Add More Grammars to `outlines.grammars`, Benchmark and Verify Their Integrity #587

lapp0 · 2024-01-25T19:04:52Z

Overview:

Adds grammars to outlines.grammars and fixes some grammars
Adds benchmarks for CFGFSM (pytest --benchmark-cfg)
Fixes UnexpectedToken error resulting from a partially complete terminal being lexed as a full terminal (see fsm.py)

There are a lot of files changed in this PR, but they're mostly new lark grammars, lark grammar improvements, and many sample test files.

Python changes:
- Add test cases
- update outlines/grammars.py with new grammars
- Prevent UnexpectedToken if terminal is partially generated and misrecognized as being a different terminal in fsm.py
Lark changes:
- Fix grammars
- Add grammars
Documentation:
- Add documentation on creating grammars (Rendered)
- Add documentation on using outlines.grammars
Add test generation files to outlines/tests/benchmark/cfg_samples/

Fixes:

Fixes Run-time structured generation benchmarks #549 (PR introduces full FSM benchmarks)
Fixes Handle regexps with lookaround #456 (PR fixes ESCAPED_STRING in common.lark and documents lookaround behavior)
Fixes Error when trying to run the grammar example with Mistral #580 (PR updates arithmetic.lark to not result in deep recursion)
Fixes Add SQL grammar #633
Fixes Add BNF grammar #636
Fixes Add JSON grammar #634
Fixes Update readme link to grammars #663

Prerequisites

allow generation of %ignore terminals #570 (Necessary for correctness of grammars)
Add grammars module #562 (Necessary infrastructure for using outlines.grammars)

Create / Fix Grammars

outlines.grammars.csv
Fix ESCAPED_STRING in outlines.grammars.json
outlines.grammars.lark
outlines.grammars.sql_select (maybe reach out to https://motherduck.com/blog/duckdb-text2sql-llm/)
remove outlines.grammars.lisp
fix outlines.grammars.arithmetic recursion issue

Other Work

End to end + benchmark tests for grammars
Enable some lookarounds by changing how CFGFSM.regex_fsm is constructed in outlines/fsm/fsm.py
Fix ESCAPED_STRING in common.lark so it's compatible with interegular
Rename all test files e.g. foo.py.test so it doesn't confuse users searching the repo
test / bench samples for outlines.grammars.arithmetic
Document lookaround behavior for https://github.com/MegaIng/interegular/

Documentation Work

~~[ ] Documentation and examples of grammars in use~~
- Will not do, I'd rather CFGFSM be LALR(1) per Make CFGFSM LALR(1) #588 before showing users sql_select
Documentation of how to create grammars

Out of scope because they require the introduction of context-sensitive indentation handling #592

outlines.grammars.python3 (and outlines.grammars.python3_interactive)
outlines.grammars.yaml

Out of scope otherwise

outlines.grammars.bash

Questions:

These tests take longer than all the other tests combined times 5. I suggest we by default only run one of these tests when pytest is run, and run all of them when pytest --do-benchmark is run. Running a single test will help ensure the changeset doesn't break CFG in general, without the need to test all grammars.
Should these be run within CI, or should PR authors manually run them when they're potentially impacting performance?

