Spaces:
Sleeping
Sleeping
File size: 3,167 Bytes
f79a242 | 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 | from pathlib import Path
from ragforge.context_budget import focused_context_budget
from ragforge.eval_metrics import missing_answer_match
from ragforge.schemas import Chunk, PipelineConfig, QueryPlan, SearchHit
ROOT = Path(__file__).resolve().parents[1]
def _hit(idx: int, source: str) -> SearchHit:
return SearchHit(
chunk=Chunk(id=f"c{idx}", text=("evidence " * (idx + 3)).strip(), source=source),
score=1.0 / (idx + 1),
dense_score=max(0.1, 0.9 - idx * 0.1),
sparse_score=max(0.0, 0.8 - idx * 0.08),
)
def _plan(task="fact_lookup", strategy="semantic") -> QueryPlan:
return QueryPlan(
route="documents",
knowledge_scope="corpus",
task_type=task,
retrieval_strategy=strategy,
web_relevance="irrelevant",
rewritten_query="test",
document_queries=["test"],
)
def test_v18_focused_context_budget_prunes_tail_but_keeps_safety_floor():
hits = [_hit(i, "a.md" if i == 0 else f"d{i}.md") for i in range(6)]
decision = focused_context_budget(hits, _plan(), PipelineConfig(top_k=6, use_context_pruning=True))
assert decision.used is True
assert decision.chunks_before == 6
assert decision.chunks_after == 3
assert decision.hits == hits[:3]
assert decision.tokens_est_after < decision.tokens_est_before
assert decision.reduction_ratio > 0
def test_v18_context_budget_does_not_prune_broad_tasks():
hits = [_hit(i, f"d{i}.md") for i in range(6)]
decision = focused_context_budget(
hits,
_plan(task="insight_synthesis", strategy="analytical"),
PipelineConfig(top_k=6, use_context_pruning=True),
)
assert decision.used is False
assert decision.hits == hits
assert decision.reason == "broad_or_multi_source_task"
def test_v18_context_budget_can_be_disabled():
hits = [_hit(i, f"d{i}.md") for i in range(6)]
decision = focused_context_budget(hits, _plan(), PipelineConfig(top_k=6, use_context_pruning=False))
assert decision.used is False
assert decision.reason == "disabled_by_user"
def test_v18_exact_grounded_absence_phrase_from_user_report_matches():
answer = (
"Based on the provided evidence, the retrieved documents do not mention any fee charged by OrbitPay "
"for opening a card dispute [D1]. Therefore, the supplied context is insufficient to answer the question."
)
assert missing_answer_match(answer, {"expected_missing_any": ["not specified"]})
def test_v18_evaluation_contains_context_budget_ablation_and_grounded_absence_fallback():
text = (ROOT / "src/ragforge/evaluation.py").read_text(encoding="utf-8")
assert "_context_budget_ablation" in text
assert 'result.trace.get("metrics", {}).get("grounded_absence", False)' in text
assert '"context_budget_ablation": context_budget_rows' in text
def test_v18_ui_exposes_context_budget_controls_and_table():
text = (ROOT / "src/ragforge/ui.py").read_text(encoding="utf-8")
assert "Focused context pruning (adaptive)" in text
assert 'with gr.Tab("Context budget")' in text
assert '"Context budget ablation": "context_budget_ablation"' in text
|