Datasets:

tuandunghcmut
/

luatcheck-embeddinggemma2-768d-indexes

Error code:   StreamingRowsError
Exception:    CastError
Message:      Couldn't cast
schema_version: int64
name: string
built_at: timestamp[s]
build_seconds: double
config_path: string
config: struct<name: string, corpora: list<item: struct<path: string, collection_id: string>>, chunking: str (... 866 chars omitted)
  child 0, name: string
  child 1, corpora: list<item: struct<path: string, collection_id: string>>
      child 0, item: struct<path: string, collection_id: string>
          child 0, path: string
          child 1, collection_id: string
  child 2, chunking: struct<max_tokens: int64, overlap_tokens: int64, max_chars: int64, overlap_chars: int64, contextual_ (... 82 chars omitted)
      child 0, max_tokens: int64
      child 1, overlap_tokens: int64
      child 2, max_chars: int64
      child 3, overlap_chars: int64
      child 4, contextual_title: bool
      child 5, special_tokens: int64
      child 6, title_token_cap: int64
      child 7, title_char_cap: int64
  child 3, retrieval: struct<embedding: struct<provider: string, model: string, device: string, batch_size: int64, max_seq (... 520 chars omitted)
      child 0, embedding: struct<provider: string, model: string, device: string, batch_size: int64, max_seq_length: int64, dt (... 181 chars omitted)
          child 0, provider: string
          child 1, model: string
          child 2, device: string
          child 3, batch_size: int64
          child 4, max_seq_length: int64
          child 5, dtype: string
          child 6, truncate_dim: int64
          child 7, normalize: bool
          
