nvidia-haystack 1.1.0__tar.gz → 1.2.0__tar.gz

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/CHANGELOG.md

@@ -1,5 +1,44 @@
 # Changelog
 
+## [integrations/nvidia-v1.1.1] - 2026-04-14
+
+### 🐛 Bug Fixes
+
+- Replace in-place dataclass mutations with dataclasses.replace() (#3112)
+
+### 🧪 Testing
+
+- Track test coverage for all integrations (#3065)
+- Add unit tests for Nvidia integration (#3162)
+
+
+## [integrations/nvidia-v1.1.0] - 2026-03-30
+
+### 🐛 Bug Fixes
+
+- Nvidia - fix structured output syntax (#3058)
+
+### 📚 Documentation
+
+- *(nvidia)* Remove explicit warm_up from examples (#2843)
+- Simplify pydoc configs (#2855)
+
+### 🧪 Testing
+
+- Test compatible integrations with python 3.14; update pyproject (#3001)
+
+### 🧹 Chores
+
+- Add ANN ruff ruleset to llama_cpp, llama_stack, mcp, meta_llama, mistral, mongodb_atlas, nvidia, ollama, openrouter, opensearch (#2991)
+- Enforce ruff docstring rules (D102/D103/D205/D209/D213/D417/D419) in integrations 21-30 (#3010)
+
+## [integrations/nvidia-v1.0.0] - 2026-01-13
+
+### 🧹 Chores
+
+- [**breaking**] Nvidia - drop Python 3.9 and use X|Y typing; fix default reranking model; improve tests (#2736)
+
+
 ## [integrations/nvidia-v0.5.0] - 2026-01-13
 
 ### 🚀 Features

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nvidia-haystack
-Version: 1.1.0
+Version: 1.2.0
 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

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/pyproject.toml

@@ -66,7 +66,8 @@ dependencies = [
 unit = 'pytest -m "not integration" {args:tests}'
 integration = 'pytest -m "integration" {args:tests}'
 all = 'pytest {args:tests}'
-cov-retry = 'pytest --cov=haystack_integrations --reruns 3 --reruns-delay 30 -x {args:tests}'
+unit-cov-retry = 'pytest --cov=haystack_integrations --reruns 3 --reruns-delay 30 -x -m "not integration" {args:tests}'
+integration-cov-append-retry = 'pytest --cov=haystack_integrations --cov-append --reruns 3 --reruns-delay 30 -x -m "integration" {args:tests}'
 
 types = """mypy -p haystack_integrations.components.embedders.nvidia \
 -p haystack_integrations.components.generators.nvidia \
@@ -152,6 +153,7 @@ ban-relative-imports = "parents"
 [tool.coverage.run]
 source = ["haystack_integrations"]
 branch = true
+relative_files = true
 parallel = false
 
 

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/src/haystack_integrations/components/embedders/nvidia/document_embedder.py

@@ -253,7 +253,7 @@ class NvidiaDocumentEmbedder:
         if not self._initialized:
             self.warm_up()
 
-        elif not isinstance(documents, list) or (documents and not isinstance(documents[0], Document)):
+        if not isinstance(documents, list) or (documents and not isinstance(documents[0], Document)):
             msg = (
                 "NvidiaDocumentEmbedder expects a list of Documents as input."
                 "In case you want to embed a string, please use the NvidiaTextEmbedder."

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/src/haystack_integrations/components/embedders/nvidia/text_embedder.py

@@ -205,13 +205,13 @@ class NvidiaTextEmbedder:
         if not self._initialized:
             self.warm_up()
 
-        elif not isinstance(text, str):
+        if not isinstance(text, str):
             msg = (
                 "NvidiaTextEmbedder expects a string as an input."
                 "In case you want to embed a list of Documents, please use the NvidiaDocumentEmbedder."
             )
             raise TypeError(msg)
-        elif not text:
+        if not text:
             msg = "Cannot embed an empty string."
             raise ValueError(msg)
 

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/src/haystack_integrations/components/generators/nvidia/generator.py

@@ -75,6 +75,12 @@ class NvidiaGenerator:
             Timeout for request calls, if not set it is inferred from the `NVIDIA_TIMEOUT` environment variable
             or set to 60 by default.
         """
+        warnings.warn(
+            "The `NvidiaGenerator` component is deprecated and will be removed in a future version. "
+            "Use `NvidiaChatGenerator` instead, which now also supports string inputs.",
+            FutureWarning,
+            stacklevel=2,
+        )
         self._model = model
         self.api_url = url_validation(api_url)
         self._api_key = api_key
@@ -132,7 +138,7 @@ class NvidiaGenerator:
 
         if not self.is_hosted and not self._model:
             if self.backend.model:
-                self.model = self.backend.model
+                self._model = self.backend.model
             else:
                 self.default_model()
 

