altair823
cf35f36f88
PR-2 of fb-41 multi-hop RAG. Decompose + retrieve + synthesize 3-stage
pipeline가 `opts.multi_hop=true` 일 때 dispatch. Dynamic decide loop
는 PR-3.
- `AskOpts.multi_hop: bool` 필드 추가 + `impl Default for AskOpts`
도입 (HOTFIXES 2026-05-07 의 known limitation 해소). 9 explicit
init site 모두 `multi_hop: false` 추가 — Default 도입으로 향후
`..Default::default()` 점진 migrate 가능.
- `RagPipeline::ask` 의 entry 에 dispatcher 한 줄
(`if opts.multi_hop { return self.ask_multi_hop(...) }`).
- `RagPipeline::ask_multi_hop` 신규 method. 1) decompose LLM call
→ JSON array of strings parse, 2) 각 sub-query 로 retrieve +
chunk_id dedup pool, 3) score gate / no-chunks 가드, 4)
pack_context (single-pass 와 helper 공유), 5) synthesize LLM
call w/ MULTI_HOP_SYNTHESIZE_SYSTEM_PROMPT, 6) citation extract
+ Answer build. `prompt_template_version` = "rag-multi-hop-v1"
로 stamp — eval `compare` 가 single-pass vs multi-hop 분리.
- Prompt const 신규: MULTI_HOP_DECOMPOSE_SYSTEM_PROMPT +
MULTI_HOP_DECOMPOSE_USER_TEMPLATE + MULTI_HOP_SYNTHESIZE_SYSTEM_PROMPT
+ PROMPT_TEMPLATE_VERSION_MULTI_HOP + MULTI_HOP_MAX_SUB_QUERIES_DEFAULT.
- `kebab_core::RefusalReason::MultiHopDecomposeFailed` variant 신규.
Cascade: kebab-store-sqlite `refusal_reason_label` + kebab-tui `ask
refusal render` exhaustive match 갱신.
- `parse_decompose_response` + `strip_markdown_json_fence` helper —
markdown code fence (```json / ```) strip + JSON array of strings
parse + trim + drop empty + cap at MULTI_HOP_MAX_SUB_QUERIES_DEFAULT.
None 반환 시 caller 가 `MultiHopDecomposeFailed` refusal.
Tests (55 passing total, 8 신규):
- 6 unit (parse_decompose_response 의 bare array / fence variants /
garbage / cap / trim 회귀 핀).
- 2 integration: `ask_multi_hop_dispatches_and_decompose_garbage_refuses`
(decompose garbage → MultiHopDecomposeFailed + 정확히 1 LLM call) +
`ask_with_multi_hop_false_keeps_single_pass_path` (회귀 핀, 기존
caller 자동 backwards-compat).
Happy-path multi-hop (decompose 성공 → synthesize) 의 integration
test 는 ScriptedLm helper 가 PR-3 의 decide loop 와 함께 도입될
때 같이 추가. 현 `MockLanguageModel` 는 canned single response 라
2-LLM-call sequence 핀 불가.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
48 lines
1.9 KiB
Rust
48 lines
1.9 KiB
Rust
//! `kb-app::ask` smoke tests.
|
|
//!
|
|
//! The pipeline's behavior is exhaustively covered by `kb-rag` tests
|
|
//! (which inject `MockLanguageModel` + `MockRetriever`). The kb-app
|
|
//! facade is a thin component wirer: it picks the retriever per
|
|
//! `opts.mode` and constructs an `OllamaLanguageModel`. Exercising
|
|
//! that wiring requires a real Ollama on `127.0.0.1:11434`, so this
|
|
//! test is `#[ignore]` by default — run with `cargo test -p kb-app
|
|
//! --test ask_smoke -- --ignored` against a live Ollama.
|
|
|
|
mod common;
|
|
|
|
use common::TestEnv;
|
|
|
|
/// Lexical-mode ask end-to-end. Requires a real Ollama on
|
|
/// `config.models.llm.endpoint` (default `127.0.0.1:11434`) running the
|
|
/// configured model. The pipeline body is otherwise covered by kb-rag's
|
|
/// integration tests; this just verifies the facade composes the
|
|
/// components correctly.
|
|
#[test]
|
|
#[ignore = "requires real Ollama on 127.0.0.1:11434"]
|
|
fn ask_lexical_smoke() {
|
|
let env = TestEnv::lexical_only();
|
|
kebab_app::ingest_with_config(env.config.clone(), env.scope(), true).unwrap();
|
|
|
|
let opts = kebab_app::AskOpts {
|
|
k: 5,
|
|
explain: false,
|
|
mode: kebab_core::SearchMode::Lexical,
|
|
temperature: Some(0.0),
|
|
seed: Some(0),
|
|
stream_sink: None,
|
|
history: Vec::new(),
|
|
conversation_id: None,
|
|
turn_index: None,
|
|
multi_hop: false,
|
|
};
|
|
// The fixture workspace contains "ownership" content; the model's
|
|
// citation behavior depends on its training, so we don't assert on
|
|
// grounded — only that the call returns a structurally-valid Answer.
|
|
let answer = kebab_app::ask_with_config(env.config.clone(), "ownership", opts)
|
|
.expect("ask returns Ok with a real Ollama backend");
|
|
// retrieval summary always populated, regardless of grounded path.
|
|
assert_eq!(answer.retrieval.mode, kebab_core::SearchMode::Lexical);
|
|
assert!(answer.retrieval.k >= 5);
|
|
assert!(answer.retrieval.trace_id.0.starts_with("ret_"));
|
|
}
|