feat(providers): add Mistral embedding provider, route registry through dynaconf
Adds a hosted Mistral embedding option (mistral-embed, 1024-dim) alongside the existing Bedrock / OpenAI / Ollama / Simple providers. Implementation mirrors OpenAIProvider: lazy dimension detection with a known-models lookup, chunked batch requests, defensive index sort, and a 429-aware retry decorator. In the same change, ProviderRegistry switches from os.getenv to the dynaconf-backed Settings dataclass so all five providers share a single configuration path. config.py gains the previously-uncovered Bedrock keys, the new Mistral keys, the missing OPENAI_GENERATION_MODEL / OLLAMA_GENERATION_MODEL, and SIMPLE_EMBEDDING_DIMENSION. Auto-detection priority: Bedrock → OpenAI → Mistral → Ollama → Simple. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
Chris Coutinho
co-authored by
Claude Opus 4.7
Claude Opus 4.7
parent
61cadf7935
commit
3268a13d11
@@ -0,0 +1,198 @@
|
||||
"""Unit tests for Mistral provider."""
|
||||
|
||||
from unittest.mock import AsyncMock, MagicMock
|
||||
|
||||
import pytest
|
||||
|
||||
from nextcloud_mcp_server.providers.mistral import (
|
||||
BATCH_SIZE,
|
||||
MISTRAL_EMBEDDING_DIMENSIONS,
|
||||
MistralProvider,
|
||||
)
|
||||
|
||||
|
||||
def _make_data(embedding: list[float], index: int) -> MagicMock:
|
||||
"""Build a mock EmbeddingResponseData entry."""
|
||||
item = MagicMock()
|
||||
item.embedding = embedding
|
||||
item.index = index
|
||||
return item
|
||||
|
||||
|
||||
def _make_response(embeddings: list[list[float]]) -> MagicMock:
|
||||
"""Build a mock EmbeddingResponse with `embeddings` indexed in order."""
|
||||
response = MagicMock()
|
||||
response.data = [_make_data(emb, i) for i, emb in enumerate(embeddings)]
|
||||
return response
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def mock_mistral_client(mocker):
|
||||
"""Mock the Mistral SDK constructor."""
|
||||
mock_client = MagicMock()
|
||||
mock_client.embeddings = MagicMock()
|
||||
mocker.patch(
|
||||
"nextcloud_mcp_server.providers.mistral.Mistral", return_value=mock_client
|
||||
)
|
||||
return mock_client
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_embedding_single(mock_mistral_client):
|
||||
"""Single text embed: round-trip through SDK with correct kwargs."""
|
||||
mock_mistral_client.embeddings.create_async = AsyncMock(
|
||||
return_value=_make_response([[0.1, 0.2, 0.3]])
|
||||
)
|
||||
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
embedding = await provider.embed("hello world")
|
||||
|
||||
assert embedding == [0.1, 0.2, 0.3]
|
||||
mock_mistral_client.embeddings.create_async.assert_awaited_once_with(
|
||||
model="mistral-embed",
|
||||
inputs=["hello world"],
|
||||
)
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_embedding_batch_single_call(mock_mistral_client):
|
||||
"""Batch smaller than BATCH_SIZE issues a single API call."""
|
||||
mock_mistral_client.embeddings.create_async = AsyncMock(
|
||||
return_value=_make_response([[0.1, 0.2], [0.3, 0.4], [0.5, 0.6]])
|
||||
)
|
||||
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
embeddings = await provider.embed_batch(["a", "b", "c"])
|
||||
|
||||
assert embeddings == [[0.1, 0.2], [0.3, 0.4], [0.5, 0.6]]
|
||||
assert mock_mistral_client.embeddings.create_async.await_count == 1
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_embedding_batch_chunking(mock_mistral_client):
|
||||
"""Batches exceeding BATCH_SIZE are split into multiple API calls."""
|
||||
|
||||
# Each call returns one embedding per input it received; capture by side
|
||||
# effect so we can inspect lengths per chunk.
|
||||
def _side_effect(*, model, inputs, **_kwargs):
|
||||
return _make_response([[float(i)] for i in range(len(inputs))])
|
||||
|
||||
mock_mistral_client.embeddings.create_async = AsyncMock(side_effect=_side_effect)
|
||||
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
total = BATCH_SIZE * 2 + 5 # forces three chunks: 64, 64, 5 (with default)
|
||||
embeddings = await provider.embed_batch([f"text-{i}" for i in range(total)])
|
||||
|
||||
assert len(embeddings) == total
|
||||
assert mock_mistral_client.embeddings.create_async.await_count == 3
|
||||
# Verify the chunk sizes the SDK was actually called with.
|
||||
chunk_sizes = [
|
||||
len(call.kwargs["inputs"])
|
||||
for call in mock_mistral_client.embeddings.create_async.await_args_list
|
||||
]
|
||||
assert chunk_sizes == [BATCH_SIZE, BATCH_SIZE, 5]
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_embedding_batch_order_preserved(mock_mistral_client):
|
||||
"""Out-of-order index in response data is sorted before returning."""