...
   child 3, build_seconds: double
          child 4, texts_per_second: double
          child 5, stores: struct<numpy: struct<backend: string, size_kb: double, write_seconds: double>, faiss: struct<backend (... 200 chars omitted)
              child 0, numpy: struct<backend: string, size_kb: double, write_seconds: double>
                  child 0, backend: string
                  child 1, size_kb: double
                  child 2, write_seconds: double
              child 1, faiss: struct<backend: string, size_kb: double, write_seconds: double>
                  child 0, backend: string
                  child 1, size_kb: double
                  child 2, write_seconds: double
              child 2, faiss_hnsw: struct<backend: string, size_kb: double, write_seconds: double>
                  child 0, backend: string
                  child 1, size_kb: double
                  child 2, write_seconds: double
              child 3, qdrant: struct<backend: string, size_kb: double, write_seconds: double>
                  child 0, backend: string
                  child 1, size_kb: double
                  child 2, write_seconds: double
pipeline_and_ui_verification: struct<retriever_class: string, search_law_top_articles: list<item: string>, read_law_resolved: bool (... 33 chars omitted)
  child 0, retriever_class: string
  child 1, search_law_top_articles: list<item: string>
      child 0, item: string
  child 2, read_law_resolved: bool
  child 3, alias_resolution_Dieu_25: bool
to
{'pipeline_and_ui_verification': {'retriever_class': Value('string'), 'search_law_top_articles': List(Value('string')), 'read_law_resolved': Value('bool'), 'alias_resolution_Dieu_25': Value('bool')}, 'benchmarks': {'bll2019_v1': {'benchmark': Value('string'), 'queries_evaluated': Value('int64'), 'methods': {'1_LuatCheck_BM25_TFIDF_RRF (current)': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64'), 'build_seconds': Value('float64')}, '2_LuatCheck_BM25_only': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64')}, '3_SQLite_FTS5_Keyword': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64')}, '4_EmbeddingGemma2_128d_Vector_NumPy': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_quer
...
e('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_query': Value('float64'), 'dimension': Value('int64')}, '9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_query': Value('float64'), 'dimension': Value('int64')}, '0_SQLite_FTS5_Full_61425_Corpus (reference)': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64'), 'corpus_size': Value('int64')}}, 'index_summary': {'index_dir': Value('string'), 'records': Value('int64'), 'chunks': Value('int64'), 'build_seconds': Value('float64'), 'texts_per_second': Value('float64'), 'stores': {'numpy': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'faiss': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'faiss_hnsw': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'qdrant': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}}}}}}
because column names don't match
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/utils.py", line 149, in get_rows_or_raise
                  return get_rows(
                      dataset=dataset,
                  ...<4 lines>...
                      column_names=column_names,
                  )
                File "/src/libs/libcommon/src/libcommon/utils.py", line 272, in decorator
                  return func(*args, **kwargs)
                File "/src/services/worker/src/worker/utils.py", line 129, in get_rows
                  rows_plus_one = list(itertools.islice(safe_iter(ds, dataset=dataset), rows_max_number + 1))
                File "/src/services/worker/src/worker/utils.py", line 544, in safe_iter
                  yield from ds.decode(False) if ds.features else ds
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 3052, in __iter__
                  for key, example in ex_iterable:
                                      ^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2559, in __iter__
                  for key, pa_table in self._iter_arrow():
                                       ~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 2584, in _iter_arrow
                  for key, pa_table in self.ex_iterable._iter_arrow():
                                       ~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 558, in _iter_arrow
                  for key, pa_table in iterator:
                                       ^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/iterable_dataset.py", line 440, in _iter_arrow
                  for key, pa_table in self.generate_tables_fn(**gen_kwags):
                                       ~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 361, in _generate_tables
                  self._cast_table(pa_table, json_field_paths=json_field_paths),
                  ~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
                File "/usr/local/lib/python3.14/site-packages/datasets/packaged_modules/json/json.py", line 132, in _cast_table
                  pa_table = table_cast(pa_table, self.info.features.arrow_schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2381, in table_cast
                  return cast_table_to_schema(table, schema)
                File "/usr/local/lib/python3.14/site-packages/datasets/table.py", line 2309, in cast_table_to_schema
                  raise CastError(
                  ...<3 lines>...
                  )
              datasets.table.CastError: Couldn't cast
              schema_version: int64
              name: string
              built_at: timestamp[s]
              build_seconds: double
              config_path: string
              config: struct<name: string, corpora: list<item: struct<path: string, collection_id: string>>, chunking: str (... 866 chars omitted)
                child 0, name: string
                child 1, corpora: list<item: struct<path: string, collection_id: string>>
                    child 0, item: struct<path: string, collection_id: string>
                        child 0, path: string
                        child 1, collection_id: string
                child 2, chunking: struct<max_tokens: int64, overlap_tokens: int64, max_chars: int64, overlap_chars: int64, contextual_ (... 82 chars omitted)
                    child 0, max_tokens: int64
                    child 1, overlap_tokens: int64
                    child 2, max_chars: int64
                    child 3, overlap_chars: int64
                    child 4, contextual_title: bool
                    child 5, special_tokens: int64
                    child 6, title_token_cap: int64
                    child 7, title_char_cap: int64
                child 3, retrieval: struct<embedding: struct<provider: string, model: string, device: string, batch_size: int64, max_seq (... 520 chars omitted)
                    child 0, embedding: struct<provider: string, model: string, device: string, batch_size: int64, max_seq_length: int64, dt (... 181 chars omitted)
                        child 0, provider: string
                        child 1, model: string
                        child 2, device: string
                        child 3, batch_size: int64
                        child 4, max_seq_length: int64
                        child 5, dtype: string
                        child 6, truncate_dim: int64
                        child 7, normalize: bool
                        
