Replaces gpt-oss on ModelConfig, qa.model and processing.title_model, and ministral-3 on the picture-description model. qa.model.vision follows the model and is now true. enable_thinking was gated on the gpt-oss name, so it did nothing for qwen3.8. With title_model's max_tokens of 100 the reasoning consumed the whole budget and title generation returned an empty string. The mapping now applies to any ollama model via reasoning_effort(): false sends "none", true sends "high". Measured on qwen3.8:27b-mlx, "low" does not disable thinking and "none" does; gpt-oss is the inverse, its template has no "none" level, so it keeps "low". Picture description bypasses get_model -- docling posts the request itself from a params dict -- so the flag was inert on that path too. vlm_api_params() carries reasoning_effort into both converters' request bodies. At max_tokens 200 the description survived either way, but the switch cut completion tokens from 141 to 45. test_search_tool_skips_binary_content_when_qa_model_is_text_only asserted the vision default rather than setting it; it now configures vision=False itself. docs/benchmarks.md keeps ministral-3: those are recorded measurements.
181 lines
5.6 KiB
Python
181 lines
5.6 KiB
Python
from contextlib import asynccontextmanager
|
|
from unittest.mock import AsyncMock, patch
|
|
|
|
import httpx
|
|
import pytest
|
|
|
|
from haiku.rag.client.downloads import download_models
|
|
from haiku.rag.config import get_config
|
|
from haiku.rag.config.models import ModelConfig
|
|
|
|
|
|
@pytest.fixture
|
|
def mock_to_thread():
|
|
"""Patch asyncio.to_thread to skip docling/tokenizer downloads."""
|
|
with patch("haiku.rag.client.downloads.asyncio.to_thread", new_callable=AsyncMock):
|
|
yield
|
|
|
|
|
|
@asynccontextmanager
|
|
async def _mock_httpx_client(stream_fn):
|
|
"""Create a mock httpx.AsyncClient context manager with a given stream function."""
|
|
mock_client = AsyncMock()
|
|
mock_client.stream = stream_fn
|
|
yield mock_client
|
|
|
|
|
|
async def test_download_models_ollama_connect_error(mock_to_thread):
|
|
"""When Ollama is not running, download_models raises ConnectionError."""
|
|
|
|
@asynccontextmanager
|
|
async def failing_stream(method, url, **kwargs):
|
|
raise httpx.ConnectError("All connection attempts failed")
|
|
yield # unreachable, but needed for generator syntax
|
|
|
|
with patch(
|
|
"haiku.rag.client.downloads.httpx.AsyncClient",
|
|
return_value=_mock_httpx_client(failing_stream),
|
|
):
|
|
with pytest.raises(
|
|
ConnectionError, match="Cannot connect to Ollama"
|
|
) as exc_info:
|
|
async for _ in download_models(get_config()):
|
|
pass
|
|
|
|
assert "ollama serve" in str(exc_info.value)
|
|
|
|
|
|
async def test_download_models_ollama_pulls_models(mock_to_thread):
|
|
"""download_models yields correct progress events for Ollama model pulls."""
