-22
 

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.4
+Metadata-Version: 2.5
 Name: nvidia-haystack
-Version: 1.4.0
+Version: 3.0.0
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/nvidia#readme
 Project-URL: Issues, https://github.com/deepset-ai/haystack-core-integrations/issues
 Project-URL: Source, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/nvidia
@@ -10,15 +10,14 @@ License-File: LICENSE.txt
 Classifier: Development Status :: 4 - Beta
 Classifier: License :: OSI Approved :: Apache Software License
 Classifier: Programming Language :: Python
-Classifier: Programming Language :: Python :: 3.10
 Classifier: Programming Language :: Python :: 3.11
 Classifier: Programming Language :: Python :: 3.12
 Classifier: Programming Language :: Python :: 3.13
 Classifier: Programming Language :: Python :: 3.14
 Classifier: Programming Language :: Python :: Implementation :: CPython
 Classifier: Programming Language :: Python :: Implementation :: PyPy
-Requires-Python: >=3.10
-Requires-Dist: haystack-ai>=2.22.0
+Requires-Python: >=3.11
+Requires-Dist: haystack-ai>=3.0.0
 Requires-Dist: requests>=2.25.0
 Requires-Dist: tqdm>=4.21.0
 Description-Content-Type: text/markdown

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/examples/chat_generator_with_structured_outputs.py

@@ -27,7 +27,7 @@ chat_messages = [
 ]
 
 component = NvidiaChatGenerator(
-    model="meta/llama-3.1-70b-instruct",
+    model="nvidia/nemotron-3.5-lightning-30b-a3b",
     generation_kwargs={"extra_body": {"nvext": {"guided_json": json_schema}}},
 )
 results = component.run(chat_messages)

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/pydoc/config_docusaurus.yml

@@ -3,7 +3,6 @@ loaders:
       - haystack_integrations.components.embedders.nvidia.document_embedder
       - haystack_integrations.components.embedders.nvidia.text_embedder
       - haystack_integrations.components.embedders.nvidia.truncate
-      - haystack_integrations.components.generators.nvidia.generator
       - haystack_integrations.components.generators.nvidia.chat.chat_generator
       - haystack_integrations.components.rankers.nvidia.ranker
       - haystack_integrations.components.rankers.nvidia.truncate

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/pyproject.toml

@@ -7,7 +7,7 @@ name = "nvidia-haystack"
 dynamic = ["version"]
 description = ''
 readme = "README.md"
-requires-python = ">=3.10"
+requires-python = ">=3.11"
 license = "Apache-2.0"
 keywords = []
 authors = [{ name = "deepset GmbH", email = "info@deepset.ai" }]
@@ -15,7 +15,6 @@ classifiers = [
   "License :: OSI Approved :: Apache Software License",
   "Development Status :: 4 - Beta",
   "Programming Language :: Python",
-  "Programming Language :: Python :: 3.10",
   "Programming Language :: Python :: 3.11",
   "Programming Language :: Python :: 3.12",
   "Programming Language :: Python :: 3.13",
@@ -23,7 +22,7 @@ classifiers = [
   "Programming Language :: Python :: Implementation :: CPython",
   "Programming Language :: Python :: Implementation :: PyPy",
 ]
-dependencies = ["haystack-ai>=2.22.0", "requests>=2.25.0", "tqdm>=4.21.0"]
+dependencies = ["haystack-ai>=3.0.0", "requests>=2.25.0", "tqdm>=4.21.0"]
 
 [project.urls]
 Documentation = "https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/nvidia#readme"
@@ -168,4 +167,4 @@ addopts = "--strict-markers"
 markers = [
   "integration: integration tests",
 ]
-log_cli = true
+log_cli = true

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/embedders/nvidia/document_embedder.py

@@ -7,8 +7,8 @@ import warnings
 from dataclasses import replace
 from typing import Any
 
-from haystack import Document, component, default_from_dict, default_to_dict, logging
-from haystack.utils import Secret, deserialize_secrets_inplace
+from haystack import Document, component, default_to_dict, logging
+from haystack.utils import Secret
 from tqdm import tqdm
 
 from haystack_integrations.components.embedders.nvidia.truncate import EmbeddingTruncateMode
@@ -28,7 +28,7 @@ class NvidiaDocumentEmbedder:
 
     doc = Document(content="I love pizza!")
 