              ...
                 child 3, build_seconds: double
                        child 4, texts_per_second: double
                        child 5, stores: struct<numpy: struct<backend: string, size_kb: double, write_seconds: double>, faiss: struct<backend (... 200 chars omitted)
                            child 0, numpy: struct<backend: string, size_kb: double, write_seconds: double>
                                child 0, backend: string
                                child 1, size_kb: double
                                child 2, write_seconds: double
                            child 1, faiss: struct<backend: string, size_kb: double, write_seconds: double>
                                child 0, backend: string
                                child 1, size_kb: double
                                child 2, write_seconds: double
                            child 2, faiss_hnsw: struct<backend: string, size_kb: double, write_seconds: double>
                                child 0, backend: string
                                child 1, size_kb: double
                                child 2, write_seconds: double
                            child 3, qdrant: struct<backend: string, size_kb: double, write_seconds: double>
                                child 0, backend: string
                                child 1, size_kb: double
                                child 2, write_seconds: double
              pipeline_and_ui_verification: struct<retriever_class: string, search_law_top_articles: list<item: string>, read_law_resolved: bool (... 33 chars omitted)
                child 0, retriever_class: string
                child 1, search_law_top_articles: list<item: string>
                    child 0, item: string
                child 2, read_law_resolved: bool
                child 3, alias_resolution_Dieu_25: bool
              to
              {'pipeline_and_ui_verification': {'retriever_class': Value('string'), 'search_law_top_articles': List(Value('string')), 'read_law_resolved': Value('bool'), 'alias_resolution_Dieu_25': Value('bool')}, 'benchmarks': {'bll2019_v1': {'benchmark': Value('string'), 'queries_evaluated': Value('int64'), 'methods': {'1_LuatCheck_BM25_TFIDF_RRF (current)': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64'), 'build_seconds': Value('float64')}, '2_LuatCheck_BM25_only': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64')}, '3_SQLite_FTS5_Keyword': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64')}, '4_EmbeddingGemma2_128d_Vector_NumPy': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_quer
              ...
              e('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_query': Value('float64'), 'dimension': Value('int64')}, '9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'search_only_ms_per_query': Value('float64'), 'latency_ms_per_query': Value('float64'), 'dimension': Value('int64')}, '0_SQLite_FTS5_Full_61425_Corpus (reference)': {'Recall@1': Value('float64'), 'Recall@5': Value('float64'), 'Recall@10': Value('float64'), 'Recall@20': Value('float64'), 'MRR@10': Value('float64'), 'NDCG@10': Value('float64'), 'MAP@10': Value('float64'), 'latency_ms_per_query': Value('float64'), 'corpus_size': Value('int64')}}, 'index_summary': {'index_dir': Value('string'), 'records': Value('int64'), 'chunks': Value('int64'), 'build_seconds': Value('float64'), 'texts_per_second': Value('float64'), 'stores': {'numpy': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'faiss': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'faiss_hnsw': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}, 'qdrant': {'backend': Value('string'), 'size_kb': Value('float64'), 'write_seconds': Value('float64')}}}}}}
              because column names don't match

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

LuatCheck 128-d Vector & Hybrid Indexes (`google/embeddinggemma-2`)

Bộ chỉ mục vector 128 chiều (Matryoshka Representation Learning truncate_dim=128) kết hợp từ khóa (SQLite FTS5) được xây dựng bằng mô hình google/embeddinggemma-2 (nhánh văn bản 270M tham số, bfloat16) cho hệ thống LuatCheck.

Danh sách các bộ Index trong Repository

Thư mục Index	Tập văn bản gốc	Số bản ghi (Records)	Số đoạn (Chunks)	Các Vector Stores đi kèm
`labor_code_2019_gemma2_128d/`	`data/corpus/labor_code_2019_official.json`	220	231	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`labour_vi_v1_gemma2_128d/`	`data/corpus/labour_vi_v1.json`	458	531	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`hcmut_academic_v1_gemma2_128d/`	`data/corpus/hcmut_academic_v1.json`	143	206	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`zalo_legal_gemma2_128d/`	`data/benchmarks/zalo_legal/corpus_eval_hard2k.json`	1000	1644	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`labor_code_2019_gemma2_768d/`	`data/corpus/labor_code_2019_official.json`	220	231	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`labour_vi_v1_gemma2_768d/`	`data/corpus/labour_vi_v1.json`	458	531	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`hcmut_academic_v1_gemma2_768d/`	`data/corpus/hcmut_academic_v1.json`	143	206	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`
`zalo_legal_gemma2_768d/`	`data/benchmarks/zalo_legal/corpus_eval_hard2k.json`	1000	1644	`numpy` (numpy), `faiss` (faiss), `faiss_hnsw` (faiss), `qdrant` (qdrant) + `sqlite_fts5`

Kết quả thực nghiệm (Retrieval Benchmark)

bll2019_v1 (Bộ luật Lao động 2019 - 220 điều) (`n = 9` truy vấn)

