unsloth/studio/backend/core/rag/embeddings.py

456 lines
14 KiB
Python

# SPDX-License-Identifier: AGPL-3.0-only
# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
"""RAG embedder singleton. Independent of the chat InferenceBackend."""
from __future__ import annotations
import logging
import threading
from typing import Any
from utils.rag.config import RAG_EMBED_BATCH_SIZE, RAG_EMBEDDING_MODEL
logger = logging.getLogger(__name__)
_lock = threading.Lock()
_model: Any | None = None
_model_name: str | None = None
_embedding_dim: int | None = None
def _load(model_name: str) -> Any:
logger.info("Loading RAG embedder: %s", model_name)
# BGE-VL's ST shim breaks across ST versions; load via AutoModel.
if model_name.startswith("BAAI/BGE-VL"):
return _BGEVLAdapter(model_name)
from unsloth import FastSentenceTransformer
# trust_remote_code: nomic-embed-text-v1.5 needs custom modeling for 8K ctx.
return FastSentenceTransformer.from_pretrained(
model_name,
for_inference = True,
trust_remote_code = True,
)
class _BGEVLAdapter:
"""SentenceTransformer-shaped adapter over BGE-VL's AutoModel."""
def __init__(self, hf_model_name: str):
from transformers import AutoModel
import torch
self._model = AutoModel.from_pretrained(
hf_model_name,
trust_remote_code = True,
)
# Required: BGE-VL's encode() raises without an installed processor.
self._model.set_processor(hf_model_name)
device = "cuda" if torch.cuda.is_available() else "cpu"
self._model.to(device).eval()
self._device = device
self._dim: int | None = None
def _normalize(self, tensor):
import torch.nn.functional as F
return F.normalize(tensor, p = 2.0, dim = -1)
# CLIP positional embedding cap; longer text triggers shape mismatch.
_CLIP_TEXT_MAX_TOKENS = 77
def encode(
self,
inputs,
*,
batch_size: int = 32,
normalize_embeddings: bool = True,
convert_to_numpy: bool = True,
show_progress_bar: bool = False,
**_ignored,
):
import io
import numpy as np
import torch
from PIL import Image
if inputs is None or len(inputs) == 0:
return np.zeros(
(0, self.get_sentence_embedding_dimension()), dtype = np.float32
)
sample = inputs[0]
is_image = isinstance(sample, Image.Image) or isinstance(
sample, (bytes, bytearray)
)
chunks_out = []
for start in range(0, len(inputs), batch_size):
batch = list(inputs[start : start + batch_size])
if is_image:
# BGE-VL's internal data_process re-opens each item with
# Image.open(...), which needs a file-like (has .read())
# or a path — NOT a pre-opened PIL Image. Pass BytesIO so
# the model's own opener works. PIL Images get rebuffered
# via an in-memory PNG round-trip.
file_likes: list[Any] = []
for b in batch:
if isinstance(b, (bytes, bytearray)):
file_likes.append(io.BytesIO(b))
elif isinstance(b, Image.Image):
buf = io.BytesIO()
b.save(buf, format = "PNG")
buf.seek(0)
file_likes.append(buf)
else:
file_likes.append(b)
with torch.no_grad():
vecs = self._model.encode(images = file_likes)
else:
vecs = self._encode_text_truncated([str(t) for t in batch])
if normalize_embeddings:
vecs = self._normalize(vecs)
chunks_out.append(vecs.detach().cpu())
out = torch.cat(chunks_out, dim = 0)
return out.numpy() if convert_to_numpy else out
def _encode_text_truncated(self, texts: list[str]):
"""Truncate to CLIP's 77-token limit; long text in multimodal mode is lossy."""