Output

Additional Benchmark Details:
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[lark_lark_self_grammar.lark.test-True]:
	Tokens / Second: 16.201
	(Num Tokens: 2771, Time: 171.035 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[lark_lark_self_grammar.lark.test-False]:
	Tokens / Second: 15.405
	(Num Tokens: 2771, Time: 179.879 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_many_features.sql.test-True]:
	Tokens / Second: 18.946
	(Num Tokens: 1294, Time: 68.300 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_many_features.sql.test-False]:
	Tokens / Second: 11.907
	(Num Tokens: 1294, Time: 108.675 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_simple.sql.test-True]:
	Tokens / Second: 21.232
	(Num Tokens: 17, Time: 0.801 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_simple.sql.test-False]:
	Tokens / Second: 2.913
	(Num Tokens: 17, Time: 5.836 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_order.sql.test-True]:
	Tokens / Second: 14.272
	(Num Tokens: 47, Time: 3.293 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_order.sql.test-False]:
	Tokens / Second: 3.614
	(Num Tokens: 47, Time: 13.004 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_having.sql.test-True]:
	Tokens / Second: 19.313
	(Num Tokens: 219, Time: 11.339 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_having.sql.test-False]:
	Tokens / Second: 6.501
	(Num Tokens: 219, Time: 33.686 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_nested_subquery.sql.test-True]:
	Tokens / Second: 17.400
	(Num Tokens: 462, Time: 26.551 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_nested_subquery.sql.test-False]:
	Tokens / Second: 8.587
	(Num Tokens: 462, Time: 53.801 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_coalesce.sql.test-True]:
	Tokens / Second: 21.573
	(Num Tokens: 230, Time: 10.662 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_coalesce.sql.test-False]:
	Tokens / Second: 7.271
	(Num Tokens: 230, Time: 31.631 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_union.sql.test-True]:
	Tokens / Second: 20.007
	(Num Tokens: 195, Time: 9.747 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[sql_select_select_union.sql.test-False]:
	Tokens / Second: 8.050
	(Num Tokens: 195, Time: 24.224 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[arithmetic_lots_of_ops.arithmetic.test-True]:
	Tokens / Second: 21.246
	(Num Tokens: 515, Time: 24.240 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[arithmetic_lots_of_ops.arithmetic.test-False]:
	Tokens / Second: 20.819
	(Num Tokens: 515, Time: 24.737 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[arithmetic_simple_math.arithmetic.test-True]:
	Tokens / Second: 21.222
	(Num Tokens: 26, Time: 1.225 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[arithmetic_simple_math.arithmetic.test-False]:
	Tokens / Second: 15.303
	(Num Tokens: 26, Time: 1.699 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit.json.test-True]:
	Tokens / Second: 35.228
	(Num Tokens: 381, Time: 10.815 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit.json.test-False]:
	Tokens / Second: 28.008
	(Num Tokens: 381, Time: 13.603 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_outlines.generate.samplers.mypy.json.test-True]:
	Tokens / Second: 20.789
	(Num Tokens: 13620, Time: 655.139 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_outlines.generate.samplers.mypy.json.test-False]:
	Tokens / Second: 17.473
	(Num Tokens: 13620, Time: 779.487 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit_no_indent.json.test-True]:
	Tokens / Second: 30.341
	(Num Tokens: 197, Time: 6.493 seconds)
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit_no_indent.json.test-False]:
	Tokens / Second: 18.822
	(Num Tokens: 197, Time: 10.466 seconds)