Phương pháp	Recall@1 (%)	Recall@5 (%)	Recall@10 (%)	Recall@20 (%)	MRR@10 (%)	NDCG@10 (%)	MAP@10 (%)	Latency (ms/q)
`1_LuatCheck_BM25_TFIDF_RRF (current)`	77.78	100.00	100.00	100.00	94.44	93.34	89.26	1.86
`2_LuatCheck_BM25_only`	77.78	100.00	100.00	100.00	94.44	93.34	89.26	1.06
`3_SQLite_FTS5_Keyword`	77.78	94.44	100.00	100.00	94.44	93.14	88.89	3.66
`4_EmbeddingGemma2_128d_Vector_NumPy`	11.11	61.11	72.22	83.33	40.37	43.65	32.04	581.53
`5_EmbeddingGemma2_128d_Hybrid_RRF (FTS5+Vector)`	33.33	83.33	88.89	100.00	61.48	65.28	55.46	585.22
`6_EmbeddingGemma2_128d_Vector_Qdrant`	11.11	61.11	72.22	83.33	40.37	43.65	32.04	582.01
`7_EmbeddingGemma2_128d_Vector_FAISS_Flat`	11.11	61.11	72.22	83.33	40.37	43.65	32.04	581.94
`8_EmbeddingGemma2_128d_Vector_FAISS_HNSW`	11.11	61.11	72.22	83.33	40.37	43.65	32.04	581.85
`9_EmbeddingGemma2_768d_Vector_NumPy_768d`	44.44	77.78	100.00	100.00	67.96	71.25	59.55	618.09
`9_EmbeddingGemma2_768d_Hybrid_RRF_768d (FTS5+Vector)`	77.78	88.89	100.00	100.00	92.59	90.55	84.92	621.78
`9_EmbeddingGemma2_768d_Vector_FAISS_Flat_768d`	44.44	77.78	100.00	100.00	67.96	71.25	59.55	618.29
`9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d`	44.44	77.78	100.00	100.00	67.96	71.25	59.55	618.32

labour_vi_v1 (Đa văn bản Pháp luật Lao động - 458 điều) (`n = 45` truy vấn)

Phương pháp	Recall@1 (%)	Recall@5 (%)	Recall@10 (%)	Recall@20 (%)	MRR@10 (%)	NDCG@10 (%)	MAP@10 (%)	Latency (ms/q)
`1_LuatCheck_BM25_TFIDF_RRF (current)`	84.44	98.89	98.89	100.00	95.19	95.69	94.07	2.72
`2_LuatCheck_BM25_only`	80.00	98.89	98.89	98.89	92.41	93.33	90.72	1.59
`3_SQLite_FTS5_Keyword`	84.44	100.00	100.00	100.00	94.81	96.43	95.19	4.83
`4_EmbeddingGemma2_128d_Vector_NumPy`	41.11	67.78	86.67	92.22	55.87	62.66	54.63	20.27
`5_EmbeddingGemma2_128d_Hybrid_RRF (FTS5+Vector)`	66.67	91.11	95.56	98.89	80.44	84.05	79.86	25.12
`6_EmbeddingGemma2_128d_Vector_Qdrant`	41.11	67.78	86.67	92.22	55.87	62.66	54.63	20.67
`7_EmbeddingGemma2_128d_Vector_FAISS_Flat`	41.11	67.78	86.67	92.22	55.87	62.66	54.63	20.33
`8_EmbeddingGemma2_128d_Vector_FAISS_HNSW`	41.11	67.78	86.67	92.22	55.87	62.66	54.63	20.36
`9_EmbeddingGemma2_768d_Vector_NumPy_768d`	54.44	84.44	90.00	98.89	71.21	74.51	68.48	133.49
`9_EmbeddingGemma2_768d_Hybrid_RRF_768d (FTS5+Vector)`	75.56	97.78	100.00	100.00	87.91	90.12	86.14	138.34
`9_EmbeddingGemma2_768d_Vector_FAISS_Flat_768d`	54.44	84.44	90.00	98.89	71.21	74.51	68.48	133.55
`9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d`	54.44	84.44	90.00	98.89	71.21	74.51	68.48	133.59

hcmut_v1 (Quy chế Học vụ ĐHBK HCMUT - 143 mục) (`n = 45` truy vấn)

