Spaces:
Running
Running
File size: 13,590 Bytes
65d5480 5be9616 65d5480 5be9616 65d5480 5be9616 77bdc95 5be9616 77bdc95 5be9616 77bdc95 5be9616 77bdc95 5be9616 c5573d3 5be9616 c5573d3 65d5480 c5573d3 65d5480 c5573d3 65d5480 c5573d3 65d5480 12a17f8 c5573d3 12a17f8 65d5480 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 | """Tests for RAG pipeline: chunker, embedder, store, retriever, and reranker."""
from __future__ import annotations
import numpy as np
import pytest
from agent_bench.rag.chunker import Chunk, chunk_fixed, chunk_recursive, chunk_text
from agent_bench.rag.embedder import Embedder
from agent_bench.rag.reranker import CrossEncoderReranker
from agent_bench.rag.retriever import Retriever
from agent_bench.rag.store import HybridStore, SearchResult
# --- Chunker tests ---
class TestChunker:
SAMPLE_TEXT = (
"FastAPI is a modern web framework.\n\n"
"It is based on standard Python type hints.\n\n"
"Path parameters are declared in the URL path using curly braces. "
"You can specify their types using annotations.\n\n"
"Query parameters are parsed automatically from the query string. "
"They support default values and optional types.\n\n"
"Request bodies use Pydantic models for validation."
)
def test_recursive_within_size_limits(self):
chunk_size = 100
overlap = 64
chunks = chunk_recursive(
self.SAMPLE_TEXT, "test.md", chunk_size=chunk_size, chunk_overlap=overlap
)
for c in chunks:
# Overlap prepend may push up to overlap chars beyond chunk_size
assert len(c.content) <= chunk_size + overlap + 1, (
f"Chunk too long: {len(c.content)} chars"
)
assert len(chunks) > 1
def test_fixed_within_size_limits(self):
chunks = chunk_fixed(self.SAMPLE_TEXT, "test.md", chunk_size=100, chunk_overlap=20)
for c in chunks:
assert len(c.content) <= 100
assert len(chunks) > 1
def test_recursive_preserves_text(self):
"""Every word in the source should appear in at least one chunk."""
chunks = chunk_recursive(self.SAMPLE_TEXT, "test.md", chunk_size=200)
all_words = set(self.SAMPLE_TEXT.split())
chunk_words = set()
for c in chunks:
chunk_words.update(c.content.split())
assert all_words.issubset(chunk_words)
def test_fixed_preserves_text_coverage(self):
"""Every word in the source should appear in at least one chunk."""
chunks = chunk_fixed(self.SAMPLE_TEXT, "test.md", chunk_size=100, chunk_overlap=20)
all_words = set(self.SAMPLE_TEXT.split())
chunk_words = set()
for c in chunks:
chunk_words.update(c.content.split())
assert all_words.issubset(chunk_words)
def test_chunk_source_is_bare_filename(self):
chunks = chunk_text(self.SAMPLE_TEXT, "fastapi_intro.md", strategy="recursive")
for c in chunks:
assert c.source == "fastapi_intro.md"
assert "/" not in c.source
def test_chunk_text_dispatcher(self):
rec = chunk_text(self.SAMPLE_TEXT, "t.md", strategy="recursive", chunk_size=200)
fix = chunk_text(self.SAMPLE_TEXT, "t.md", strategy="fixed", chunk_size=200)
assert all(c.metadata.get("strategy") == "recursive" for c in rec)
assert all(c.metadata.get("strategy") == "fixed" for c in fix)
def test_empty_text(self):
assert chunk_recursive("", "empty.md") == []
assert chunk_fixed("", "empty.md") == []
# --- Embedder tests ---
class TestEmbedder:
def test_embed_produces_correct_shape(self, mock_embedder: Embedder):
vec = mock_embedder.embed("test sentence")
assert vec.shape == (384,)
def test_embed_is_normalized(self, mock_embedder: Embedder):
vec = mock_embedder.embed("test sentence")
norm = np.linalg.norm(vec)
assert norm == pytest.approx(1.0, abs=1e-5)
def test_embed_batch_shape(self, mock_embedder: Embedder):
vecs = mock_embedder.embed_batch(["sentence one", "sentence two", "sentence three"])
assert vecs.shape == (3, 384)
def test_cache_hit_skips_model(self, mock_embedding_model, tmp_path):
"""Second embed() call for same text should use cache, not model."""
