rager 0.1.0__py3-none-any.whl

rager/__init__.py

@@ -0,0 +1,55 @@
+"""Caching based RAG primitives."""
+
+import logging
+
+from rager.chunkers import Chunker, SemanticChunker
+from rager.embedders import (
+    Embedder,
+    SentenceTransformerDenseEmbedder,
+    SpladeSparseEmbedder,
+)
+from rager.fusers import BordaCountFuser, Fuser, ReciprocalRankFuser
+from rager.generators import Generator, TransformersGenerator
+from rager.indexes import DenseIndex, Index, SparseIndex
+from rager.parsers import (
+    CsvFileParser,
+    MarkdownFileParser,
+    Parser,
+    PdfFileParser,
+    UnstructuredFileParser,
+)
+from rager.scorers import CrossEncoderScorer, Scorer
+from rager.stores import ChunkStore, MetadataStore, Store
+from rager.types import DenseEmbedding, Hash, Metadata, SparseEmbedding
+
+logging.getLogger(__name__).addHandler(logging.NullHandler())
+
+__all__ = [
+    "BordaCountFuser",
+    "ChunkStore",
+    "Chunker",
+    "CrossEncoderScorer",
+    "CsvFileParser",
+    "DenseEmbedding",
+    "DenseIndex",
+    "Embedder",
+    "Fuser",
+    "Generator",
+    "Hash",
+    "Index",
+    "MarkdownFileParser",
+    "Metadata",
+    "MetadataStore",
+    "Parser",
+    "PdfFileParser",
+    "ReciprocalRankFuser",
+    "Scorer",
+    "SemanticChunker",
+    "SentenceTransformerDenseEmbedder",
+    "SparseEmbedding",
+    "SparseIndex",
+    "SpladeSparseEmbedder",
+    "Store",
+    "TransformersGenerator",
+    "UnstructuredFileParser",
+]

rager/chunkers.py

@@ -0,0 +1,71 @@
+"""Chunkers for splitting units into smaller chunks."""
+
+import logging
+from functools import cached_property
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol, cast
+
+import belljar
+from semantic_chunker import get_chunker
+
+if TYPE_CHECKING:
+    from semantic_text_splitter import TextSplitter
+
+logger = logging.getLogger(__name__)
+
+
+class Chunker(Protocol):
+    """Protocol for chunking units into smaller chunks."""
+
+    def chunk(self, unit: str) -> list[str]:
+        """Split a unit into smaller chunks."""
+        ...
+
+
+class SemanticChunker:
+    """Chunker that splits units based on semantic boundaries."""
+
+    def __init__(
+        self,
+        model_name: str = "gpt-3.5-turbo",
+        chunk_size: int = 1000,
+        overlap: int = 0,
+    ) -> None:
+        """Initialize the semantic chunker."""
+        self.model_name = model_name
+        self.chunk_size = chunk_size
+        self.overlap = overlap
+
+    @cached_property
+    def model(self) -> TextSplitter:
+        """Get the text splitter instance."""
+        logger.info(
+            "Loading semantic chunker for model %r (chunk_size=%d, overlap=%d)",
+            self.model_name,
+            self.chunk_size,
+            self.overlap,
+        )
+        return cast(
+            "TextSplitter",
+            get_chunker(
+                self.model_name,
+                chunking_type="text",
+                tree_sitter_language=None,
+                max_tokens=self.chunk_size,
+                overlap=self.overlap,
+                trim=True,
+            ),
+        )
+
+    @belljar.store(Path(".jar/chunkers"))
+    def chunk(self, unit: str) -> list[str]:
+        """Split a unit into semantically meaningful chunks."""
+        belljar.include(self.model_name)
+        belljar.include(self.chunk_size)
+        belljar.include(self.overlap)
+        belljar.include(unit)
+        belljar.check()
+        logger.debug("Cache miss; chunking unit of %d characters", len(unit))
+        chunks = self.model.chunks(unit)
+        logger.debug("Split unit into %d chunks", len(chunks))
+        return chunks