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/src/haystack_integrations/components/rankers/nvidia/ranker.py

@@ -4,6 +4,7 @@
 
 import os
 import warnings
+from dataclasses import replace
 from typing import Any
 
 from haystack import Document, component, default_from_dict, default_to_dict, logging
@@ -236,11 +237,8 @@ class NvidiaRanker:
 
         # rank result is list[{index: int, logit: float}] sorted by logit
         sorted_indexes_and_scores = self.backend.rank(query_text=query_text, document_texts=document_texts)
-        sorted_documents = []
-        for item in sorted_indexes_and_scores[:top_k]:
-            # mutate (don't copy) the document because we're only updating the score
-            doc = documents[item["index"]]
-            doc.score = item["logit"]
-            sorted_documents.append(doc)
+        sorted_documents = [
+            replace(documents[item["index"]], score=item["logit"]) for item in sorted_indexes_and_scores[:top_k]
+        ]
 
         return {"documents": sorted_documents}

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_base_url.py

@@ -28,7 +28,8 @@ from haystack_integrations.components.rankers.nvidia import NvidiaRanker
 def test_base_url_invalid_not_hosted(base_url: str, component) -> None:
     with pytest.warns(UserWarning) as msg:
         component(api_url=base_url, model="x")
-    assert "you may have inference and listing issues" in str(msg[0].message)
+    user_warnings = [w for w in msg if issubclass(w.category, UserWarning)]
+    assert any("you may have inference and listing issues" in str(w.message) for w in user_warnings)
 
 
 @pytest.mark.parametrize(

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_document_embedder.py

@@ -354,6 +354,13 @@ class TestNvidiaDocumentEmbedder:
         with pytest.raises(TypeError, match="NvidiaDocumentEmbedder expects a list of Documents as input"):
             embedder.run(documents=list_integers_input)
 
+    def test_run_validates_input_without_prior_warm_up(self):
+        api_key = Secret.from_token("fake-api-key")
+        embedder = NvidiaDocumentEmbedder("nvidia/nv-embedqa-e5-v5", api_key=api_key)
+
+        with pytest.raises(TypeError, match="NvidiaDocumentEmbedder expects a list of Documents as input"):
+            embedder.run(documents="text")
+
     def test_run_empty_document(self, caplog):
         model = "nvidia/nv-embedqa-e5-v5"
         api_key = Secret.from_token("fake-api-key")

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_generator.py

@@ -193,7 +193,7 @@ class TestNvidiaGenerator:
     @pytest.mark.integration
     def test_run_integration_with_api_catalog(self):
         generator = NvidiaGenerator(
-            model="meta/llama3-8b-instruct",
+            model="meta/llama-3.1-8b-instruct",
             api_url="https://integrate.api.nvidia.com/v1",
             api_key=Secret.from_env_var("NVIDIA_API_KEY"),
             model_arguments={
@@ -230,3 +230,73 @@ class TestNvidiaGenerator:
         )
         with pytest.raises(ValueError):
             generator1.warm_up()
+
+    @pytest.mark.usefixtures("mock_local_models")
+    def test_warm_up_falls_back_to_default_model(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        generator = NvidiaGenerator(api_url="http://localhost:8080/v1")
+
+        with pytest.warns(UserWarning, match="Default model is set as:"):
+            generator.warm_up()
+
+        assert generator._model == "model1"
+        assert generator.backend.model == "model1"
+        assert generator.to_dict()["init_parameters"]["model"] == "model1"
+
+    def test_default_model_raises_when_no_valid_models(self, monkeypatch, requests_mock):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        requests_mock.get(
+            "http://localhost:8080/v1/models",
+            json={"data": [{"id": "derived-model", "object": "model", "root": "base-model"}]},
+        )
+        generator = NvidiaGenerator(api_url="http://localhost:8080/v1")
+
+        with pytest.raises(ValueError, match="No locally hosted model was found"):
+            generator.warm_up()
+
+    def test_warm_up_is_idempotent(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        generator = NvidiaGenerator("meta/llama-3.1-8b-instruct")
+        generator.warm_up()
+        backend = generator.backend
+        generator.warm_up()
+        assert generator.backend is backend
+
+    def test_available_models_without_backend(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        generator = NvidiaGenerator("meta/llama-3.1-8b-instruct")
+        assert generator.available_models == []
+
+    @pytest.mark.usefixtures("mock_local_models")
+    def test_available_models_with_backend(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        generator = NvidiaGenerator(model="model1", api_url="http://localhost:8080/v1")
+        generator.warm_up()
+        models = generator.available_models
+        assert len(models) == 1
+        assert models[0].id == "model1"
+
+    def test_from_dict(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        data = {
+            "type": "haystack_integrations.components.generators.nvidia.generator.NvidiaGenerator",
+            "init_parameters": {
+                "api_key": {"env_vars": ["NVIDIA_API_KEY"], "strict": True, "type": "env_var"},
+                "api_url": "https://my.url.com/v1",
+                "model": "meta/llama-3.1-8b-instruct",
+                "model_arguments": {"temperature": 0.5},
+            },
+        }
+        generator = NvidiaGenerator.from_dict(data)
+        assert generator._model == "meta/llama-3.1-8b-instruct"
+        assert generator.api_url == "https://my.url.com/v1"
+        assert generator._model_arguments == {"temperature": 0.5}
+
+    def test_run(self, monkeypatch, mock_local_chat_completion):  # noqa: ARG002
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        generator = NvidiaGenerator(model="model1", api_url="http://localhost:8080/v1")
+
+        result = generator.run(prompt="What is the answer?")
+
+        assert result["replies"] == ["Hello!", "How are you?"]
+        assert len(result["meta"]) == 2