embedder = Embedder(model=mock_embedding_model, cache_dir=str(tmp_path))
_ = embedder.embed("cache test")
calls_after_first = mock_embedding_model.call_count
_ = embedder.embed("cache test")
assert mock_embedding_model.call_count == calls_after_first
def test_different_texts_produce_different_embeddings(self, mock_embedder: Embedder):
v1 = mock_embedder.embed("path parameters")
v2 = mock_embedder.embed("query parameters")
assert not np.allclose(v1, v2)
# --- Store tests ---
class TestHybridStore:
def test_add_and_semantic_search(self, test_store: HybridStore, mock_embedder: Embedder):
"""Semantic search returns relevant result for a known query."""
query_vec = mock_embedder.embed("path parameters curly braces")
results = test_store.search(
query_embedding=query_vec,
query_text="path parameters curly braces",
top_k=3,
strategy="semantic",
)
assert len(results) > 0
assert all(isinstance(r, SearchResult) for r in results)
# Should have scores and ranks
assert results[0].rank == 1
assert results[0].retrieval_strategy == "semantic"
def test_keyword_search(self, test_store: HybridStore, mock_embedder: Embedder):
"""BM25 keyword search finds chunks with matching terms."""
query_vec = mock_embedder.embed("Pydantic models validation")
results = test_store.search(
query_embedding=query_vec,
query_text="Pydantic models validation",
top_k=3,
strategy="keyword",
)
assert len(results) > 0
# Top result should be the request body chunk (mentions Pydantic)
assert "Pydantic" in results[0].chunk.content
def test_hybrid_returns_results_from_both(
self, test_store: HybridStore, mock_embedder: Embedder
):
"""RRF hybrid search returns results — both dense and sparse contribute."""
query_vec = mock_embedder.embed("path parameters FastAPI")
results = test_store.search(
query_embedding=query_vec,
query_text="path parameters FastAPI",
top_k=5,
strategy="hybrid",
)
assert len(results) > 0
assert all(r.retrieval_strategy == "hybrid" for r in results)
# RRF scores should be positive and sorted descending
for i in range(len(results) - 1):
assert results[i].score >= results[i + 1].score
def test_empty_store(self):
store = HybridStore(dimension=384)
dummy_vec = np.random.randn(384).astype(np.float32)
results = store.search(
query_embedding=dummy_vec, query_text="test", top_k=5, strategy="hybrid"
)
assert results == []
def test_save_load_roundtrip(self, test_store: HybridStore, mock_embedder: Embedder, tmp_path):
"""Save and load preserves all data and produces same search results."""
store_path = tmp_path / "test_store"
# Search before save
query_vec = mock_embedder.embed("path parameters")
results_before = test_store.search(
query_embedding=query_vec,
query_text="path parameters",
top_k=3,
strategy="hybrid",
)
# Save and reload
test_store.save(store_path)
loaded = HybridStore.load(store_path, rrf_k=60)
# Stats match
assert loaded.stats().total_chunks == test_store.stats().total_chunks
assert loaded.stats().faiss_index_size == test_store.stats().faiss_index_size
# Search after load
results_after = loaded.search(
query_embedding=query_vec,
query_text="path parameters",
top_k=3,
strategy="hybrid",
)
assert len(results_after) == len(results_before)
assert [r.chunk.id for r in results_after] == [r.chunk.id for r in results_before]
def test_stats(self, test_store: HybridStore):
stats = test_store.stats()
assert stats.total_chunks == 5
assert stats.faiss_index_size == 5
assert stats.unique_sources == 4 # 4 unique source files in sample chunks
# --- Retriever tests ---
class TestRetriever:
@pytest.mark.asyncio
async def test_search_returns_results(self, test_retriever: Retriever):
result = await test_retriever.search("path parameters", top_k=3)
assert len(result.results) > 0
assert all(isinstance(r, SearchResult) for r in result.results)
@pytest.mark.asyncio
async def test_search_strategy_override(self, test_retriever: Retriever):
result = await test_retriever.search("Pydantic models", top_k=3, strategy="keyword")
assert len(result.results) > 0
assert all(r.retrieval_strategy == "keyword" for r in result.results)
# --- Reranker tests ---
class MockCrossEncoder:
"""Mock cross-encoder that returns deterministic scores based on content length."""