rager/embedders.py

@@ -0,0 +1,129 @@
+"""Embedders for converting text chunks into vector representations."""
+
+import logging
+from datetime import timedelta
+from functools import cached_property
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol
+
+import belljar
+import concresce
+from sentence_transformers import SentenceTransformer, SparseEncoder
+
+if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable, Iterable, Iterator
+
+    from torch import Tensor
+
+    from rager import DenseEmbedding
+    from rager.types import SparseEmbedding
+
+logger = logging.getLogger(__name__)
+
+
+class Embedder[E](Protocol):
+    """Protocol for embedders."""
+
+    async def embed(self, chunk: str) -> E:
+        """Convert a text chunk into a vector representation."""
+        ...
+
+
+class SentenceTransformerDenseEmbedder:
+    """Dense embedding using SentenceTransformer."""
+
+    def __init__(self, model_name: str = "all-MiniLM-L6-v2") -> None:
+        """Initialize the dense embedder with a specific model."""
+        self.model_name = model_name
+
+    @cached_property
+    def model(self) -> SentenceTransformer:
+        """Load the SentenceTransformer model."""
+        logger.info("Loading SentenceTransformer model %r", self.model_name)
+        return SentenceTransformer(self.model_name)
+
+    @cached_property
+    def _encode(self) -> Callable[[str], Awaitable[DenseEmbedding]]:
+        """Coalesce concurrent calls into per-instance encoding batches."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._encode_batch)
+
+    async def _encode_batch(self, chunk: str) -> DenseEmbedding:
+        """Convert a text chunk into a vector representation."""
+        chunks = await concresce.collect(chunk)
+        logger.debug(
+            "Encoding batch of %d chunks with %r", len(chunks), self.model_name
+        )
+        embeddings = self.model.encode(chunks, normalize_embeddings=True).tolist()
+        return concresce.scatter(embeddings)
+
+    @belljar.store(Path(".jar/embedders"))
+    async def embed(self, chunk: str) -> DenseEmbedding:
+        """Convert a text chunk into a vector representation."""
+        belljar.include(self.model_name)
+        belljar.include(chunk)
+        belljar.check()
+        logger.debug("Cache miss; embedding chunk of %d characters", len(chunk))
+        return await self._encode(chunk)
+
+
+class SpladeSparseEmbedder:
+    """Sparse embedding using a SPLADE SparseEncoder."""
+
+    def __init__(self, model_name: str = "prithivida/Splade_PP_en_v1") -> None:
+        """Initialize the sparse embedder with a specific model."""
+        self.model_name = model_name
+
+    @cached_property
+    def model(self) -> SparseEncoder:
+        """Load the SPLADE SparseEncoder model."""
+        logger.info("Loading SPLADE SparseEncoder model %r", self.model_name)
+        return SparseEncoder(self.model_name)
+
+    @staticmethod
+    def _decode_entries(coalesced: Tensor) -> Iterator[tuple[int, int, float]]:
+        """Yield (row, token_id, weight) triples from a coalesced sparse tensor."""
+        indices = coalesced.indices()
+        values = coalesced.values()
+        for row, token, weight in zip(indices[0], indices[1], values, strict=True):
+            yield int(row), int(token), float(weight)
+
+    @staticmethod
+    def _group_by_row(
+        entries: Iterable[tuple[int, int, float]], count: int
+    ) -> list[SparseEmbedding]:
+        """Bucket (row, token_id, weight) triples into one weight map per row."""
+        embeddings: list[SparseEmbedding] = [{} for _ in range(count)]
+        for row, token, weight in entries:
+            embeddings[row][token] = weight
+        return embeddings
+
+    @classmethod
+    def _coalesced_to_embeddings(
+        cls, coalesced: Tensor, count: int
+    ) -> list[SparseEmbedding]:
+        """Split a coalesced sparse tensor into one weight map per input row."""
+        return cls._group_by_row(cls._decode_entries(coalesced), count)
+
+    @cached_property
+    def _encode(self) -> Callable[[str], Awaitable[SparseEmbedding]]:
+        """Coalesce concurrent calls into per-instance encoding batches."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._encode_batch)
+
+    async def _encode_batch(self, chunk: str) -> SparseEmbedding:
+        """Convert a text chunk into a sparse vector representation."""
+        chunks = await concresce.collect(chunk)
+        logger.debug(
+            "Encoding batch of %d chunks with %r", len(chunks), self.model_name
+        )
+        coalesced = self.model.encode(chunks).coalesce()
+        embeddings = self._coalesced_to_embeddings(coalesced, len(chunks))
+        return concresce.scatter(embeddings)
+
+    @belljar.store(Path(".jar/embedders"))
+    async def embed(self, chunk: str) -> SparseEmbedding:
+        """Convert a text chunk into a sparse vector representation."""
+        belljar.include(self.model_name)
+        belljar.include(chunk)
+        belljar.check()
+        logger.debug("Cache miss; embedding chunk of %d characters", len(chunk))
+        return await self._encode(chunk)