-------------------------------------------------------------------------------------------------------------------------- benchmark: 26 tests ---------------------------------------------------------------------------------------------------------------------------
Name (time in ms)                                                                                Min                     Max                    Mean            StdDev                  Median               IQR            Outliers     OPS            Rounds  Iterations
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
test_benchmark_cfg_generation[sql_select_select_simple.sql.test-True]                       800.6749 (1.0)          800.6749 (1.0)          800.6749 (1.0)      0.0000 (1.0)          800.6749 (1.0)      0.0000 (1.0)           0;0  1.2489 (1.0)           1           1
test_benchmark_cfg_generation[arithmetic_simple_math.arithmetic.test-True]                1,225.1380 (1.53)       1,225.1380 (1.53)       1,225.1380 (1.53)     0.0000 (1.0)        1,225.1380 (1.53)     0.0000 (1.0)           0;0  0.8162 (0.65)          1           1
test_benchmark_cfg_generation[arithmetic_simple_math.arithmetic.test-False]               1,699.0278 (2.12)       1,699.0278 (2.12)       1,699.0278 (2.12)     0.0000 (1.0)        1,699.0278 (2.12)     0.0000 (1.0)           0;0  0.5886 (0.47)          1           1
test_benchmark_cfg_generation[sql_select_select_order.sql.test-True]                      3,293.2230 (4.11)       3,293.2230 (4.11)       3,293.2230 (4.11)     0.0000 (1.0)        3,293.2230 (4.11)     0.0000 (1.0)           0;0  0.3037 (0.24)          1           1
test_benchmark_cfg_generation[sql_select_select_simple.sql.test-False]                    5,836.3914 (7.29)       5,836.3914 (7.29)       5,836.3914 (7.29)     0.0000 (1.0)        5,836.3914 (7.29)     0.0000 (1.0)           0;0  0.1713 (0.14)          1           1
test_benchmark_cfg_generation[json_simple_fruit_no_indent.json.test-True]                 6,492.7985 (8.11)       6,492.7985 (8.11)       6,492.7985 (8.11)     0.0000 (1.0)        6,492.7985 (8.11)     0.0000 (1.0)           0;0  0.1540 (0.12)          1           1
test_benchmark_cfg_generation[sql_select_select_union.sql.test-True]                      9,746.7852 (12.17)      9,746.7852 (12.17)      9,746.7852 (12.17)    0.0000 (1.0)        9,746.7852 (12.17)    0.0000 (1.0)           0;0  0.1026 (0.08)          1           1
test_benchmark_cfg_generation[json_simple_fruit_no_indent.json.test-False]               10,466.2372 (13.07)     10,466.2372 (13.07)     10,466.2372 (13.07)    0.0000 (1.0)       10,466.2372 (13.07)    0.0000 (1.0)           0;0  0.0955 (0.08)          1           1
test_benchmark_cfg_generation[sql_select_select_coalesce.sql.test-True]                  10,661.6278 (13.32)     10,661.6278 (13.32)     10,661.6278 (13.32)    0.0000 (1.0)       10,661.6278 (13.32)    0.0000 (1.0)           0;0  0.0938 (0.08)          1           1
test_benchmark_cfg_generation[json_simple_fruit.json.test-True]                          10,815.3759 (13.51)     10,815.3759 (13.51)     10,815.3759 (13.51)    0.0000 (1.0)       10,815.3759 (13.51)    0.0000 (1.0)           0;0  0.0925 (0.07)          1           1
test_benchmark_cfg_generation[sql_select_select_having.sql.test-True]                    11,339.4663 (14.16)     11,339.4663 (14.16)     11,339.4663 (14.16)    0.0000 (1.0)       11,339.4663 (14.16)    0.0000 (1.0)           0;0  0.0882 (0.07)          1           1
test_benchmark_cfg_generation[sql_select_select_order.sql.test-False]                    13,004.0021 (16.24)     13,004.0021 (16.24)     13,004.0021 (16.24)    0.0000 (1.0)       13,004.0021 (16.24)    0.0000 (1.0)           0;0  0.0769 (0.06)          1           1