def predict(self, pairs: list[tuple[str, str]]) -> list[float]:
# Score based on content length — longer content scores higher
# This gives a deterministic, predictable reordering
return [float(len(content)) for _, content in pairs]
class TestCrossEncoderReranker:
def _make_chunks(self, contents: list[str]) -> list[Chunk]:
return [
Chunk(id=f"c{i}", content=c, source=f"doc_{i}.md", chunk_index=0)
for i, c in enumerate(contents)
]
def test_reranker_reorders(self):
"""Reranker reorders chunks by cross-encoder score."""
chunks = self._make_chunks(["short", "a medium length chunk", "longest chunk content here"])
reranker = CrossEncoderReranker(model=MockCrossEncoder())
result = reranker.rerank("test query", chunks, top_k=3)
# MockCrossEncoder scores by content length, so longest first
assert result[0][0].content == "longest chunk content here"
assert result[1][0].content == "a medium length chunk"
assert result[2][0].content == "short"
def test_reranker_top_k(self):
"""Reranker returns exactly top_k results from a larger input."""
chunks = self._make_chunks([f"content {i}" for i in range(20)])
reranker = CrossEncoderReranker(model=MockCrossEncoder())
result = reranker.rerank("test query", chunks, top_k=5)
assert len(result) == 5
def test_reranker_disabled(self, mock_embedder: Embedder, test_store: HybridStore):
"""Retriever without reranker preserves RRF order."""
retriever_no_reranker = Retriever(embedder=mock_embedder, store=test_store)
retriever_with_none = Retriever(
embedder=mock_embedder, store=test_store, reranker=None,
)
import asyncio
results_a = asyncio.get_event_loop().run_until_complete(
retriever_no_reranker.search("path parameters", top_k=3)
)
results_b = asyncio.get_event_loop().run_until_complete(
retriever_with_none.search("path parameters", top_k=3)
)
assert [r.chunk.id for r in results_a.results] == [r.chunk.id for r in results_b.results]
def test_reranker_empty_input(self):
"""Empty chunk list returns empty list."""
reranker = CrossEncoderReranker(model=MockCrossEncoder())
result = reranker.rerank("test query", [], top_k=5)
assert result == []
@pytest.mark.asyncio
async def test_reranked_results_preserve_rrf_scores(
self, mock_embedder: Embedder, test_store: HybridStore,
):
"""Reranked results carry original RRF scores, not 0.0.
This is critical: the refusal gate in SearchTool checks max_score
from the returned results. If reranking zeroes out scores, the
refusal gate would reject every reranked query.
"""
reranker = CrossEncoderReranker(model=MockCrossEncoder())
retriever = Retriever(
embedder=mock_embedder,
store=test_store,
reranker=reranker,
reranker_top_k=3,
)
result = await retriever.search("path parameters", top_k=3)
assert len(result.results) > 0
# All scores must be positive (preserved from RRF), not 0.0
scores = [r.score for r in result.results]
assert all(r.score > 0 for r in result.results), (
f"Reranked scores should be positive RRF scores, got: {scores}"
)
@pytest.mark.asyncio
async def test_refusal_with_reranker_enabled(self):
"""Integration: out-of-scope query with reranker on still refuses.
The refusal gate fires on RRF max_score BEFORE reranking (go/no-go
decision). This test validates the Feature 1 + Feature 2 interaction.
"""
from agent_bench.tools.search import SearchTool
from tests.test_tools import MockChunk, MockRetriever, MockSearchResult
# Low scores — should trigger refusal regardless of reranker
low_score_results = [
MockSearchResult(
chunk=MockChunk(content="Unrelated content", source="irrelevant.md"),
score=0.005,
),
]
retriever = MockRetriever(results=low_score_results)
tool = SearchTool(retriever=retriever, refusal_threshold=0.02)
result = await tool.execute(query="how to cook pasta")
assert result.metadata["refused"] is True
assert "No relevant documents found" in result.result
|