Phương pháp	Recall@1 (%)	Recall@5 (%)	Recall@10 (%)	Recall@20 (%)	MRR@10 (%)	NDCG@10 (%)	MAP@10 (%)	Latency (ms/q)
`1_LuatCheck_BM25_TFIDF_RRF (current)`	85.56	98.89	100.00	100.00	95.19	95.51	93.33	1.17
`2_LuatCheck_BM25_only`	86.67	98.89	98.89	100.00	95.19	95.16	93.15	1.07
`3_SQLite_FTS5_Keyword`	90.00	100.00	100.00	100.00	97.78	97.64	96.30	1.57
`4_EmbeddingGemma2_128d_Vector_NumPy`	40.00	74.44	88.89	88.89	57.71	64.63	56.15	20.27
`5_EmbeddingGemma2_128d_Hybrid_RRF (FTS5+Vector)`	65.56	91.11	92.22	95.56	80.44	81.59	76.72	21.86
`6_EmbeddingGemma2_128d_Vector_Qdrant`	40.00	74.44	88.89	88.89	57.71	64.63	56.15	20.61
`7_EmbeddingGemma2_128d_Vector_FAISS_Flat`	40.00	74.44	88.89	88.89	57.71	64.63	56.15	20.34
`8_EmbeddingGemma2_128d_Vector_FAISS_HNSW`	40.00	74.44	88.89	88.89	57.71	64.63	56.15	20.37
`9_EmbeddingGemma2_768d_Vector_NumPy_768d`	61.11	88.89	94.44	95.56	77.08	79.74	73.74	137.29
`9_EmbeddingGemma2_768d_Hybrid_RRF_768d (FTS5+Vector)`	77.78	95.56	96.67	98.89	89.63	90.29	87.24	138.88
`9_EmbeddingGemma2_768d_Vector_FAISS_Flat_768d`	61.11	88.89	94.44	95.56	77.08	79.74	73.74	137.34
`9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d`	61.11	88.89	94.44	95.56	77.08	79.74	73.74	137.37

zalo_legal_hard1k (GreenNode Zalo AI Legal - 1,000 Hard-Negative Pool, 818 queries) (`n = 818` truy vấn)

Phương pháp	Recall@1 (%)	Recall@5 (%)	Recall@10 (%)	Recall@20 (%)	MRR@10 (%)	NDCG@10 (%)	MAP@10 (%)	Latency (ms/q)
`1_LuatCheck_BM25_TFIDF_RRF (current)`	62.47	93.22	98.17	99.27	75.43	80.99	75.34	5.24
`2_LuatCheck_BM25_only`	58.68	91.63	96.64	98.41	72.62	78.50	72.53	1.85
`3_SQLite_FTS5_Keyword`	68.83	95.78	98.78	99.14	80.20	84.73	80.05	3.58
`4_EmbeddingGemma2_128d_Vector_NumPy`	39.98	69.01	80.75	88.33	53.61	60.01	53.35	8.52
`5_EmbeddingGemma2_128d_Hybrid_RRF (FTS5+Vector)`	59.90	91.26	96.27	99.27	73.34	78.83	73.05	12.12
`6_EmbeddingGemma2_128d_Vector_Qdrant`	39.98	69.01	80.75	88.33	53.61	60.01	53.35	8.99
`7_EmbeddingGemma2_128d_Vector_FAISS_Flat`	39.98	69.01	80.75	88.33	53.61	60.01	53.35	8.55
`8_EmbeddingGemma2_128d_Vector_FAISS_HNSW`	39.98	69.01	80.87	88.33	53.62	60.04	53.37	8.6
`9_EmbeddingGemma2_768d_Vector_NumPy_768d`	52.93	83.68	90.28	94.07	66.13	71.88	65.86	18.18
`9_EmbeddingGemma2_768d_Hybrid_RRF_768d (FTS5+Vector)`	66.99	94.56	98.53	99.39	78.92	83.68	78.72	21.77
`9_EmbeddingGemma2_768d_Vector_FAISS_Flat_768d`	52.93	83.68	90.28	94.07	66.13	71.88	65.86	18.26
`9_EmbeddingGemma2_768d_Vector_FAISS_HNSW_768d`	52.93	83.68	90.28	94.19	66.13	71.88	65.86	18.28
`0_SQLite_FTS5_Full_61425_Corpus (reference)`	56.11	83.44	88.45	92.11	68.07	72.98	67.88	152.24

Cách nạp và sử dụng trong LuatCheck

from src.indexing.builder import load_index

with load_index('data/index/labor_code_2019_gemma2_128d', stores=['numpy', 'qdrant']) as idx:
    searcher = idx.searcher('numpy')
    hits = searcher.search('thời gian thử việc tối đa trình độ cao đẳng trở lên', top_k=5, mode='hybrid')

Downloads last month: 142

Total file size:

99.5 MB

LuatCheck 128-d Vector & Hybrid Indexes (google/embeddinggemma-2)

Danh sách các bộ Index trong Repository

Kết quả thực nghiệm (Retrieval Benchmark)

bll2019_v1 (Bộ luật Lao động 2019 - 220 điều) (n = 9 truy vấn)

labour_vi_v1 (Đa văn bản Pháp luật Lao động - 458 điều) (n = 45 truy vấn)

hcmut_v1 (Quy chế Học vụ ĐHBK HCMUT - 143 mục) (n = 45 truy vấn)