|
||||
response = MagicMock()
|
||||
response.data = [
|
||||
_make_data([0.3, 0.3], 2),
|
||||
_make_data([0.1, 0.1], 0),
|
||||
_make_data([0.2, 0.2], 1),
|
||||
]
|
||||
mock_mistral_client.embeddings.create_async = AsyncMock(return_value=response)
|
||||
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
embeddings = await provider.embed_batch(["x", "y", "z"])
|
||||
|
||||
assert embeddings == [[0.1, 0.1], [0.2, 0.2], [0.3, 0.3]]
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_supports_capabilities(mock_mistral_client):
|
||||
"""Mistral provider advertises embeddings only."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
assert provider.supports_embeddings is True
|
||||
assert provider.supports_generation is False
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_generate_not_implemented(mock_mistral_client):
|
||||
"""generate() always raises NotImplementedError."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
with pytest.raises(NotImplementedError, match="does not support generation"):
|
||||
await provider.generate("test prompt")
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_get_dimension_known_model(mock_mistral_client):
|
||||
"""Known model: dimension available without an API call."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
assert provider.get_dimension() == MISTRAL_EMBEDDING_DIMENSIONS["mistral-embed"]
|
||||
mock_mistral_client.embeddings.create_async.assert_not_called()
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_get_dimension_unknown_model_detected(mock_mistral_client):
|
||||
"""Unknown model: dimension detected on first embed() call."""
|
||||
mock_mistral_client.embeddings.create_async = AsyncMock(
|
||||
return_value=_make_response([[0.1] * 768])
|
||||
)
|
||||
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="custom-mistral")
|
||||
|
||||
with pytest.raises(RuntimeError, match="not detected yet"):
|
||||
provider.get_dimension()
|
||||
|
||||
await provider.embed("test")
|
||||
assert provider.get_dimension() == 768
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_no_embeddings_disabled():
|
||||
"""Setting embedding_model=None disables the embedding capability."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model=None)
|
||||
assert provider.supports_embeddings is False
|
||||
|
||||
with pytest.raises(NotImplementedError, match="no embedding_model configured"):
|
||||
await provider.embed("test")
|
||||
with pytest.raises(NotImplementedError, match="no embedding_model configured"):
|
||||
await provider.embed_batch(["test"])
|
||||
with pytest.raises(NotImplementedError, match="no embedding_model configured"):
|
||||
provider.get_dimension()
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_empty_batch(mock_mistral_client):
|
||||
"""An empty batch returns [] without calling the API."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
assert await provider.embed_batch([]) == []
|
||||
mock_mistral_client.embeddings.create_async.assert_not_called()
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_close_no_error(mock_mistral_client):
|
||||
"""close() is best-effort and does not raise."""
|
||||
provider = MistralProvider(api_key="test-key", embedding_model="mistral-embed")
|
||||
# No __aexit__ on the mock by default → close() should silently no-op.
|
||||
await provider.close()
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
async def test_mistral_base_url_passed_to_sdk(mocker):
|
||||
"""base_url is forwarded as server_url to the Mistral SDK constructor."""
|
||||
mock_ctor = mocker.patch(
|
||||
"nextcloud_mcp_server.providers.mistral.Mistral", return_value=MagicMock()
|
||||
)
|
||||
|
||||
MistralProvider(
|
||||
api_key="test-key",
|
||||
embedding_model="mistral-embed",
|
||||
base_url="https://example.com/mistral",
|
||||
)
|
||||
mock_ctor.assert_called_once_with(
|
||||
api_key="test-key",
|
||||
server_url="https://example.com/mistral",
|
||||
)
|
||||
@@ -0,0 +1,130 @@
|
||||
"""Unit tests for ProviderRegistry — dynaconf-driven auto-detection."""
|
||||
|
||||
import pytest
|
||||
|
||||
from nextcloud_mcp_server.config import _reload_config
|
||||
from nextcloud_mcp_server.providers import (
|
||||
BedrockProvider,
|
||||
MistralProvider,
|
||||
OllamaProvider,
|
||||
OpenAIProvider,
|
||||
SimpleProvider,
|
||||
get_provider,
|
||||
reset_provider,
|
||||
)
|
||||
from nextcloud_mcp_server.providers.bedrock import BOTO3_AVAILABLE
|
||||
|
||||
|
||||
def _clear_provider_envs(monkeypatch: pytest.MonkeyPatch) -> None:
|
||||
"""Strip every provider-selection env var so each test starts clean."""