rager/fusers.py

@@ -0,0 +1,53 @@
+"""Fusers for combining ranked lists of values."""
+
+import logging
+from typing import TYPE_CHECKING, Protocol
+
+if TYPE_CHECKING:
+    from collections.abc import Hashable
+
+logger = logging.getLogger(__name__)
+
+
+class Fuser[V](Protocol):
+    """Protocol for fusing values into ranked lists."""
+
+    def fuse(self, *values: list[V]) -> list[V]:
+        """Fuse multiple values into one ranked list."""
+        ...
+
+
+class ReciprocalRankFuser[V: Hashable]:
+    """Fuser that combines ranked lists by reciprocal rank fusion."""
+
+    def __init__(self, k: int = 60) -> None:
+        """Initialize the fuser with the reciprocal rank smoothing constant."""
+        self.k = k
+
+    def fuse(self, *values: list[V]) -> list[V]:
+        """Fuse multiple ranked lists into one list ranked by reciprocal rank."""
+        logger.debug(
+            "Fusing %d rankings by reciprocal rank (k=%d)", len(values), self.k
+        )
+        scores: dict[V, float] = {}
+        for ranking in values:
+            for rank, value in enumerate(ranking, start=1):
+                scores[value] = scores.get(value, 0.0) + 1 / (self.k + rank)
+        fused = sorted(scores, key=lambda value: scores[value], reverse=True)
+        logger.debug("Fused rankings into %d distinct values", len(fused))
+        return fused
+
+
+class BordaCountFuser[V: Hashable]:
+    """Fuser that combines ranked lists by Borda count."""
+
+    def fuse(self, *values: list[V]) -> list[V]:
+        """Fuse multiple ranked lists into one list ranked by Borda count."""
+        logger.debug("Fusing %d rankings by Borda count", len(values))
+        scores: dict[V, int] = {}
+        for ranking in values:
+            for rank, value in enumerate(ranking):
+                scores[value] = scores.get(value, 0) + len(ranking) - rank
+        fused = sorted(scores, key=lambda value: scores[value], reverse=True)
+        logger.debug("Fused rankings into %d distinct values", len(fused))
+        return fused

rager/generators.py

@@ -0,0 +1,80 @@
+"""Generators for producing answers to prompts."""
+
+import logging
+from datetime import timedelta
+from functools import cached_property
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol
+
+import belljar
+import concresce
+from transformers import pipeline
+
+if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable
+
+    from transformers import TextGenerationPipeline
+
+logger = logging.getLogger(__name__)
+
+
+class Generator(Protocol):
+    """Protocol for prompting content generators."""
+
+    async def prompt(self, query: str) -> str:
+        """Generate content based on the query."""
+        ...
+
+
+class TransformersGenerator:
+    """Generator that prompts a local transformers text-generation model."""
+
+    def __init__(
+        self,
+        model_name: str = "HuggingFaceTB/SmolLM2-135M-Instruct",
+        max_new_tokens: int = 512,
+    ) -> None:
+        """Initialize the generator with a specific model."""
+        self.model_name = model_name
+        self.max_new_tokens = max_new_tokens
+
+    @cached_property
+    def model(self) -> TextGenerationPipeline:
+        """Load the text-generation pipeline."""
+        logger.info("Loading text-generation pipeline for model %r", self.model_name)
+        return pipeline("text-generation", model=self.model_name)
+
+    @cached_property
+    def _generate(self) -> Callable[[str], Awaitable[str]]:
+        """Coalesce concurrent calls into per-instance generation batches."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._generate_batch)
+
+    async def _generate_batch(self, query: str) -> str:
+        """Generate an answer for each query in the collected batch."""
+        queries = await concresce.collect(query)
+        logger.debug(
+            "Generating answers for a batch of %d queries with %r (max_new_tokens=%d)",
+            len(queries),
+            self.model_name,
+            self.max_new_tokens,
+        )
+        chats = [[{"role": "user", "content": query}] for query in queries]
+        outputs = self.model(
+            chats,
+            max_new_tokens=self.max_new_tokens,
+            do_sample=False,
+        )
+        answers = [output[0]["generated_text"][-1]["content"] for output in outputs]
+        return concresce.scatter(answers)
+
+    @belljar.store(Path(".jar/generators"))
+    async def prompt(self, query: str) -> str:
+        """Generate content based on the query."""
+        belljar.include(self.model_name)
+        belljar.include(self.max_new_tokens)
+        belljar.include(query)
+        belljar.check()
+        logger.debug(
+            "Cache miss; generating answer for query of %d characters", len(query)
+        )
+        return await self._generate(query)

