Fix llamacpp caching by making LlamaCppTokenizer an outlines Tokenizer

outlines/integrations/llamacpp.py

@@ -26,7 +26,7 @@
 """
 
 import math
-from typing import TYPE_CHECKING, Dict, Optional, Set, Type, Union
+from typing import TYPE_CHECKING, Optional, Type, Union
 
 import numpy as np
 import torch
@@ -36,37 +36,12 @@
 from outlines.fsm.guide import CFGGuide, Guide, RegexGuide
 from outlines.fsm.json_schema import build_regex_from_schema
 from outlines.integrations.utils import convert_json_schema_to_str
+from outlines.models.llamacpp import LlamaCppTokenizer
 
 if TYPE_CHECKING:
     from llama_cpp import Llama
 
 
-class LlamaCppTokenizer:
-    def __init__(self, model: "Llama"):
-        self.eos_token_id = model.token_eos()
-        self.eos_token = model.tokenizer().decode([self.eos_token_id])
-        self.pad_token_id = self.eos_token_id
-        self.special_tokens: Set[int] = set()
-
-        self.vocabulary: Dict[str, int] = dict()
-
-        tokenizer = model.tokenizer()
-
-        self.decode = tokenizer.decode
-
-        # TODO: Remove when https://github.com/ggerganov/llama.cpp/pull/5613 is resolved
-        try:
-            self.vocabulary = model.tokenizer_.hf_tokenizer.get_vocab()
-        except AttributeError:
-            # ###
-            for t in range(model.n_vocab()):
-                token_piece = model.tokenizer().decode([t])
-                self.vocabulary[token_piece] = t
-
-    def convert_token_to_string(self, token: str) -> str:
-        return token
-
-
 class LogitsProcessor:
     """Bias LlamaCpp generation using a finite state machine.
 

outlines/models/llamacpp.py

@@ -1,15 +1,102 @@
 import dataclasses
+import pickle
 import warnings
-from typing import TYPE_CHECKING, Iterator, List, Optional, TypedDict, Union
+from typing import (
+    TYPE_CHECKING,
+    Dict,
+    Iterator,
+    List,
+    Optional,
+    Set,
+    Tuple,
+    TypedDict,
+    Union,
+)
 
 from typing_extensions import Unpack
 
 from outlines.generate.api import GenerationParameters, SamplingParameters
+from outlines.models.tokenizer import Tokenizer
 
 if TYPE_CHECKING:
     from llama_cpp import Llama, LogitsProcessorList
 
 
+class LlamaCppTokenizer(Tokenizer):
+    def __init__(self, model: "Llama"):
+        self.eos_token_id = model.token_eos()
+        self.eos_token = model.tokenizer().decode([self.eos_token_id])
+        self.pad_token_id = self.eos_token_id
+        self.special_tokens: Set[int] = set()
+
+        self.vocabulary: Dict[str, int] = dict()
+
+        self.tokenizer = model.tokenizer()
+
+        # TODO: Remove when https://github.com/ggerganov/llama.cpp/pull/5613 is resolved
+        try:
+            self.vocabulary = model.tokenizer_.hf_tokenizer.get_vocab()
+        except AttributeError:
+            # ###
+            for t in range(model.n_vocab()):
+                token_piece = model.tokenizer().decode([t])
+                self.vocabulary[token_piece] = t
+
+        # ensure stable ordering of vocabulary
+        self.vocabulary = {
+            tok: tok_id
+            for tok, tok_id in sorted(self.vocabulary.items(), key=lambda x: x[1])
+        }
+
+        self._hash = None
+
+    def decode(self, token_ids: List[int]) -> List[str]:
+        decoded_bytes = self.tokenizer.detokenize(token_ids)
+        return [decoded_bytes.decode("utf-8", errors="ignore")]
+
+    def encode(
+        self, prompt: Union[str, List[str]], add_bos: bool = True, special: bool = True
+    ) -> Tuple[List[int], List[int]]:
+        if isinstance(prompt, list):
+            raise NotImplementedError(
+                "llama-cpp-python tokenizer doesn't support batch tokenization"
+            )
+        token_ids = self.tokenizer.tokenize(
+            prompt.encode("utf-8", errors="ignore"), add_bos=add_bos, special=special
+        )
+        # generate attention mask, missing from llama-cpp-python
+        attention_mask = [
+            1 if token_id != self.pad_token_id else 0 for token_id in token_ids
+        ]
+        return token_ids, attention_mask
+
+    def convert_token_to_string(self, token: str) -> str:
+        return token
+
+    def __eq__(self, other):
+        if not isinstance(other, LlamaCppTokenizer):
+            return False
+        return self.__getstate__() == other.__getstate__()
+
+    def __hash__(self):
+        if self._hash is None:
+            self._hash = hash(pickle.dumps(self))
+        return self._hash
+
+    def __getstate__(self):
+        """Create a stable representation for outlines.caching"""
+        return (
+            self.vocabulary,
+            self.eos_token_id,
+            self.eos_token,
+            self.pad_token_id,
+            sorted(self.special_tokens),
+        )
+
+    def __setstate__(self, state):
+        raise NotImplementedError("Cannot load a pickled llamacpp tokenizer")
+
+
 class LlamaCppParams(TypedDict, total=False):
     suffix: Optional[str]
     temperature: float