test_benchmark_cfg_generation[json_simple_fruit.json.test-False]                         13,603.4424 (16.99)     13,603.4424 (16.99)     13,603.4424 (16.99)    0.0000 (1.0)       13,603.4424 (16.99)    0.0000 (1.0)           0;0  0.0735 (0.06)          1           1
test_benchmark_cfg_generation[sql_select_select_union.sql.test-False]                    24,224.4266 (30.26)     24,224.4266 (30.26)     24,224.4266 (30.26)    0.0000 (1.0)       24,224.4266 (30.26)    0.0000 (1.0)           0;0  0.0413 (0.03)          1           1
test_benchmark_cfg_generation[arithmetic_lots_of_ops.arithmetic.test-True]               24,239.8884 (30.27)     24,239.8884 (30.27)     24,239.8884 (30.27)    0.0000 (1.0)       24,239.8884 (30.27)    0.0000 (1.0)           0;0  0.0413 (0.03)          1           1
test_benchmark_cfg_generation[arithmetic_lots_of_ops.arithmetic.test-False]              24,736.5881 (30.89)     24,736.5881 (30.89)     24,736.5881 (30.89)    0.0000 (1.0)       24,736.5881 (30.89)    0.0000 (1.0)           0;0  0.0404 (0.03)          1           1
test_benchmark_cfg_generation[sql_select_select_nested_subquery.sql.test-True]           26,551.4368 (33.16)     26,551.4368 (33.16)     26,551.4368 (33.16)    0.0000 (1.0)       26,551.4368 (33.16)    0.0000 (1.0)           0;0  0.0377 (0.03)          1           1
test_benchmark_cfg_generation[sql_select_select_coalesce.sql.test-False]                 31,631.2438 (39.51)     31,631.2438 (39.51)     31,631.2438 (39.51)    0.0000 (1.0)       31,631.2438 (39.51)    0.0000 (1.0)           0;0  0.0316 (0.03)          1           1
test_benchmark_cfg_generation[sql_select_select_having.sql.test-False]                   33,686.4307 (42.07)     33,686.4307 (42.07)     33,686.4307 (42.07)    0.0000 (1.0)       33,686.4307 (42.07)    0.0000 (1.0)           0;0  0.0297 (0.02)          1           1
test_benchmark_cfg_generation[sql_select_select_nested_subquery.sql.test-False]          53,801.4904 (67.20)     53,801.4904 (67.20)     53,801.4904 (67.20)    0.0000 (1.0)       53,801.4904 (67.20)    0.0000 (1.0)           0;0  0.0186 (0.01)          1           1
test_benchmark_cfg_generation[sql_select_select_many_features.sql.test-True]             68,299.6835 (85.30)     68,299.6835 (85.30)     68,299.6835 (85.30)    0.0000 (1.0)       68,299.6835 (85.30)    0.0000 (1.0)           0;0  0.0146 (0.01)          1           1
test_benchmark_cfg_generation[sql_select_select_many_features.sql.test-False]           108,675.0438 (135.73)   108,675.0438 (135.73)   108,675.0438 (135.73)   0.0000 (1.0)      108,675.0438 (135.73)   0.0000 (1.0)           0;0  0.0092 (0.01)          1           1
test_benchmark_cfg_generation[lark_lark_self_grammar.lark.test-True]                    171,034.5827 (213.61)   171,034.5827 (213.61)   171,034.5827 (213.61)   0.0000 (1.0)      171,034.5827 (213.61)   0.0000 (1.0)           0;0  0.0058 (0.00)          1           1
test_benchmark_cfg_generation[lark_lark_self_grammar.lark.test-False]                   179,879.0275 (224.66)   179,879.0275 (224.66)   179,879.0275 (224.66)   0.0000 (1.0)      179,879.0275 (224.66)   0.0000 (1.0)           0;0  0.0056 (0.00)          1           1
test_benchmark_cfg_generation[json_outlines.generate.samplers.mypy.json.test-True]      655,138.7894 (818.23)   655,138.7894 (818.23)   655,138.7894 (818.23)   0.0000 (1.0)      655,138.7894 (818.23)   0.0000 (1.0)           0;0  0.0015 (0.00)          1           1
test_benchmark_cfg_generation[json_outlines.generate.samplers.mypy.json.test-False]     779,487.4373 (973.54)   779,487.4373 (973.54)   779,487.4373 (973.54)   0.0000 (1.0)      779,487.4373 (973.54)   0.0000 (1.0)           0;0  0.0013 (0.00)          1           1
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------