-    text_embedder = NvidiaDocumentEmbedder(model="nvidia/nv-embedqa-e5-v5", api_url="https://integrate.api.nvidia.com/v1")
+    text_embedder = NvidiaDocumentEmbedder(model="nvidia/nemotron-3-embed-1b", api_url="https://integrate.api.nvidia.com/v1")
     # Components warm up automatically on first run.
 
     result = document_embedder.run([doc])
@@ -97,8 +97,7 @@ class NvidiaDocumentEmbedder:
             truncate = EmbeddingTruncateMode.from_str(truncate)
         self.truncate = truncate
 
-        self.backend: Any | None = None
-        self._initialized = False
+        self.backend: NimBackend | None = None
 
         if timeout is None:
             timeout = float(os.environ.get("NVIDIA_TIMEOUT", "60.0"))
@@ -134,7 +133,7 @@ class NvidiaDocumentEmbedder:
         """
         Initializes the component.
         """
-        if self._initialized:
+        if self.backend is not None:
             return
 
         model_kwargs = {"input_type": "passage"}
@@ -152,11 +151,15 @@ class NvidiaDocumentEmbedder:
         if not self.model and self.backend.model:
             self.model = self.backend.model
 
-        self._initialized = True
-
         if not self.model:
             self.default_model()
 
+    def close(self) -> None:
+        """Close the backend and release its resources."""
+        if self.backend is not None:
+            self.backend.close()
+            self.backend = None
+
     def to_dict(self) -> dict[str, Any]:
         """
         Serializes the component to a dictionary.
@@ -186,21 +189,6 @@ class NvidiaDocumentEmbedder:
         """
         return self.backend.models() if self.backend else []
 
-    @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> "NvidiaDocumentEmbedder":
-        """
-        Deserializes the component from a dictionary.
-
-        :param data:
-            The dictionary to deserialize from.
-        :returns:
-            The deserialized component.
-        """
-        init_parameters = data.get("init_parameters", {})
-        if init_parameters:
-            deserialize_secrets_inplace(data["init_parameters"], keys=["api_key"])
-        return default_from_dict(cls, data)
-
     def _prepare_texts_to_embed(self, documents: list[Document]) -> list[str]:
         texts_to_embed = []
         for doc in documents:
@@ -250,8 +238,7 @@ class NvidiaDocumentEmbedder:
         :raises TypeError:
             If the input is not a list of Documents.
         """
-        if not self._initialized:
-            self.warm_up()
+        self.warm_up()
 
         if not isinstance(documents, list) or (documents and not isinstance(documents[0], Document)):
             msg = (

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/embedders/nvidia/text_embedder.py

@@ -6,8 +6,8 @@ import os
 import warnings
 from typing import Any
 
-from haystack import component, default_from_dict, default_to_dict, logging
-from haystack.utils import Secret, deserialize_secrets_inplace
+from haystack import component, default_to_dict, logging
+from haystack.utils import Secret
 
 from haystack_integrations.components.embedders.nvidia.truncate import EmbeddingTruncateMode
 from haystack_integrations.utils.nvidia import DEFAULT_API_URL, Client, Model, NimBackend, url_validation
@@ -29,7 +29,7 @@ class NvidiaTextEmbedder:
 
     text_to_embed = "I love pizza!"
 
-    text_embedder = NvidiaTextEmbedder(model="nvidia/nv-embedqa-e5-v5", api_url="https://integrate.api.nvidia.com/v1")
+    text_embedder = NvidiaTextEmbedder(model="nvidia/nemotron-3-embed-1b", api_url="https://integrate.api.nvidia.com/v1")
     # Components warm up automatically on first run.
 
     print(text_embedder.run(text_to_embed))
@@ -80,8 +80,7 @@ class NvidiaTextEmbedder:
             truncate = EmbeddingTruncateMode.from_str(truncate)
         self.truncate = truncate
 