rager/indexes.py

@@ -0,0 +1,262 @@
+"""Embedding Index protocol definitions."""
+
+import logging
+from datetime import timedelta
+from functools import cached_property
+from typing import TYPE_CHECKING, Protocol
+
+import blake3
+import concresce
+import faiss
+import numpy as np
+
+if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable
+
+    from rager import DenseEmbedding, SparseEmbedding
+
+logger = logging.getLogger(__name__)
+
+
+class Index[E, K](Protocol):
+    """Protocol for an embedding index.
+
+    Similarity is measured by inner product, which equals cosine similarity
+    only when the embeddings are L2-normalized. Normalize your embeddings
+    before adding them if you want cosine ranking.
+    """
+
+    async def add(self, embedding: E) -> K:
+        """Add an embedding to the index and return its key."""
+        ...
+
+    async def remove(self, key: K) -> None:
+        """Remove an embedding from the index by its key."""
+        ...
+
+    async def similar(self, embedding: E, results: int = 100) -> list[K]:
+        """Retrieve the keys of the ``results`` most similar embeddings."""
+        ...
+
+
+class DenseIndex:
+    """Dense embedding index backed by a flat FAISS inner-product index.
+
+    Ranking uses inner product, so it matches cosine similarity only for
+    L2-normalized embeddings (as produced by the bundled dense embedder).
+
+    The embedding dimension is inferred from the first embedding added. Pass
+    ``dimensions`` to validate that every embedding matches an expected width.
+    """
+
+    def __init__(self, dimensions: int | None = None) -> None:
+        """Initialize the index, optionally fixing the embedding dimension."""
+        self.dimensions = dimensions
+        self._index: faiss.IndexIDMap2 | None = None
+
+    @cached_property
+    def add(self) -> Callable[[DenseEmbedding], Awaitable[int]]:
+        """Add an embedding to the index and return its key."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._add)
+
+    @cached_property
+    def remove(self) -> Callable[[int], Awaitable[None]]:
+        """Remove an embedding from the index by its key."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._remove)
+
+    async def similar(self, embedding: DenseEmbedding, results: int = 100) -> list[int]:
+        """Retrieve the keys of the ``results`` most similar embeddings."""
+        return await self._similar(embedding, results)
+
+    @cached_property
+    def _similar(self) -> Callable[[DenseEmbedding, int], Awaitable[list[int]]]:
+        """Coalesce concurrent queries into one batched FAISS search."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._similar_batch)
+
+    def _ensure_index(self, dimensions: int) -> faiss.IndexIDMap2:
+        """Return the index, building it to match the first embedding's width."""
+        if self.dimensions is None:
+            logger.debug("Inferred embedding dimension %d from first batch", dimensions)
+            self.dimensions = dimensions
+        if dimensions != self.dimensions:
+            message = (
+                f"Embedding has {dimensions} dimensions, "
+                f"but the index expects {self.dimensions}."
+            )
+            logger.error(message)
+            raise ValueError(message)
+        if self._index is None:
+            logger.info("Created dense FAISS index with %d dimensions", self.dimensions)
+            self._index = faiss.IndexIDMap2(faiss.IndexFlatIP(self.dimensions))
+        return self._index
+
+    @staticmethod
+    def _to_rows(embeddings: list[DenseEmbedding]) -> np.ndarray:
+        """Convert embeddings into a float32 matrix for FAISS."""
+        return np.asarray(embeddings, dtype=np.float32)
+
+    @staticmethod
+    def _key(row: np.ndarray) -> int:
+        """Derive a deterministic int64 key from the embedding's content."""
+        digest = blake3.blake3(row.tobytes()).digest()
+        return int.from_bytes(digest[:8], "little", signed=True)
+
+    async def _add(self, embedding: DenseEmbedding) -> int:
+        """Add an embedding to the index and return its key."""
+        embeddings = await concresce.collect(embedding)
+        rows = self._to_rows(embeddings)
+        index = self._ensure_index(rows.shape[1])
+        keys = [self._key(row) for row in rows]
+        unique = dict(zip(keys, rows, strict=True))
+        ids = np.asarray(list(unique), dtype=np.int64)
+        index.remove_ids(faiss.IDSelectorBatch(ids))
+        index.add_with_ids(self._to_rows(list(unique.values())), ids)
+        logger.debug(
+            "Added %d embeddings (%d unique) to the dense index; total is now %d",
+            len(keys),
+            len(unique),
+            index.ntotal,
+        )
+        return concresce.scatter(keys)
+
+    async def _remove(self, key: int) -> None:
+        """Remove an embedding from the index by its key."""
+        keys = await concresce.collect(key)
+        if self._index is None:
+            logger.warning(
+                "Ignoring removal of %d keys from an empty dense index", len(keys)
+            )
+        else:
+            ids = np.asarray(keys, dtype=np.int64)
+            removed = self._index.remove_ids(faiss.IDSelectorBatch(ids))
+            logger.debug(
+                "Removed %d of %d keys from the dense index; total is now %d",
+                removed,
