prompt-firewall 0.1.4__py3-none-any.whl → 0.1.5__py3-none-any.whl

{prompt_firewall-0.1.4.dist-info → prompt_firewall-0.1.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: prompt-firewall
-Version: 0.1.4
+Version: 0.1.5
 Summary: Prompt guard library for composing multiple safety checks.
 Requires-Python: >=3.13
 Description-Content-Type: text/markdown
@@ -22,8 +22,6 @@ Composable prompt-safety guards as a lightweight Python library. It provides:
 from prompt_guard import GuardPipeline, PromptGuardConfig
 
 cfg = PromptGuardConfig(
-    phrases_path="phrases.txt",
-    sentences_path="sentences.txt",
     embed_model_name="mxbai-embed-large",
     judge_model_name="qwen2.5:3b-instruct",
 )

{prompt_firewall-0.1.4.dist-info → prompt_firewall-0.1.5.dist-info}/RECORD

@@ -1,6 +1,6 @@
 prompt_guard/__init__.py,sha256=MMcONUgjoE5sG8LQX8n6pF7mvawil6XLp4Mxa3KX1Vw,174
-prompt_guard/config.py,sha256=YgdulueF9H984P6RBA1MiehFmKbdqrGwEy3Tp0ur7sY,727
-prompt_guard/pipeline.py,sha256=k91cdza08UopgoRByciYLpdWcuqZCFd5QD21_OpiWkU,3195
+prompt_guard/config.py,sha256=MCewtkFFL51yXm0I1smFvcEQCTRcYz7Kl1xWij4OQ-g,2144
+prompt_guard/pipeline.py,sha256=zR9a958oG8faAH1hXBcZIpCAxl42e-Gw9coj5-_2tPI,3227
 prompt_guard/protocols.py,sha256=xHOKaIyexvl1HPp7dAdswr2lLDdGc542VtF1SQOMGqE,396
 prompt_guard/data/phrases.txt,sha256=HRP32xImm2AxB42oNVhYXAUX2ikrJWpyjw4BbkN6vnU,348
 prompt_guard/data/sentences.txt,sha256=sbrzWRh8jirNX_xY32PPjp7VGEWxWorN9t-IyVjH4_4,1275
@@ -8,7 +8,7 @@ prompt_guard/guards/__init__.py,sha256=DZp9g5I-H6AM3MkNyZnLi48kc7-bAECP48b6SO6YD
 prompt_guard/guards/judge_guard.py,sha256=dtlcZjG8s8l_6SijyRfVwE_yz7gqFiKtO8uwL0Jlo_E,1870
 prompt_guard/guards/rag_guard.py,sha256=bfDDdLZQvWTqCk6j7N8uCeCNzpkOTOFqh-jUYk4pyPk,960
 prompt_guard/guards/tfidf_guard.py,sha256=9nis8bobAHCgBtpUAYHXOHaQHvs4WQUUXvHWme-OET0,1351
-prompt_firewall-0.1.4.dist-info/METADATA,sha256=bRnGV47DGJiHeRNA6yYgkAhbxrptYDZmvwXKQUgINkk,1329
-prompt_firewall-0.1.4.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
-prompt_firewall-0.1.4.dist-info/top_level.txt,sha256=5XNPQyMJgYV0nzEbniQJ3NjJcthIHv4h0WcPXOErxsk,13
-prompt_firewall-0.1.4.dist-info/RECORD,,
+prompt_firewall-0.1.5.dist-info/METADATA,sha256=krtbsTgMufdnmgS0EWfZMvcY4ig0zDsg5bDGfGegWAw,1261
+prompt_firewall-0.1.5.dist-info/WHEEL,sha256=wUyA8OaulRlbfwMtmQsvNngGrxQHAvkKcvRmdizlJi0,92
+prompt_firewall-0.1.5.dist-info/top_level.txt,sha256=5XNPQyMJgYV0nzEbniQJ3NjJcthIHv4h0WcPXOErxsk,13
+prompt_firewall-0.1.5.dist-info/RECORD,,

prompt_guard/config.py

@@ -1,30 +1,77 @@
-from dataclasses import dataclass
+from pydantic import BaseModel, Field
 
 
-DEFAULT_PHRASES_RESOURCE = "data/phrases.txt"
-DEFAULT_SENTENCES_RESOURCE = "data/sentences.txt"
+BUILTIN_DEFAULT_PHRASES_RESOURCE = "data/phrases.txt"
+BUILTIN_DEFAULT_SENTENCES_RESOURCE = "data/sentences.txt"
 