-        self.backend: Any | None = None
-        self._initialized = False
+        self.backend: NimBackend | None = None
 
         if timeout is None:
             timeout = float(os.environ.get("NVIDIA_TIMEOUT", "60.0"))
@@ -123,7 +122,7 @@ class NvidiaTextEmbedder:
         """
         Initializes the component.
         """
-        if self._initialized:
+        if self.backend is not None:
             return
 
         model_kwargs = {"input_type": "query"}
@@ -138,14 +137,18 @@ class NvidiaTextEmbedder:
             timeout=self.timeout,
             client=Client.NVIDIA_TEXT_EMBEDDER,
         )
-        self._initialized = True
-
         if not self.model:
             if self.backend.model:
                 self.model = self.backend.model
             else:
                 self.default_model()
 
+    def close(self) -> None:
+        """Close the backend and release its resources."""
+        if self.backend is not None:
+            self.backend.close()
+            self.backend = None
+
     def to_dict(self) -> dict[str, Any]:
         """
         Serializes the component to a dictionary.
@@ -171,21 +174,6 @@ class NvidiaTextEmbedder:
         """
         return self.backend.models() if self.backend else []
 
-    @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> "NvidiaTextEmbedder":
-        """
-        Deserializes the component from a dictionary.
-
-        :param data:
-            The dictionary to deserialize from.
-        :returns:
-            The deserialized component.
-        """
-        init_parameters = data.get("init_parameters", {})
-        if init_parameters:
-            deserialize_secrets_inplace(data["init_parameters"], keys=["api_key"])
-        return default_from_dict(cls, data)
-
     @component.output_types(embedding=list[float], meta=dict[str, Any])
     def run(self, text: str) -> dict[str, list[float] | dict[str, Any]]:
         """
@@ -202,8 +190,7 @@ class NvidiaTextEmbedder:
         :raises ValueError:
             If the input string is empty.
         """
-        if not self._initialized:
-            self.warm_up()
+        self.warm_up()
 
         if not isinstance(text, str):
             msg = (

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/generators/nvidia/__init__.py

@@ -3,6 +3,5 @@
 # SPDX-License-Identifier: Apache-2.0
 
 from .chat.chat_generator import NvidiaChatGenerator
-from .generator import NvidiaGenerator
 
-__all__ = ["NvidiaChatGenerator", "NvidiaGenerator"]
+__all__ = ["NvidiaChatGenerator"]

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/generators/nvidia/chat/chat_generator.py

@@ -53,7 +53,7 @@ class NvidiaChatGenerator(OpenAIChatGenerator):
         self,
         *,
         api_key: Secret = Secret.from_env_var("NVIDIA_API_KEY"),
-        model: str = "meta/llama-3.1-8b-instruct",
+        model: str = "nvidia/nemotron-3.5-lightning-30b-a3b",
         streaming_callback: StreamingCallbackT | None = None,
         api_base_url: str | None = os.getenv("NVIDIA_API_URL", DEFAULT_API_URL),
         generation_kwargs: dict[str, Any] | None = None,

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/rankers/nvidia/ranker.py

@@ -6,8 +6,8 @@ import os
 from dataclasses import replace
 from typing import Any
 
-from haystack import Document, component, default_from_dict, default_to_dict
-from haystack.utils import Secret, deserialize_secrets_inplace
+from haystack import Document, component, default_to_dict
+from haystack.utils import Secret
 
 from haystack_integrations.components.rankers.nvidia.truncate import RankerTruncateMode
 from haystack_integrations.utils.nvidia import DEFAULT_API_URL, Client, NimBackend, is_hosted, url_validation
@@ -25,7 +25,7 @@ class NvidiaRanker:
     from haystack.utils import Secret
 
     ranker = NvidiaRanker(
-        model="nvidia/nv-rerankqa-mistral-4b-v3",
+        model="nvidia/llama-nemotron-rerank-vl-1b-v2",
         api_key=Secret.from_env_var("NVIDIA_API_KEY"),
     )
     # Components warm up automatically on first run.
@@ -108,8 +108,7 @@ class NvidiaRanker:
         #  - assume we won't call backend.models()
         self.api_url = url_validation(api_url)
         self.top_k = top_k
-        self._initialized = False
-        self.backend: Any | None = None
+        self.backend: NimBackend | None = None
         self.is_hosted = is_hosted(api_url)
 
         self.query_prefix = query_prefix
@@ -145,42 +144,35 @@ class NvidiaRanker:
             timeout=self.timeout,
         )
 
-    @classmethod
-    def from_dict(cls, data: dict[str, Any]) -> "NvidiaRanker":
-        """
-        Deserialize the ranker from a dictionary.
-
-        :param data: A dictionary containing the ranker's attributes.
-        :returns: The deserialized ranker.
-        """
-        init_parameters = data.get("init_parameters", {})
-        if init_parameters:
-            deserialize_secrets_inplace(data["init_parameters"], keys=["api_key"])
-        return default_from_dict(cls, data)
-
     def warm_up(self) -> None:
         """
         Initialize the ranker.
 