+                len(keys),
+                self._index.ntotal,
+            )
+        return concresce.scatter([None] * len(keys))
+
+    async def _similar_batch(
+        self, embedding: DenseEmbedding, results: int
+    ) -> list[int]:
+        """Retrieve the most similar embedding keys to the given embedding."""
+        collected = await concresce.collect((embedding, results))
+        embeddings = [query for query, _ in collected]
+        counts = [count for _, count in collected]
+        if self._index is None:
+            logger.warning(
+                "Similarity search of %d queries on an empty dense index",
+                len(collected),
+            )
+            return concresce.scatter([[] for _ in collected])
+        logger.debug(
+            "Searching the dense index of %d embeddings with %d queries",
+            self._index.ntotal,
+            len(collected),
+        )
+        _, ids = self._index.search(self._to_rows(embeddings), max(counts))
+        rankings = [
+            [int(i) for i in row[:count] if i != -1]
+            for row, count in zip(ids, counts, strict=True)
+        ]
+        return concresce.scatter(rankings)
+
+
+class SparseIndex:
+    """Sparse embedding index using inner-product similarity over weight maps."""
+
+    def __init__(self) -> None:
+        """Initialize the sparse index."""
+        self._embeddings: dict[int, SparseEmbedding] = {}
+
+    @cached_property
+    def add(self) -> Callable[[SparseEmbedding], Awaitable[int]]:
+        """Add an embedding to the index and return its key."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._add)
+
+    @cached_property
+    def remove(self) -> Callable[[int], Awaitable[None]]:
+        """Remove an embedding from the index by its key."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._remove)
+
+    async def similar(
+        self, embedding: SparseEmbedding, results: int = 100
+    ) -> list[int]:
+        """Retrieve the keys of the ``results`` most similar embeddings."""
+        return await self._similar(embedding, results)
+
+    @cached_property
+    def _similar(self) -> Callable[[SparseEmbedding, int], Awaitable[list[int]]]:
+        """Coalesce concurrent queries into one batched ranking pass."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._similar_batch)
+
+    @staticmethod
+    def _key(embedding: SparseEmbedding) -> int:
+        """Derive a deterministic int64 key from the embedding's content."""
+        tokens = sorted(embedding)
+        weights = [embedding[token] for token in tokens]
+        data = np.asarray(tokens, dtype=np.int64).tobytes()
+        data += np.asarray(weights, dtype=np.float32).tobytes()
+        digest = blake3.blake3(data).digest()
+        return int.from_bytes(digest[:8], "little", signed=True)
+
+    @staticmethod
+    def _score(query: SparseEmbedding, stored: SparseEmbedding) -> float:
+        """Compute the inner product between two weight maps."""
+        if len(stored) < len(query):
+            query, stored = stored, query
+        return sum(weight * stored.get(token, 0.0) for token, weight in query.items())
+
+    def _top_keys(self, query: SparseEmbedding, results: int) -> list[int]:
+        """Rank stored keys by inner product with the query, dropping misses."""
+        scores = {
+            key: self._score(query, stored) for key, stored in self._embeddings.items()
+        }
+        ranked = sorted(scores, key=lambda key: scores[key], reverse=True)
+        return [key for key in ranked[:results] if scores[key] > 0]
+
+    async def _add(self, embedding: SparseEmbedding) -> int:
+        """Add an embedding to the index and return its key."""
+        embeddings = await concresce.collect(embedding)
+        keys = [self._key(e) for e in embeddings]
+        self._embeddings.update(zip(keys, embeddings, strict=True))
+        logger.debug(
+            "Added %d embeddings to the sparse index; total is now %d",
+            len(keys),
+            len(self._embeddings),
+        )
+        return concresce.scatter(keys)
+
+    async def _remove(self, key: int) -> None:
+        """Remove an embedding from the index by its key."""
+        keys = await concresce.collect(key)
+        removed = 0
+        for k in keys:
+            removed += self._embeddings.pop(k, None) is not None
+        logger.debug(
+            "Removed %d of %d keys from the sparse index; total is now %d",
+            removed,
+            len(keys),
+            len(self._embeddings),
+        )
+        return concresce.scatter([None] * len(keys))
+
+    async def _similar_batch(
+        self, embedding: SparseEmbedding, results: int
+    ) -> list[int]:
+        """Retrieve the most similar embedding keys to the given embedding."""
+        collected = await concresce.collect((embedding, results))
+        if not self._embeddings:
+            logger.warning(
+                "Similarity search of %d queries on an empty sparse index",
+                len(collected),
+            )
+        else:
+            logger.debug(
+                "Searching the sparse index of %d embeddings with %d queries",
+                len(self._embeddings),
+                len(collected),
+            )
+        rankings = [self._top_keys(query, count) for query, count in collected]
+        return concresce.scatter(rankings)