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_nim_backend.py

@@ -233,6 +233,36 @@ class TestNimBackend:
                 timeout=60.0,
             )
 
+    def test_embed_raises_on_http_error(self, monkeypatch):
+        error_response = requests.Response()
+        error_response.status_code = 500
+        error_response._content = b"server exploded"
+        with patch("requests.sessions.Session.post", return_value=error_response):
+            monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+            backend = NimBackend(model="nvidia/nv-embedqa-e5-v5", api_url=DEFAULT_API_URL, client="NvidiaTextEmbedder")
+            with pytest.raises(ValueError, match="Failed to query embedding endpoint"):
+                backend.embed(texts=["a"])
+
+    def test_generate_raises_on_http_error(self, monkeypatch):
+        error_response = requests.Response()
+        error_response.status_code = 500
+        error_response._content = b"server exploded"
+        with patch("requests.sessions.Session.post", return_value=error_response):
+            monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+            backend = NimBackend(model="meta/llama3-8b-instruct", api_url=DEFAULT_API_URL, client="NvidiaGenerator")
+            with pytest.raises(ValueError, match="Failed to query chat completion endpoint"):
+                backend.generate(prompt="hi")
+
+    def test_models_raises_when_empty(self, monkeypatch):
+        empty_response = requests.Response()
+        empty_response.status_code = 200
+        empty_response._content = json.dumps({"data": []}).encode()
+        with patch("requests.sessions.Session.get", return_value=empty_response):
+            monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+            backend = NimBackend(model="custom-model", api_url="http://localhost:8000")
+            with pytest.raises(ValueError, match="No hosted model were found"):
+                backend.models()
+
     def test_rank(self, monkeypatch):
         with patch("requests.sessions.Session.post", side_effect=mock_rank_post_response) as mock_post:
             monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
@@ -258,3 +288,27 @@ class TestNimBackend:
                 },
                 timeout=60.0,
             )
+
+    def test_rank_raises_on_http_error(self, monkeypatch):
+        error_response = requests.Response()
+        error_response.status_code = 500
+        error_response._content = b"server exploded"
+        with patch("requests.sessions.Session.post", return_value=error_response):
+            monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+            backend = NimBackend(
+                model="nvidia/llama-3.2-nv-rerankqa-1b-v2", api_url=DEFAULT_API_URL, client="NvidiaRanker"
+            )
+            with pytest.raises(ValueError, match="Failed to rank endpoint"):
+                backend.rank(query_text="q", document_texts=["a"])
+
+    def test_rank_raises_when_rankings_missing(self, monkeypatch):
+        response = requests.Response()
+        response.status_code = 200
+        response._content = json.dumps({"unexpected": "payload"}).encode()
+        with patch("requests.sessions.Session.post", return_value=response):
+            monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+            backend = NimBackend(
+                model="nvidia/llama-3.2-nv-rerankqa-1b-v2", api_url=DEFAULT_API_URL, client="NvidiaRanker"
+            )
+            with pytest.raises(ValueError, match="Expected 'rankings' in response"):
+                backend.rank(query_text="q", document_texts=["a"])