         :raises ValueError: If the API key is required for hosted NVIDIA NIMs.
         """
-        if not self._initialized:
-            model_kwargs: dict[str, Any] = {}
-            if self.truncate is not None:
-                model_kwargs.update(truncate=str(self.truncate))
-            self.backend = NimBackend(
-                model=self.model,
-                model_type="ranking",
-                api_url=self.api_url,
-                api_key=self.api_key,
-                model_kwargs=model_kwargs,
-                timeout=self.timeout,
-                client=Client.NVIDIA_RANKER,
-            )
-            if not self.is_hosted and not self.model:
-                if self.backend.model:
-                    self.model = self.backend.model
-            self._initialized = True
+        if self.backend is not None:
+            return
+
+        model_kwargs: dict[str, Any] = {}
+        if self.truncate is not None:
+            model_kwargs.update(truncate=str(self.truncate))
+        self.backend = NimBackend(
+            model=self.model,
+            model_type="ranking",
+            api_url=self.api_url,
+            api_key=self.api_key,
+            model_kwargs=model_kwargs,
+            timeout=self.timeout,
+            client=Client.NVIDIA_RANKER,
+        )
+        if not self.is_hosted and not self.model and self.backend.model:
+            self.model = self.backend.model
+
+    def close(self) -> None:
+        """Close the backend and release its resources."""
+        if self.backend is not None:
+            self.backend.close()
+            self.backend = None
 
     def _prepare_documents_to_embed(self, documents: list[Document]) -> list[str]:
         document_texts = []
@@ -208,8 +200,7 @@ class NvidiaRanker:
 
         :returns: A dictionary containing the ranked documents.
         """
-        if not self._initialized:
-            self.warm_up()
+        self.warm_up()
 
         if not isinstance(query, str):
             msg = "NvidiaRanker expects the `query` parameter to be a string."

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/components/rankers/nvidia/truncate.py

@@ -5,7 +5,7 @@
 from enum import Enum
 
 
-class RankerTruncateMode(str, Enum):
+class RankerTruncateMode(str, Enum):  # noqa: UP042 # StrEnum would change str() of this public enum
     """
     Specifies how inputs to the NVIDIA ranker components are truncated.
 

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/utils/nvidia/models.py

@@ -308,6 +308,13 @@ CHAT_MODEL_TABLE = {
         supports_tools=True,
         supports_structured_output=True,
     ),
+    "nvidia/nemotron-3.5-lightning-30b-a3b": Model(
+        id="nvidia/nemotron-3.5-lightning-30b-a3b",
+        model_type="chat",
+        client=Client.NVIDIA_GENERATOR,
+        supports_tools=True,
+        supports_structured_output=True,
+    ),
     "meta/llama-3.1-405b-instruct": Model(
         id="meta/llama-3.1-405b-instruct",
         model_type="chat",
@@ -430,6 +437,10 @@ EMBEDDING_MODEL_TABLE = {
         id="nvidia/nv-embedqa-mistral-7b-v2",
         model_type="embedding",
     ),
+    "nvidia/nemotron-3-embed-1b": Model(
+        id="nvidia/nemotron-3-embed-1b",
+        model_type="embedding",
+    ),
     "nvidia/nv-embedqa-e5-v5": Model(
         id="nvidia/nv-embedqa-e5-v5",
         model_type="embedding",
@@ -452,6 +463,12 @@ EMBEDDING_MODEL_TABLE = {
     ),
 }
 RANKING_MODEL_TABLE = {
+    "nvidia/llama-nemotron-rerank-vl-1b-v2": Model(
+        id="nvidia/llama-nemotron-rerank-vl-1b-v2",
+        model_type="ranking",
+        client=Client.NVIDIA_RANKER,
+        endpoint="https://ai.api.nvidia.com/v1/retrieval/nvidia/llama-nemotron-rerank-vl-1b-v2/reranking",
+    ),
     "nv-rerank-qa-mistral-4b:1": Model(
         id="nv-rerank-qa-mistral-4b:1",
         model_type="ranking",
@@ -480,9 +497,9 @@ RANKING_MODEL_TABLE = {
 }
 