rager/parsers.py

@@ -0,0 +1,53 @@
+"""Parsers for extracting units from files."""
+
+import logging
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol
+
+import belljar
+import blake3
+from unstructured.partition.auto import partition
+
+if TYPE_CHECKING:
+    from rager.types import Hash
+
+logger = logging.getLogger(__name__)
+
+
+class Parser[F](Protocol):
+    """Protocol for parsers."""
+
+    def units(self, file: F) -> list[str]:
+        """Return the extracted text units."""
+        ...
+
+    def id(self, file: F) -> Hash:
+        """Return the file ID."""
+        ...
+
+
+class UnstructuredFileParser:
+    """Parser for unstructured files."""
+
+    @belljar.store(Path(".jar/parsers"))
+    def units(self, file: Path) -> list[str]:
+        """Extract text units from the unstructured file."""
+        file_id = self.id(file)
+        belljar.include(file_id.digest())
+        belljar.check()
+        logger.info("Cache miss; parsing file %s", file)
+        elements = partition(filename=str(file))
+        units = [element.text for element in elements]
+        logger.debug("Extracted %d units from %s", len(units), file)
+        return units
+
+    def id(self, file: Path) -> Hash:
+        """Return the content ID for the unstructured file."""
+        data = file.read_bytes()
+        logger.debug("Hashing %d bytes from %s", len(data), file)
+        return blake3.blake3(data)
+
+
+PdfFileParser = UnstructuredFileParser
+MarkdownFileParser = UnstructuredFileParser
+CsvFileParser = UnstructuredFileParser

rager/scorers.py

@@ -0,0 +1,64 @@
+"""Rerankers for scoring chunks based on relevance."""
+
+import logging
+from datetime import timedelta
+from functools import cached_property
+from pathlib import Path
+from typing import TYPE_CHECKING, Protocol
+
+import belljar
+import concresce
+from sentence_transformers import CrossEncoder
+
+if TYPE_CHECKING:
+    from collections.abc import Awaitable, Callable
+
+logger = logging.getLogger(__name__)
+
+
+class Scorer(Protocol):
+    """Protocol for chunk scorers."""
+
+    async def score(self, query: str, chunk: str) -> float:
+        """Score a chunk based on its semantic similarity to the query."""
+        ...
+
+
+class CrossEncoderScorer:
+    """Scorer that reranks chunks with a sentence-transformers cross-encoder."""
+
+    def __init__(self, model_name: str = "cross-encoder/ms-marco-MiniLM-L6-v2") -> None:
+        """Initialize the scorer with a specific model."""
+        self.model_name = model_name
+
+    @cached_property
+    def model(self) -> CrossEncoder:
+        """Load the CrossEncoder model."""
+        logger.info("Loading CrossEncoder model %r", self.model_name)
+        return CrossEncoder(self.model_name)
+
+    @cached_property
+    def _predict(self) -> Callable[[str, str], Awaitable[float]]:
+        """Coalesce concurrent calls into per-instance prediction batches."""
+        return concresce.batch(window=timedelta(milliseconds=1))(self._predict_batch)
+
+    async def _predict_batch(self, query: str, chunk: str) -> float:
+        """Score a query-chunk pair with the cross-encoder."""
+        pairs = await concresce.collect((query, chunk))
+        logger.debug(
+            "Scoring batch of %d query-chunk pairs with %r",
+            len(pairs),
+            self.model_name,
+        )
+        scores = self.model.predict(pairs).tolist()
+        return concresce.scatter(scores)
+
+    @belljar.store(Path(".jar/scorers"))
+    async def score(self, query: str, chunk: str) -> float:
+        """Score a chunk based on its semantic similarity to the query."""
+        belljar.include(self.model_name)
+        belljar.include(query)
+        belljar.include(chunk)
+        belljar.check()
+        logger.debug("Cache miss; scoring chunk of %d characters", len(chunk))
+        return await self._predict(query, chunk)