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_ranker.py

@@ -330,6 +330,20 @@ class TestNvidiaRanker:
         client.warm_up()
         assert client.backend.timeout == 45.0
 
+    def test_run_on_empty_list(self, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        client = NvidiaRanker()
+        client.warm_up()
+        assert client.run(query="q", documents=[]) == {"documents": []}
+
+    def test_run_without_prior_warm_up(self, requests_mock, monkeypatch):
+        monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
+        requests_mock.post(re.compile(r".*ranking"), json={"rankings": [{"index": 0, "logit": 1.0}]})
+        client = NvidiaRanker()
+        result = client.run(query="q", documents=[Document(content="doc")])
+        assert client._initialized is True
+        assert len(result["documents"]) == 1
+
     def test_prepare_texts_to_embed_w_metadata(self):
         documents = [
             Document(content=f"document number {i}:\ncontent", meta={"meta_field": f"meta_value {i}"}) for i in range(5)

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_text_embedder.py

@@ -183,6 +183,20 @@ class TestNvidiaTextEmbedder:
         with pytest.raises(ValueError, match="empty string"):
             embedder.run(text="")
 
+    def test_run_validates_input_without_prior_warm_up(self):
+        api_key = Secret.from_token("fake-api-key")
+        embedder = NvidiaTextEmbedder("nvidia/nv-embedqa-e5-v5", api_key=api_key)
+
+        with pytest.raises(TypeError, match="NvidiaTextEmbedder expects a string as an input"):
+            embedder.run(text=[1, 2, 3])
+
+    def test_run_rejects_empty_string_without_prior_warm_up(self):
+        api_key = Secret.from_token("fake-api-key")
+        embedder = NvidiaTextEmbedder("nvidia/nv-embedqa-e5-v5", api_key=api_key)
+
+        with pytest.raises(ValueError, match="empty string"):
+            embedder.run(text="")
+
     def test_setting_timeout(self, monkeypatch):
         monkeypatch.setenv("NVIDIA_API_KEY", "fake-api-key")
         embedder = NvidiaTextEmbedder(timeout=10.0)

{nvidia_haystack-1.1.0 → nvidia_haystack-1.2.0}/tests/test_utils.py

@@ -5,7 +5,12 @@
 import pytest
 
 from haystack_integrations.utils.nvidia import Client, is_hosted
-from haystack_integrations.utils.nvidia.models import CHAT_MODEL_TABLE, EMBEDDING_MODEL_TABLE, RANKING_MODEL_TABLE
+from haystack_integrations.utils.nvidia.models import (
+    CHAT_MODEL_TABLE,
+    EMBEDDING_MODEL_TABLE,
+    RANKING_MODEL_TABLE,
+    Model,
+)
 from haystack_integrations.utils.nvidia.utils import (
     determine_model,
     lookup_model,
@@ -110,3 +115,39 @@ def test_validate_hosted_model_with_client() -> None:
     model = validate_hosted_model("meta/codellama-70b", Client.NVIDIA_GENERATOR)
     assert model is not None
     assert model.client == Client.NVIDIA_GENERATOR
+
+
+# Model
+def test_model_hash_uses_id() -> None:
+    assert hash(Model(id="foo")) == hash("foo")
+
+
+def test_model_validate_with_enum_client() -> None:
+    model = Model(id="foo", model_type="chat", client=Client.NVIDIA_GENERATOR)
+    assert model.validate() == hash("foo")
+
+
+def test_model_validate_with_string_client() -> None:
+    model = Model(id="foo", model_type="chat", client="NvidiaGenerator")
+    assert model.validate() == hash("foo")
+
+
+def test_model_validate_raises_on_incompatible_type() -> None:
+    model = Model(id="foo", model_type="embedding", client=Client.NVIDIA_GENERATOR)
+    with pytest.raises(ValueError, match="not supported by client"):
+        model.validate()
+
+
+def test_model_validate_without_client() -> None:
+    model = Model(id="foo", model_type="chat")
+    assert model.validate() == hash("foo")
+
+
+# Client
+def test_client_from_str_invalid() -> None:
+    with pytest.raises(ValueError, match="Unknown client"):
+        Client.from_str("NotARealClient")
+
+
+def test_client_str_returns_value() -> None:
+    assert str(Client.NVIDIA_GENERATOR) == "NvidiaGenerator"