 
-@dataclass(slots=True)
-class PromptGuardConfig:
-    phrases_path: str | None = None
-    sentences_path: str | None = None
+class PromptGuardConfig(BaseModel):
+    phrases_path: str | None = Field(
+        default=None,
+        examples=["/path/to/phrases.txt"],
+        description="Path to phrases file (one phrase per line).",
+    )
+    sentences_path: str | None = Field(
+        default=None,
+        examples=["/path/to/sentences.txt"],
+        description="Path to sentences file (one sentence per line).",
+    )
 
-    tfidf_top_k: int = 5
-    tfidf_ngram_range: tuple[int, int] = (1, 3)
-    rag_top_k: int = 5
+    tfidf_top_k: int = Field(
+        default=5,
+        examples=[5],
+        description="Number of top TF-IDF matches to keep.",
+    )
+    tfidf_ngram_range: tuple[int, int] = Field(
+        default=(1, 3),
+        examples=[(1, 3)],
+        description="N-gram range for TF-IDF vectorizer.",
+    )
+    rag_top_k: int = Field(
+        default=5,
+        examples=[5],
+        description="Number of top RAG matches to retrieve.",
+    )
 
-    embed_model_name: str = "mxbai-embed-large"
-    judge_model_name: str = "qwen2.5:3b-instruct"
-    judge_temperature: float = 0.0
-    judge_max_tokens: int = 256
+    embed_model_name: str = Field(
+        default="mxbai-embed-large",
+        examples=["mxbai-embed-large"],
+        description="Ollama embedding model name.",
+    )
+    judge_model_name: str = Field(
+        default="qwen2.5:3b-instruct",
+        examples=["qwen2.5:3b-instruct"],
+        description="Ollama judge model name.",
+    )
+    judge_temperature: float = Field(
+        default=0.0,
+        examples=[0.0],
+        description="LLM judge temperature.",
+    )
+    judge_max_tokens: int = Field(
+        default=256,
+        examples=[256],
+        description="LLM judge max tokens.",
+    )
 
-    base_url: str | None = None
+    base_url: str | None = Field(
+        default=None,
+        examples=["http://localhost:11434"],
+        description="Optional base URL for Ollama.",
+    )
 
-    tfidf_limit: float = 0.0
-    rag_limit: float = 0.0
-    judge_limit: float = 0.0
-
-    enable_tfidf: bool = True
-    enable_rag: bool = True
-    enable_judge: bool = True
+    enable_tfidf: bool = Field(
+        default=True,
+        examples=[True],
+        description="Enable TF-IDF guard.",
+    )
+    enable_rag: bool = Field(
+        default=True,
+        examples=[True],
+        description="Enable RAG guard.",
+    )
+    enable_judge: bool = Field(
+        default=True,
+        examples=[True],
+        description="Enable LLM judge guard.",
+    )

prompt_guard/pipeline.py

@@ -6,8 +6,8 @@ from llama_index.embeddings.ollama import OllamaEmbedding
 from sklearn.feature_extraction.text import TfidfVectorizer
 
 from .config import (
-    DEFAULT_PHRASES_RESOURCE,
-    DEFAULT_SENTENCES_RESOURCE,
+    BUILTIN_DEFAULT_PHRASES_RESOURCE,
+    BUILTIN_DEFAULT_SENTENCES_RESOURCE,
     PromptGuardConfig,
 )
 from .guards.judge_guard import LlmJudgeGuard
@@ -43,7 +43,7 @@ class GuardPipeline:
     def _build_tfidf_guard(self) -> TfIdfGuard:
         phrases = self._load_lines(
             self._config.phrases_path,
-            default_resource=DEFAULT_PHRASES_RESOURCE,
+            default_resource=BUILTIN_DEFAULT_PHRASES_RESOURCE,
         )
         vectorizer = TfidfVectorizer(ngram_range=self._config.tfidf_ngram_range)
         phrase_matrix = vectorizer.fit_transform(phrases)
@@ -57,7 +57,7 @@ class GuardPipeline:
     def _build_rag_guard(self) -> RagGuard:
         lines = self._load_lines(
             self._config.sentences_path,
-            default_resource=DEFAULT_SENTENCES_RESOURCE,
+            default_resource=BUILTIN_DEFAULT_SENTENCES_RESOURCE,
         )
         docs = [Document(text=line) for line in lines]
         embed_kwargs = {"model_name": self._config.embed_model_name}