rager/stores.py

@@ -0,0 +1,103 @@
+"""Key-value store protocol definitions."""
+
+import logging
+from typing import Protocol
+
+from rager.types import DenseEmbedding, Metadata, SparseEmbedding
+
+logger = logging.getLogger(__name__)
+
+
+class Store[V, K](Protocol):
+    """Protocol for a key-value store mapping keys of type K to values of type V."""
+
+    def add(self, key: K, value: V) -> None:
+        """Store a value with the given key."""
+        ...
+
+    def get(self, key: K) -> V | None:
+        """Retrieve a value by its key."""
+        ...
+
+    def remove(self, key: K) -> None:
+        """Remove a value by its key."""
+        ...
+
+
+class EmbeddingStore[E]:
+    """In-memory store mapping index keys to embeddings."""
+
+    def __init__(self) -> None:
+        """Initialize the store with no embeddings."""
+        self._embeddings: dict[int, E] = {}
+
+    def add(self, key: int, value: E) -> None:
+        """Store an embedding with the given key."""
+        logger.debug("Storing embedding for key %d", key)
+        self._embeddings[key] = value
+
+    def get(self, key: int) -> E | None:
+        """Retrieve an embedding by its key."""
+        embedding = self._embeddings.get(key)
+        if embedding is None:
+            logger.debug("No embedding stored for key %d", key)
+        return embedding
+
+    def remove(self, key: int) -> None:
+        """Remove an embedding by its key."""
+        logger.debug("Removing embedding for key %d", key)
+        self._embeddings.pop(key, None)
+
+
+DenseEmbeddingStore = EmbeddingStore[DenseEmbedding]
+SparseEmbeddingStore = EmbeddingStore[SparseEmbedding]
+
+
+class ChunkStore:
+    """In-memory store mapping chunk indexes to chunk text."""
+
+    def __init__(self) -> None:
+        """Initialize the store with no chunks."""
+        self._chunks: dict[int, str] = {}
+
+    def add(self, key: int, value: str) -> None:
+        """Store chunk text with the given key."""
+        logger.debug("Storing chunk of %d characters for key %d", len(value), key)
+        self._chunks[key] = value
+
+    def get(self, key: int) -> str | None:
+        """Retrieve chunk text by its key."""
+        chunk = self._chunks.get(key)
+        if chunk is None:
+            logger.debug("No chunk stored for key %d", key)
+        return chunk
+
+    def remove(self, key: int) -> None:
+        """Remove chunk text by its key."""
+        logger.debug("Removing chunk for key %d", key)
+        self._chunks.pop(key, None)
+
+
+class MetadataStore[M: Metadata]:
+    """In-memory store mapping embedding keys to chunk metadata."""
+
+    def __init__(self) -> None:
+        """Initialize the store with no metadata."""
+        self._metadata: dict[int, M] = {}
+
+    def add(self, key: int, value: M) -> None:
+        """Store metadata with the given key."""
+        logger.debug("Storing metadata for key %d", key)
+        self._metadata[key] = value
+
+    def get(self, key: int) -> M | None:
+        """Retrieve metadata by its key."""
+        metadata = self._metadata.get(key)
+        if metadata is None:
+            logger.debug("No metadata stored for key %d", key)
+        return metadata
+
+    def remove(self, key: int) -> None:
+        """Remove metadata by its key."""
+        logger.debug("Removing metadata for key %d", key)
+        self._metadata.pop(key, None)

rager/types.py

@@ -0,0 +1,23 @@
+"""Type definitions for rager."""
+
+from typing import Protocol
+
+from blake3 import blake3
+
+Hash = blake3
+DenseEmbedding = list[float]
+SparseEmbedding = dict[int, float]
+
+
+class Metadata(Protocol):
+    """Protocol for the fields every chunk metadata class must provide."""
+
+    @property
+    def chunk(self) -> str:
+        """The chunk text this metadata describes."""
+        ...
+
+    @property
+    def file_id(self) -> Hash:
+        """The content ID of the file the chunk was extracted from."""
+        ...