Profile

$ pytest tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit.json-False] --benchmark-cprofile=cumtime

<removed noise>

--------------------------------------------------------------------------------- cProfile (time in s) ---------------------------------------------------------------------------------
tests/benchmark/test_benchmark_cfg_generation.py::test_benchmark_cfg_generation[json_simple_fruit.json-False]
ncalls	tottime	percall	cumtime	percall	filename:lineno(function)
1	0.0000	0.0000	14.3733	14.3733	outlines/tests/benchmark/test_benchmark_cfg_generation.py:130(<lambda>)
1	0.1738	0.1738	14.3732	14.3732	outlines/tests/benchmark/test_benchmark_cfg_generation.py:51(run_until_eos)
327	0.1854	0.0006	14.1685	0.0433	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/fsm.py:222(allowed_token_ids)
208	0.7643	0.0037	13.7202	0.0660	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/fsm.py:94(__init__)
208	0.0036	0.0000	7.3470	0.0353	outlines/.myenv/lib/python3.11/site-packages/outlines/caching.py:64(wrapper)
989/860	5.6061	0.0065	5.6063	0.0065	~:0(<built-in method builtins.sorted>)
208	0.2597	0.0012	3.2571	0.0157	outlines/.myenv/lib/python3.11/site-packages/outlines/caching.py:39(hash_arguments)
8	0.0003	0.0000	3.1981	0.3998	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/fsm.py:95(create_states_mapping)
8	0.0158	0.0020	3.0903	0.3863	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/regex.py:558(create_fsm_index_tokenizer)
8	0.3168	0.0396	2.9276	0.3660	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/regex.py:494(create_fsm_index_end_to_end)
416	0.0037	0.0000	2.7863	0.0067	outlines/.myenv/lib/python3.11/site-packages/cloudpickle/cloudpickle.py:1464(dumps)
416	0.0007	0.0000	2.7790	0.0067	outlines/.myenv/lib/python3.11/site-packages/cloudpickle/cloudpickle.py:1243(dump)
416	2.7783	0.0067	2.7783	0.0067	~:0(<function Pickler.dump at 0x7f49efb03ba0>)
84	2.4870	0.0296	2.4873	0.0296	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/regex.py:461(state_scan_tokens)
200	0.0008	0.0000	0.8564	0.0043	outlines/.myenv/lib/python3.11/site-packages/diskcache/core.py:1224(__getitem__)
200	0.0018	0.0000	0.8556	0.0043	outlines/.myenv/lib/python3.11/site-packages/diskcache/core.py:1123(get)
200	0.0018	0.0000	0.8514	0.0043	outlines/.myenv/lib/python3.11/site-packages/diskcache/core.py:254(fetch)
200	0.8406	0.0042	0.8406	0.0042	~:0(<built-in method _pickle.load>)
416	0.2078	0.0005	0.2078	0.0005	~:0(<method 'update' of '_hashlib.HASH' objects>)
8	0.0001	0.0000	0.1214	0.0152	outlines/.myenv/lib/python3.11/site-packages/outlines/models/transformers.py:177(__hash__)
8	0.0000	0.0000	0.1213	0.0152	outlines/.myenv/lib/python3.11/site-packages/datasets/fingerprint.py:226(hash)
8	0.0000	0.0000	0.1203	0.0150	outlines/.myenv/lib/python3.11/site-packages/datasets/utils/_dill.py:105(dumps)
8	0.0001	0.0000	0.1203	0.0150	outlines/.myenv/lib/python3.11/site-packages/datasets/utils/_dill.py:100(dump)
8	0.0000	0.0000	0.1201	0.0150	outlines/.myenv/lib/python3.11/site-packages/dill/_dill.py:416(dump)
8	0.0001	0.0000	0.1200	0.0150	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:476(dump)
825	0.1125	0.0001	0.1125	0.0001	~:0(<method '__reduce_ex__' of 'object' objects>)
1050404	0.0843	0.0000	0.0843	0.0000	~:0(<method 'add' of 'set' objects>)
825601	0.0397	0.0000	0.0397	0.0000	~:0(<method 'setdefault' of 'dict' objects>)
534	0.0320	0.0001	0.0324	0.0001	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/fsm.py:128(allowed_token_ids)
17970/601	0.0269	0.0000	0.0740	0.0001	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/copy.py:128(deepcopy)