import torch
tokenizer = self._get_text_tokenizer()
inputs = tokenizer(
texts,
return_tensors = "pt",
padding = True,
truncation = True,
max_length = self._CLIP_TEXT_MAX_TOKENS,
)
inputs = {k: v.to(self._device) for k, v in inputs.items()}
if any(len(t.split()) > 30 for t in texts):
logger.info(
"BGE-VL text encode: truncating chunks to %d tokens (CLIP cap)",
self._CLIP_TEXT_MAX_TOKENS,
)
with torch.no_grad():
return self._model.get_text_features(**inputs)
def _get_text_tokenizer(self):
processor = getattr(self._model, "processor", None)
if processor is not None:
tok = getattr(processor, "tokenizer", None)
if tok is not None:
return tok
tok = getattr(self._model, "tokenizer", None)
if tok is not None:
return tok
raise AttributeError("BGE-VL adapter could not locate a text tokenizer")
def get_sentence_embedding_dimension(self) -> int:
if self._dim is None:
v = self.encode(["dim-probe"], batch_size = 1)
self._dim = int(v.shape[-1])
return self._dim
def tokenize(self, texts):
return self._get_text_tokenizer()(
texts,
return_tensors = "pt",
padding = True,
)
def get_embedder(model_name: str | None = None) -> Any:
global _model, _model_name, _embedding_dim
target = model_name or RAG_EMBEDDING_MODEL
with _lock:
if _model is None or _model_name != target:
_model = _load(target)
_model_name = target
try:
_embedding_dim = int(_model.get_sentence_embedding_dimension())
except Exception:
_embedding_dim = None
return _model
def get_embedding_dim(model_name: str | None = None) -> int:
model = get_embedder(model_name)
global _embedding_dim
if _embedding_dim is None:
_embedding_dim = int(model.get_sentence_embedding_dimension())
return _embedding_dim
def get_active_model_name() -> str | None:
return _model_name
def encode(
texts: list[str],
*,
model_name: str | None = None,
batch_size: int | None = None,
normalize: bool = True,
):
model = get_embedder(model_name)
return model.encode(
texts,
batch_size = batch_size or RAG_EMBED_BATCH_SIZE,
normalize_embeddings = normalize,
convert_to_numpy = True,
show_progress_bar = False,
)
def encode_images(
image_bytes_list: list[bytes],
*,
model_name: str | None = None,
batch_size: int | None = None,
normalize: bool = True,
):
"""Embed image bytes via a CLIP-family multimodal encoder."""
from io import BytesIO
from PIL import Image
if not image_bytes_list:
return []
model = get_embedder(model_name)
images = [Image.open(BytesIO(b)).convert("RGB") for b in image_bytes_list]
return model.encode(
images,
batch_size = batch_size or RAG_EMBED_BATCH_SIZE,
normalize_embeddings = normalize,
convert_to_numpy = True,
show_progress_bar = False,
)
def token_counter(model_name: str | None = None):
"""Return a token-count callable backed by the embedder's tokenizer."""
model = get_embedder(model_name)
def _count(text: str) -> int:
try:
tokens = model.tokenize([text])
ids = tokens.get("input_ids")
if ids is None:
return max(1, len(text) // 4)
return int(ids.shape[1])
except Exception:
return max(1, len(text) // 4)
return _count
# --- Late chunking (Jina technique) ---
_LATE_WINDOW_OVERLAP_TOKENS = 512
def late_chunk_encode(
doc_text: str,
char_spans: list[tuple[int, int]],
*,
model_name: str | None = None,
normalize: bool = True,
):
"""Single forward pass over the doc, mean-pool token embeddings per chunk span."""
import numpy as np
if not char_spans:
return []
model = get_embedder(model_name)
tokenizer = model.tokenizer
max_length = int(getattr(model, "max_seq_length", None) or 8192)
encoded = tokenizer(
doc_text,
return_tensors = "pt",
return_offsets_mapping = True,
add_special_tokens = True,
truncation = False,
)
offsets = encoded.pop("offset_mapping")[0].tolist()
n_tokens = int(encoded["input_ids"].shape[1])
if n_tokens <= max_length:
token_embeddings = _encode_tokens(model, encoded)
return _pool_spans(
token_embeddings,
offsets,
char_spans,
normalize = normalize,
np_module = np,
model = model,
doc_text = doc_text,
)
logger.info(
"Late chunking: doc has %d tokens > model max %d; using windowed pass",
n_tokens,
max_length,
)
return _windowed_late_chunk_encode(
doc_text = doc_text,
char_spans = char_spans,
model = model,
max_length = max_length,
normalize = normalize,
np_module = np,
)
def _encode_tokens(model, encoded):
import torch
transformer = model[0].auto_model
device = next(transformer.parameters()).device
inputs_on_device = {k: v.to(device) for k, v in encoded.items()}
with torch.no_grad():
outputs = transformer(**inputs_on_device)
return outputs.last_hidden_state[0].detach().cpu().numpy()
def _pool_spans(
token_embeddings,
offsets,
char_spans,
*,
normalize: bool,
np_module,
model,
doc_text: str,
token_index_offset: int = 0,
):
"""Mean-pool token embeddings per (char_start, char_end) span."""