rager-0.1.0.dist-info/METADATA

@@ -0,0 +1,67 @@
+Metadata-Version: 2.4
+Name: rager
+Version: 0.1.0
+Summary: Caching based RAG primitives
+Keywords: rag,retrieval-augmented-generation,caching,embeddings,faiss,nlp
+Author: Wannes Vantorre
+Author-email: Wannes Vantorre <vantorrewannes@gmail.com>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Developers
+Classifier: Operating System :: OS Independent
+Classifier: Programming Language :: Python :: 3
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Programming Language :: Python :: 3 :: Only
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Software Development :: Libraries :: Python Modules
+Classifier: Typing :: Typed
+Requires-Dist: belljar>=3.0.0
+Requires-Dist: blake3>=1.0.9
+Requires-Dist: concresce>=4.0.0
+Requires-Dist: faiss-cpu>=1.14.3
+Requires-Dist: numpy>=2.4.6
+Requires-Dist: semantic-chunker>=0.2.0
+Requires-Dist: sentence-transformers>=5.6.0
+Requires-Dist: tokenizers>=0.22.2
+Requires-Dist: transformers>=5.13.1
+Requires-Dist: unstructured-inference>=1.6.13
+Requires-Dist: unstructured[csv,md,pdf]>=0.24.0
+Requires-Dist: torch>=2.13.0 ; extra == 'cpu'
+Requires-Dist: torchvision>=0.28.0 ; extra == 'cpu'
+Requires-Dist: torch>=2.13.0 ; extra == 'cu130'
+Requires-Dist: torchvision>=0.28.0 ; extra == 'cu130'
+Requires-Dist: torch>=2.13.0 ; extra == 'rocm'
+Requires-Dist: torchvision>=0.28.0 ; extra == 'rocm'
+Requires-Dist: triton>=3.7.1 ; extra == 'rocm'
+Requires-Dist: triton-rocm>=2.13.0 ; extra == 'rocm'
+Maintainer: Wannes Vantorre
+Maintainer-email: Wannes Vantorre <vantorrewannes@gmail.com>
+Requires-Python: >=3.14
+Project-URL: Homepage, https://github.com/VantorreWannes/rager
+Project-URL: Repository, https://github.com/VantorreWannes/rager
+Project-URL: Issues, https://github.com/VantorreWannes/rager/issues
+Project-URL: Changelog, https://github.com/VantorreWannes/rager/releases
+Provides-Extra: cpu
+Provides-Extra: cu130
+Provides-Extra: rocm
+Description-Content-Type: text/markdown
+
+# rager
+
+Caching based RAG primitives — composable building blocks for retrieval-augmented generation with caching baked in.
+
+## Install
+
+```
+uv add rager
+```
+
+## Getting started
+
+```
+uv run pytest
+uv run prek install
+```
+
+Add an `--extra` to pin the default device: `rocm` (AMD), `cu130` (NVIDIA), or `cpu`. With `cpu`, torch supplies automatic fallback. For example: `uv run --extra cpu pytest`.

rager-0.1.0.dist-info/RECORD

@@ -0,0 +1,15 @@
+rager/__init__.py,sha256=gsRs2bdN9JTsMhl_uShsTb9JEy669_vCirAjLeFjWjg,1354
+rager/chunkers.py,sha256=TaMH31upTtcusOEz1prr5u4sYg2pdvXTY0bgQduVHTA,2087
+rager/embedders.py,sha256=8zPM5nFQ7jDl2_cKajB42sm6Sa4_hZeapJgDGo8xqgc,5031
+rager/fusers.py,sha256=IO5nrcBd-nCqq0FF291H8n6AcYSdTlMBDLEIjMOPcLw,1951
+rager/generators.py,sha256=azxIO5JLRvWt8H0-esVde59Vmtum1unFfxaD0ucAun8,2710
+rager/indexes.py,sha256=FiH7YBvas6a9Z9512Mwe8h6UPw_Xn_QLEooqP0gxBgU,10603
+rager/parsers.py,sha256=uyPYbbQT5dy8ksOK2Kus2sgcNRRPcAkJUjOCBYp8fZw,1478
+rager/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+rager/scorers.py,sha256=OhQMOi2XiXcmL7ffppQlAPZEZfHuWuaxqeNqvHpTTwM,2233
+rager/stores.py,sha256=oerEsb7BQsw4d1O_A9ey1UHHVTAckY4Pps_fIFKTmZw,3253
+rager/types.py,sha256=sylAee7jneBmmFbgu9iW6dc0N4jjl8N08mfsIDscKHA,510
+rager-0.1.0.dist-info/licenses/LICENSE,sha256=JYbLOUUlsHr9JxFTkfwg2e37zQtkf6a2VHw6wGJkEwI,11345
+rager-0.1.0.dist-info/WHEEL,sha256=CoDSoyhtC_eO_tlxRYzsTraPv1fPJRXFx91k6ISeAvA,81
+rager-0.1.0.dist-info/METADATA,sha256=9_zPWzt5NOPhXPyAreczUt51hLa6F6Rg42yndkEyrH4,2442
+rager-0.1.0.dist-info/RECORD,,

rager-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: uv 0.11.28
+Root-Is-Purelib: true
+Tag: py3-none-any

rager-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Wannes Vantorre
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.