8	0.0237	0.0030	0.0237	0.0030	outlines/.myenv/lib/python3.11/site-packages/outlines/fsm/regex.py:584(<dictcomp>)
6204	0.0214	0.0000	0.0333	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/parsers/lalr_parser_state.py:67(feed_token)
5811	0.0188	0.0000	0.0707	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:590(next_token)
730	0.0162	0.0000	0.0463	0.0001	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:972(crawl)
8838	0.0125	0.0000	0.0125	0.0000	~:0(<method 'match' of '_regex.Pattern' objects>)
440	0.0109	0.0000	0.0109	0.0000	~:0(<method 'execute' of 'sqlite3.Connection' objects>)
8	0.0108	0.0013	0.0108	0.0013	~:0(<built-in method _pickle.dumps>)
15053	0.0106	0.0000	0.0142	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:213(_future_new)
11867	0.0081	0.0000	0.0090	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:350(follow)
76602	0.0052	0.0000	0.0052	0.0000	~:0(<method 'append' of 'list' objects>)
48965	0.0043	0.0000	0.0043	0.0000	~:0(<method 'get' of 'dict' objects>)
48586	0.0039	0.0000	0.0039	0.0000	~:0(<built-in method builtins.id>)
45129	0.0035	0.0000	0.0035	0.0000	~:0(<built-in method builtins.len>)
27369	0.0046	0.0000	0.0094	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:117(<genexpr>)
23214	0.0038	0.0000	0.0055	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:103(__getitem__)
19879	0.0037	0.0000	0.0048	0.0000	~:0(<built-in method builtins.isinstance>)
19423	0.0037	0.0000	0.0037	0.0000	~:0(<built-in method builtins.getattr>)
17970	0.0076	0.0000	0.0101	0.0000	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/copy.py:243(_keep_alive)
16402	0.0016	0.0000	0.0016	0.0000	~:0(<built-in method builtins.issubclass>)
15894	0.0037	0.0000	0.0037	0.0000	~:0(<built-in method __new__ of type object at 0x7f4a8f1b1ba0>)
15053	0.0079	0.0000	0.0221	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:202(__new__)
8849	0.0051	0.0000	0.0164	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:262(__deepcopy__)
8799	0.0059	0.0000	0.0059	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:928(follow)
8784	0.0080	0.0000	0.0220	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:387(match)
8784	0.0070	0.0000	0.0095	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:292(feed)
8784	0.0055	0.0000	0.0313	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:587(match)
8784	0.0018	0.0000	0.0039	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:581(scanner)
8784	0.0016	0.0000	0.0016	0.0000	~:0(<method 'group' of '_regex.Match' objects>)
99	0.0002	0.0000	0.0010	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:116(_get_alphabet)
99	0.0002	0.0000	0.0002	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:133(<dictcomp>)
99	0.0001	0.0000	0.0006	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:131(from_groups)
99	0.0000	0.0000	0.0000	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:126(_get_lengths)
99	0.0000	0.0000	0.0000	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:168(simplify)
983	0.0002	0.0000	0.0002	0.0000	~:0(<method 'isupper' of 'str' objects>)
981	0.0012	0.0000	0.0013	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/fsm.py:191(__init__)
98/32	0.0001	0.0000	0.0004	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:361(_get_lengths)
95	0.0001	0.0000	0.0001	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:157(<dictcomp>)
936	0.0008	0.0000	0.0008	0.0000	~:0(<method 'write' of '_io.BytesIO' objects>)
92/8	0.0001	0.0000	0.0143	0.0018	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:517(group)
90	0.0000	0.0000	0.0000	0.0000	outlines/.myenv/lib/python3.11/site-packages/lark/lexer.py:70(_get_flags)