tests/generate/conftest.py

@@ -0,0 +1,24 @@
+from importlib import reload
+
+import pytest
+
+
+@pytest.fixture
+def temp_cache_dir():
+    import os
+    import tempfile
+
+    import outlines.caching
+    import outlines.fsm.guide
+
+    with tempfile.TemporaryDirectory() as tempdir:
+        os.environ["OUTLINES_CACHE_DIR"] = tempdir
+        outlines.caching.get_cache.cache_clear()
+        reload(outlines)
+        reload(outlines.fsm.guide)
+        cache_status = outlines.caching._caching_enabled
+        try:
+            outlines.caching._caching_enabled = True
+            yield
+        finally:
+            outlines.caching._caching_enabled = cache_status

tests/generate/test_integration_llamacpp.py

@@ -279,3 +279,58 @@ def test_llama_cpp_pre_tokenizer_remains_broken():
     model = models.llamacpp(repo, model_path)
     with pytest.raises(RuntimeError):
         generate.choice(model, ["skirt", "dress", "pen", "jacket"])
+
+
+def test_RegexGuide_caching(model, temp_cache_dir):
+    import llama_cpp
+
+    import outlines.caching
+    from outlines.fsm.guide import create_states_mapping
+
+    assert outlines.caching._caching_enabled
+
+    regex = r"((25[0-5]|2[0-4]\d|[01]?\d\d?)\.){3}(25[0-5]|2[0-4]\d|[01]?\d\d?)"
+    prompt = "What is the IP address of the Google DNS servers? "
+
+    cache = outlines.caching.get_cache()
+
+    # Returns (hits, misses)
+    _ = cache.stats(enable=True)
+    assert cache.statistics
+
+    assert create_states_mapping.__memory__ is cache
+
+    generator = generate.regex(model, regex, sampler=samplers.greedy())
+    assert cache.stats() == (0, 1)
+
+    model_2 = models.llamacpp(
+        "Qwen/Qwen1.5-0.5B-Chat-GGUF",
+        "*q2*.gguf",
+        tokenizer=llama_cpp.llama_tokenizer.LlamaHFTokenizer.from_pretrained(
+            "Qwen/Qwen1.5-0.5B-Chat"
+        ),
+    )
+    generator_2 = generate.regex(model_2, regex, sampler=samplers.greedy())
+    assert cache.stats() == (0, 2)
+
+    # These two different models and tokenizers should not have the same state
+    # mapping results
+    assert (
+        generator.logits_processor.fsm.states_to_token_maps
+        != generator_2.logits_processor.fsm.states_to_token_maps
+    )
+
+    generator_3 = generate.regex(model_2, regex, sampler=samplers.greedy())
+    assert cache.stats() == (1, 2)
+    assert (
+        generator_2.logits_processor.fsm.states_to_token_maps
+        == generator_3.logits_processor.fsm.states_to_token_maps
+    )
+
+    # Just for fun...
+    structured = generator(prompt, max_tokens=30)
+    structured_2 = generator_2(prompt, max_tokens=30)
+
+    assert re.fullmatch(regex, structured)
+    assert re.fullmatch(regex, structured_2)
+    assert structured != structured_2

tests/generate/test_integration_transformers.py

@@ -1,7 +1,6 @@
 import datetime
 import re
 from enum import Enum
-from importlib import reload
 from typing import List, Union
 
 import pytest
@@ -15,27 +14,6 @@
 from outlines.samplers import beam_search, greedy, multinomial
 
 
-@pytest.fixture
-def temp_cache_dir():
-    import os
-    import tempfile
-
-    import outlines.caching
-    import outlines.fsm.guide
-
-    with tempfile.TemporaryDirectory() as tempdir:
-        os.environ["OUTLINES_CACHE_DIR"] = tempdir
-        outlines.caching.get_cache.cache_clear()
-        reload(outlines)
-        reload(outlines.fsm.guide)
-        cache_status = outlines.caching._caching_enabled
-        try:
-            outlines.caching._caching_enabled = True
-            yield
-        finally:
-            outlines.caching._caching_enabled = cache_status
-
-
 def test_transformers_integration_text():
     rng = torch.Generator()
     rng.manual_seed(10000)  # Choosen so <EOS> is generated