 DEFAULT_MODELS = {
-    "embedding": "nvidia/nv-embedqa-e5-v5",
-    "ranking": "nv-rerank-qa-mistral-4b:1",
-    "chat": "meta/llama3-8b-instruct",
+    "embedding": "nvidia/nemotron-3-embed-1b",
+    "ranking": "nvidia/llama-nemotron-rerank-vl-1b-v2",
+    "chat": "nvidia/nemotron-3.5-lightning-30b-a3b",
 }
 
 

{nvidia_haystack-1.4.0 → nvidia_haystack-3.0.0}/src/haystack_integrations/utils/nvidia/nim_backend.py

@@ -38,9 +38,6 @@ class NimBackend:
         if api_key:
             headers["authorization"] = f"Bearer {api_key.resolve_value()}"
 
-        self.session = requests.Session()
-        self.session.headers.update(headers)
-
         self.api_url = api_url
         # Hosted ranking models are assigned a per-model endpoint that already includes the
         # full path (e.g. ".../nv-rerankqa-mistral-4b-v3/reranking"), unlike other hosted
@@ -79,6 +76,12 @@ class NimBackend:
         if timeout is None:
             timeout = float(os.environ.get("NVIDIA_TIMEOUT", REQUEST_TIMEOUT))
         self.timeout = timeout
+        self.session = requests.Session()
+        self.session.headers.update(headers)
+
+    def close(self) -> None:
+        """Close the HTTP session owned by this backend."""
+        self.session.close()
 
     def embed(self, texts: list[str]) -> tuple[list[list[float]], dict[str, Any]]:
         """Compute embeddings for a list of texts via the NIM API."""
@@ -111,65 +114,6 @@ class NimBackend:
 
         return embeddings, {"usage": data["usage"]}
 
-    def generate(self, prompt: str) -> tuple[list[str], list[dict[str, Any]]]:
-        """Generate text completions for a prompt via the NIM API."""
-        # We're using the chat completion endpoint as the NIM API doesn't support
-        # the /completions endpoint. So both the non-chat and chat generator will use this.
-        # This is the same for local containers and the cloud API.
-        url = f"{self.api_url}/chat/completions"
-
-        try:
-            res = self.session.post(
-                url,
-                json={
-                    "model": self.model,
-                    "messages": [
-                        {
-                            "role": "user",
-                            "content": prompt,
-                        },
-                    ],
-                    **self.model_kwargs,
-                },
-                timeout=self.timeout,
-            )
-            res.raise_for_status()
-        except requests.HTTPError as e:
-            response_text = e.response.text.strip() if e.response is not None else ""
-            error_details = f": {response_text}" if response_text else ""
-            logger.error(
-                "Error when calling NIM chat completion endpoint{error}",
-                error=error_details,
-            )
-            msg = f"Failed to query chat completion endpoint{error_details}"
-            raise ValueError(msg) from e
-
-        completions = res.json()
-        choices = completions["choices"]
-        # Sort the choices by index, we don't know whether they're out of order or not
-        choices.sort(key=lambda c: c["index"])
-        replies = []
-        meta = []
-        for choice in choices:
-            message = choice["message"]
-            replies.append(message["content"])
-            choice_meta = {
-                "role": message["role"],
-                "usage": {
-                    "prompt_tokens": completions["usage"]["prompt_tokens"],
-                    "total_tokens": completions["usage"]["total_tokens"],
-                },
-            }
-            # These fields could be null, the others will always be present
-            if "finish_reason" in choice:
-                choice_meta["finish_reason"] = choice["finish_reason"]
-            if "completion_tokens" in completions["usage"]:
-                choice_meta["usage"]["completion_tokens"] = completions["usage"]["completion_tokens"]
-
-            meta.append(choice_meta)
-
-        return replies, meta
-
     def models(self) -> list[Model]:
         """Retrieve available models from the NIM API."""
         url = f"{self.api_url}/models"