888	0.0003	0.0000	0.0004	0.0000	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:241(write)
88	0.0027	0.0000	0.0032	0.0000	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:706(chargroup_inner)
88	0.0000	0.0000	0.0000	0.0000	~:0(<method 'end' of 're.Match' objects>)
86/8	0.0003	0.0000	0.0852	0.0106	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:447(to_fsm)
4	0.0015	0.0004	0.0042	0.0010	outlines/.myenv/lib/python3.11/site-packages/diskcache/core.py:230(_write)
1434/8	0.0019	0.0002	0.0147	0.0018	outlines/.myenv/lib/python3.11/site-packages/interegular/utils/simple_parser.py:34(w)
696/8	0.0016	0.0002	0.1196	0.0150	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:535(save)
144/8	0.0010	0.0001	0.0145	0.0018	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:503(conc)
8	0.0008	0.0001	0.0008	0.0001	~:0(<method 'update' of 'xxhash.xxh64' objects>)
355/8	0.0007	0.0001	0.0106	0.0013	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:69(get_alphabet)
696/8	0.0007	0.0001	0.1198	0.0150	outlines/.myenv/lib/python3.11/site-packages/datasets/utils/_dill.py:30(save)
12	0.0008	0.0001	0.0008	0.0001	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:172(_combine_char_groups)
696/8	0.0005	0.0001	0.1198	0.0150	outlines/.myenv/lib/python3.11/site-packages/dill/_dill.py:365(save)
339/16	0.0009	0.0001	0.0145	0.0009	outlines/.myenv/lib/python3.11/site-packages/interegular/patterns.py:512(obj)
8	0.0001	0.0000	0.1195	0.0149	outlines/.myenv/lib/python3.11/site-packages/datasets/utils/_dill.py:197(_save_transformersPreTrainedTokenizerBase)
24/8	0.0002	0.0000	0.1190	0.0149	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:621(save_reduce)
24/8	0.0001	0.0000	0.1182	0.0148	outlines/.myenv/lib/python3.11/site-packages/dill/_dill.py:1190(save_module_dict)
24/8	0.0001	0.0000	0.1181	0.0148	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:965(save_dict)
24/8	0.0000	0.0000	0.1180	0.0148	outlines/.myenv/lib/python3.11/site-packages/datasets/utils/_dill.py:71(_batch_setitems)
24/8	0.0002	0.0000	0.1180	0.0147	/nix/store/qp5zys77biz7imbk6yy85q5pdv7qk84j-python3-3.11.6/lib/python3.11/pickle.py:978(_batch_setitems)
3814	0.0018	0.0000	0.0018	0.0000	~:0(<method 'write' of '_io.BufferedWriter' objects>)
231	0.0002	0.0000	0.0002	0.0000	~:0(<method 'values' of 'dict' objects>)
8162	0.0006	0.0000	0.0006	0.0000	~:0(<method 'update' of 'set' objects>)
24	0.0000	0.0000	0.0000	0.0000	~:0(<method 'union' of 'set' objects>)
468	0.0006	0.0000	0.0011	0.0000	~:0(<method 'union' of 'frozenset' objects>)
21	0.0000	0.0000	0.0000	0.0000	~:0(<method 'translate' of 'str' objects>)
720	0.0001	0.0000	0.0001	0.0000	~:0(<method 'tell' of '_io.BytesIO' objects>)
1825	0.0010	0.0000	0.0010	0.0000	~:0(<method 'startswith' of 'str' objects>)
64	0.0000	0.0000	0.0000	0.0000	~:0(<method 'split' of 'str' objects>)
12	0.0000	0.0000	0.0000	0.0000	~:0(<method 'rstrip' of 'str' objects>)
24	0.0000	0.0000	0.0000	0.0000	~:0(<method 'rpartition' of 'str' objects>)
1981	0.0005	0.0000	0.0005	0.0000	~:0(<method 'rindex' of 'str' objects>)
12	0.0000	0.0000	0.0000	0.0000	~:0(<method 'rfind' of 'str' objects>)
3260	0.0003	0.0000	0.0003	0.0000	~:0(<method 'replace' of 'str' objects>)
2	0.0000	0.0000	0.0000	0.0000	~:0(<method 'remove' of 'set' objects>)
84	0.0000	0.0000	0.0000	0.0000	~:0(<method 'pop' of 'set' objects>)
62	0.0000	0.0000	0.0000	0.0000	~:0(<method 'pop' of 'list' objects>)
350	0.0001	0.0000	0.0001	0.0000	~:0(<method 'pop' of 'dict' objects>)
120	0.0001	0.0000	0.0001	0.0000	~:0(<method 'match' of 're.Pattern' objects>)
1750	0.0003	0.0000	0.0003	0.0000	~:0(<method 'keys' of 'dict' objects>)