|
||||
for name in (
|
||||
"AWS_REGION",
|
||||
"AWS_ACCESS_KEY_ID",
|
||||
"AWS_SECRET_ACCESS_KEY",
|
||||
"BEDROCK_EMBEDDING_MODEL",
|
||||
"BEDROCK_GENERATION_MODEL",
|
||||
"OPENAI_API_KEY",
|
||||
"OPENAI_BASE_URL",
|
||||
"OPENAI_EMBEDDING_MODEL",
|
||||
"OPENAI_GENERATION_MODEL",
|
||||
"MISTRAL_API_KEY",
|
||||
"MISTRAL_BASE_URL",
|
||||
"MISTRAL_EMBEDDING_MODEL",
|
||||
"OLLAMA_BASE_URL",
|
||||
"OLLAMA_EMBEDDING_MODEL",
|
||||
"OLLAMA_GENERATION_MODEL",
|
||||
"OLLAMA_VERIFY_SSL",
|
||||
"SIMPLE_EMBEDDING_DIMENSION",
|
||||
):
|
||||
monkeypatch.delenv(name, raising=False)
|
||||
|
||||
|
||||
@pytest.fixture
|
||||
def clean_provider_env(monkeypatch):
|
||||
"""Reset provider singleton + dynaconf cache around each test."""
|
||||
_clear_provider_envs(monkeypatch)
|
||||
reset_provider()
|
||||
_reload_config()
|
||||
yield monkeypatch
|
||||
reset_provider()
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_falls_back_to_simple(clean_provider_env):
|
||||
"""No provider env set → SimpleProvider (with default dimension)."""
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, SimpleProvider)
|
||||
assert provider.get_dimension() == 384
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_picks_simple_with_custom_dimension(clean_provider_env):
|
||||
"""SIMPLE_EMBEDDING_DIMENSION flows through dynaconf to SimpleProvider."""
|
||||
clean_provider_env.setenv("SIMPLE_EMBEDDING_DIMENSION", "512")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, SimpleProvider)
|
||||
assert provider.get_dimension() == 512
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_picks_mistral_when_api_key_set(clean_provider_env):
|
||||
"""MISTRAL_API_KEY alone is enough to select MistralProvider."""
|
||||
clean_provider_env.setenv("MISTRAL_API_KEY", "test-key")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, MistralProvider)
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_picks_ollama_when_base_url_set(clean_provider_env, mocker):
|
||||
"""OLLAMA_BASE_URL selects OllamaProvider."""
|
||||
# OllamaProvider eagerly probes /api/tags in __init__; stub it out.
|
||||
mocker.patch(
|
||||
"nextcloud_mcp_server.providers.ollama.OllamaProvider._check_model_is_loaded"
|
||||
)
|
||||
clean_provider_env.setenv("OLLAMA_BASE_URL", "http://localhost:11434")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, OllamaProvider)
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_openai_wins_over_mistral_and_ollama(clean_provider_env):
|
||||
"""OpenAI takes priority when multiple provider env vars are set."""
|
||||
clean_provider_env.setenv("OPENAI_API_KEY", "openai-key")
|
||||
clean_provider_env.setenv("MISTRAL_API_KEY", "mistral-key")
|
||||
clean_provider_env.setenv("OLLAMA_BASE_URL", "http://localhost:11434")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, OpenAIProvider)
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_mistral_wins_over_ollama(clean_provider_env):
|
||||
"""Mistral takes priority over Ollama when both are configured."""
|
||||
clean_provider_env.setenv("MISTRAL_API_KEY", "mistral-key")
|
||||
clean_provider_env.setenv("OLLAMA_BASE_URL", "http://localhost:11434")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, MistralProvider)
|
||||
|
||||
|
||||
@pytest.mark.unit
|
||||
def test_registry_bedrock_wins_when_aws_region_set(clean_provider_env):
|
||||
"""AWS_REGION alone routes to Bedrock, even with other providers configured."""
|
||||
if not BOTO3_AVAILABLE:
|
||||
pytest.skip("boto3 not installed")
|
||||
|
||||
clean_provider_env.setenv("AWS_REGION", "us-east-1")
|
||||
clean_provider_env.setenv("OPENAI_API_KEY", "openai-key")
|
||||
clean_provider_env.setenv("MISTRAL_API_KEY", "mistral-key")
|
||||
_reload_config()
|
||||
|
||||
provider = get_provider()
|
||||
assert isinstance(provider, BedrockProvider)
|
||||
Reference in New Issue
Block a user