vectors = []
n_rows = token_embeddings.shape[0]
for char_start, char_end in char_spans:
# Skip special tokens whose offsets are (0, 0).
indices = [
i - token_index_offset
for i, (ts, te) in enumerate(offsets)
if te > ts and te > char_start and ts < char_end
]
indices = [i for i in indices if 0 <= i < n_rows]
if not indices:
vec = model.encode(
doc_text[char_start:char_end],
normalize_embeddings = normalize,
convert_to_numpy = True,
show_progress_bar = False,
)
vectors.append(vec)
continue
pooled = token_embeddings[indices].mean(axis = 0)
if normalize:
denom = float(np_module.linalg.norm(pooled))
if denom > 0:
pooled = pooled / denom
vectors.append(pooled)
return vectors
def _windowed_late_chunk_encode(
*,
doc_text: str,
char_spans: list[tuple[int, int]],
model,
max_length: int,
normalize: bool,
np_module,
):
"""Doc > ctx window: pool each chunk against the window containing most of its tokens."""
import torch
tokenizer = model.tokenizer
transformer = model[0].auto_model
device = next(transformer.parameters()).device
full = tokenizer(
doc_text,
return_tensors = "pt",
return_offsets_mapping = True,
add_special_tokens = False,
truncation = False,
)
all_input_ids = full["input_ids"][0]
all_offsets = full["offset_mapping"][0].tolist()
n_tokens = int(all_input_ids.shape[0])
stride = max(1, max_length - _LATE_WINDOW_OVERLAP_TOKENS)
windows: list[tuple[int, int]] = []
pos = 0
while pos < n_tokens:
end = min(pos + max_length, n_tokens)
windows.append((pos, end))
if end >= n_tokens:
break
pos += stride
window_embeddings: dict[int, "np_module.ndarray"] = {}
def _window_embeddings(window_index: int):
if window_index in window_embeddings:
return window_embeddings[window_index]
ws, we = windows[window_index]
win_ids = all_input_ids[ws:we].unsqueeze(0).to(device)
win_attn = torch.ones_like(win_ids)
with torch.no_grad():
outputs = transformer(input_ids = win_ids, attention_mask = win_attn)
emb = outputs.last_hidden_state[0].detach().cpu().numpy()
window_embeddings[window_index] = emb
return emb
vectors = []
for char_start, char_end in char_spans:
chunk_token_indices = [
i
for i, (ts, te) in enumerate(all_offsets)
if te > ts and te > char_start and ts < char_end
]
if not chunk_token_indices:
vec = model.encode(
doc_text[char_start:char_end],
normalize_embeddings = normalize,
convert_to_numpy = True,
show_progress_bar = False,
)
vectors.append(vec)
continue
best_window = 0
best_overlap = 0
for wi, (ws, we) in enumerate(windows):
overlap = sum(1 for ti in chunk_token_indices if ws <= ti < we)
if overlap > best_overlap:
best_overlap = overlap
best_window = wi
ws, _we = windows[best_window]
emb = _window_embeddings(best_window)
local_indices = [
ti - ws for ti in chunk_token_indices if ws <= ti < ws + emb.shape[0]
]
if not local_indices:
vec = model.encode(
doc_text[char_start:char_end],
normalize_embeddings = normalize,
convert_to_numpy = True,
show_progress_bar = False,
)
vectors.append(vec)
continue
pooled = emb[local_indices].mean(axis = 0)
if normalize:
denom = float(np_module.linalg.norm(pooled))
if denom > 0:
pooled = pooled / denom
vectors.append(pooled)
return vectors