…o extend current terminal vs transition to next

…e cache key

…ognized as being a different terminal

…mark-cfg-generation

rlouf · 2024-03-18T09:06:55Z

Closing due to inactivity. Feel free to re-open.

rlouf added enhancement structured generation Linked to structured generation grammar labels Jan 26, 2024

Andrew Lapp and others added 27 commits February 6, 2024 19:02

minimal grammars implementation

c4f1ffe

fixes + broken test case

2e7fcfb

standardize first state as well instead of hardcoded

1bfa37a

fix cfg incomplete edge case; can now allow model to sample whether t…

0fc7f2b

…o extend current terminal vs transition to next

fix test case

1c79934

move from outlines/lark -> outlines/grammars

5c9e3c8

update test comment

a423c1b

add docs

e0f447c

fix docs

acbd416

add reference to available grammars

efabafd

add smoketest example result to docs

ba5159d

read grammar files to a string, no path handling in CFGFSM

d348b85

revert to old var pass

0c3bf0a

fix beam search and multiple concurrent sequences using token_id tupl…

90ecd40

…e cache key

include viktor-ferenczi refactor

98808bb

fix tests

7e9fea7

don't recurse, assume previous input was handled by logits processor

26fe60e

fixes + broken test case

e80cea9

add base benchmark suite

f2fd2d9

disable cache

066c1b9

update caching order

e74f495

ensure numba is compiles so it doesn't skew benchmarks

4cffe70

add grammars and incomplete benchmark test

aaeff67

basic functioning CFGFSM tests

9629806

fix missing lark files

0885981

disable cache

3b9f720

update caching order

07da67b

Andrew Lapp added 13 commits February 6, 2024 19:20

revert bad rebase

a6bd0c9

better docs

733f0f7

Prevent UnexpectedToken if terminal is partially generated and misrec…

b9ce75d

…ognized as being a different terminal

remove incomplete bash.lark

5463478

revert vllm changes from different branch

2dd5afb

fix incorrect test name resulting from rebase

8fd9968

fix bad rebase

94917bf

fix docstring

cad42a7

fix test_fsm.py

8fdf081

improve docs

8e8dd86

provide

b560aa8

fix docs

e7f1a3c

fix test_fsm.py

b24af0d

This was referenced Feb 7, 2024

Improve CFGFSM Performance #620

Open

Apply functools.lru_cache to RegexFSM to Improve CFGFSM performance #621

Closed

Andrew Lapp added 6 commits February 7, 2024 15:10

Apply lru_cache to RegexFSM

8ed3c3d

Speed up CFGFSM By Incorporating FastLark

fea092b

Merge branch 'use-fast-copy-lalr-parser' into benchmark-cfg-generation

9c6c8c0

correct lark.lexer.Token import path

06b1653

Only Calculate TransformerTokenizer.__hash__ Once

6bd4a6e

Merge branch 'calculate-transsformers-tokenizer-hash-once' into bench…

7fcae10

…mark-cfg-generation

lapp0 mentioned this pull request Feb 7, 2024

Calculate TranssformersTokenizer.__hash__ Once to Improve CFGFSM performance #623

Closed

1 task

add todo

06d0251

lapp0 mentioned this pull request Feb 11, 2024

Add BNF grammar #636

Closed

don't reference non-existent grammar repo in docs

8470f92

rlouf closed this Mar 18, 2024

lapp0 mentioned this pull request May 29, 2024

Added benchmarks for larger regex fsm and runtime benchmarks for the same #925

Open

lapp0 mentioned this pull request Jun 23, 2024

Stable cfg lapp0/outlines#34

Draft

10 tasks

lapp0 mentioned this pull request Jul 25, 2024

Update CFGGuide to use outlines.fsm.parsing. Enable generate.cfg #1067

Merged

2 tasks

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add More Grammars to `outlines.grammars`, Benchmark and Verify Their Integrity #587

Add More Grammars to `outlines.grammars`, Benchmark and Verify Their Integrity #587

lapp0 commented Jan 25, 2024 •

edited

Loading

rlouf commented Mar 18, 2024

Add More Grammars to outlines.grammars, Benchmark and Verify Their Integrity #587

Add More Grammars to outlines.grammars, Benchmark and Verify Their Integrity #587

Conversation

lapp0 commented Jan 25, 2024 • edited Loading

Output

Profile

rlouf commented Mar 18, 2024

Add More Grammars to `outlines.grammars`, Benchmark and Verify Their Integrity #587

Add More Grammars to `outlines.grammars`, Benchmark and Verify Their Integrity #587

lapp0 commented Jan 25, 2024 •

edited

Loading