|
|
stream_lines = [
|
|
'{"status": "pulling manifest"}',
|
|
"",
|
|
'{"status": "downloading", "digest": "sha256:abc", "total": 1000, "completed": 500}',
|
|
'{"status": "downloading", "digest": "sha256:abc", "total": 1000, "completed": 1000}',
|
|
"not valid json",
|
|
'{"status": "verifying sha256 digest"}',
|
|
'{"status": "writing manifest"}',
|
|
'{"status": "success"}',
|
|
]
|
|
|
|
@asynccontextmanager
|
|
async def mock_stream(method, url, **kwargs):
|
|
mock_resp = AsyncMock()
|
|
|
|
async def aiter_lines():
|
|
for line in stream_lines:
|
|
yield line
|
|
|
|
mock_resp.aiter_lines = aiter_lines
|
|
yield mock_resp
|
|
|
|
with patch(
|
|
"haiku.rag.client.downloads.httpx.AsyncClient",
|
|
return_value=_mock_httpx_client(mock_stream),
|
|
):
|
|
events = []
|
|
async for progress in download_models(get_config()):
|
|
events.append(progress)
|
|
|
|
# Default config has embeddings=qwen3-embedding:4b, qa=qwen3.8
|
|
ollama_models = {"qwen3.8", "qwen3-embedding:4b"}
|
|
ollama_events = [e for e in events if e.model in ollama_models]
|
|
pulling_events = [e for e in ollama_events if e.status == "pulling"]
|
|
done_events = [e for e in ollama_events if e.status == "done"]
|
|
download_events = [e for e in ollama_events if e.status == "downloading"]
|
|
|
|
assert len(pulling_events) == 2
|
|
assert len(done_events) == 2
|
|
assert len(download_events) > 0
|
|
|
|
for de in download_events:
|
|
assert de.digest == "sha256:abc"
|
|
assert de.total == 1000
|
|
assert de.completed > 0
|
|
|
|
|
|
async def test_download_models_no_ollama_models(mock_to_thread):
|
|
"""When no Ollama models are configured, no Ollama pull events are yielded."""
|
|
from haiku.rag.config import AppConfig
|
|
|
|
config = AppConfig()
|
|
config.embeddings.model.provider = "openai"
|
|
config.qa.model.provider = "openai"
|
|
|
|
events = []
|
|
async for progress in download_models(config):
|
|
events.append(progress)
|
|
|
|
models = {e.model for e in events}
|
|
assert "qwen3-embedding:4b" not in models
|
|
assert "qwen3.8" not in models
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"configure,expected_model",
|
|
[
|
|
(
|
|
lambda c: setattr(
|
|
c.reranking,
|
|
"model",
|
|
ModelConfig(provider="ollama", name="rerank-model"),
|
|
),
|
|
"rerank-model",
|
|
),
|
|
(
|
|
lambda c: (
|
|
setattr(c.processing, "pictures", "description"),
|
|
setattr(
|
|
c.processing.conversion_options.picture_description.model,
|
|
"provider",
|
|
"ollama",
|
|
),
|
|
setattr(
|
|
c.processing.conversion_options.picture_description.model,
|
|
"name",
|
|
"vision-model",
|
|
),
|
|
),
|
|
"vision-model",
|
|
),
|
|
(
|
|
lambda c: (
|
|
setattr(c.processing, "auto_title", True),
|
|
setattr(c.processing.title_model, "provider", "ollama"),
|
|
setattr(c.processing.title_model, "name", "title-model"),
|
|
),
|
|
"title-model",
|
|
),
|
|
],
|
|
ids=["reranker", "picture_description", "auto_title"],
|
|
)
|
|
async def test_ollama_models_from_every_config_slot_are_pulled(
|
|
mock_to_thread, configure, expected_model
|
|
):
|
|
"""Each config slot that can name an ollama model contributes to the pull set."""
|
|
from haiku.rag.config import AppConfig
|
|
|
|
config = AppConfig()
|
|
configure(config)
|
|
|
|
@asynccontextmanager
|
|
async def mock_stream(method, url, **kwargs):
|
|
mock_resp = AsyncMock()
|
|
|
|
async def aiter_lines():
|
|
yield '{"status": "success"}'
|
|
|
|
mock_resp.aiter_lines = aiter_lines
|
|
yield mock_resp
|
|
|
|
with patch(
|
|
"haiku.rag.client.downloads.httpx.AsyncClient",
|
|
return_value=_mock_httpx_client(mock_stream),
|
|
):
|
|
pulled = {
|
|
progress.model
|
|
async for progress in download_models(config)
|
|
if progress.status == "pulling"
|
|
}
|
|
|
|
assert expected_model in pulled
|