deeplotx 0.5.3__tar.gz → 0.5.6__tar.gz

{deeplotx-0.5.3 → deeplotx-0.5.6}/PKG-INFO

@@ -1,18 +1,19 @@
 Metadata-Version: 2.4
 Name: deeplotx
-Version: 0.5.3
+Version: 0.5.6
 Summary: Easy-2-use long text NLP toolkit.
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: hf-xet>=1.0.5
+Requires-Dist: hf-xet
 Requires-Dist: jupyter
 Requires-Dist: numpy
 Requires-Dist: protobuf
-Requires-Dist: python-dotenv>=1.1.0
+Requires-Dist: python-dotenv
 Requires-Dist: torch
 Requires-Dist: transformers
-Requires-Dist: typing-extensions>=4.13.2
+Requires-Dist: typing-extensions
+Requires-Dist: vortezwohl>=0.0.6
 Dynamic: license-file
 
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/vortezwohl/DeepLoTX)
@@ -264,7 +265,8 @@ Dynamic: license-file
     long_text_encoder = LongTextEncoder(
         max_length=2048,  # 最大文本大小, 超出截断
         chunk_size=448,  # 块大小 (按 Token 计)
-        overlapping=32  # 块间重叠大小 (按 Token 计)
+        overlapping=32,  # 块间重叠大小 (按 Token 计)
+        cache_capacity=512  # 缓存大小
     )
 
     trainer = TextBinaryClassifierTrainer(

{deeplotx-0.5.3 → deeplotx-0.5.6}/README.md

@@ -247,7 +247,8 @@
     long_text_encoder = LongTextEncoder(
         max_length=2048,  # 最大文本大小, 超出截断
         chunk_size=448,  # 块大小 (按 Token 计)
-        overlapping=32  # 块间重叠大小 (按 Token 计)
+        overlapping=32,  # 块间重叠大小 (按 Token 计)
+        cache_capacity=512  # 缓存大小
     )
 
     trainer = TextBinaryClassifierTrainer(

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx/__init__.py

@@ -3,7 +3,7 @@ import os
 
 __ROOT__ = os.path.dirname(os.path.abspath(__file__))
 
-from .encoder import BertEncoder, LongTextEncoder, LongformerEncoder
+from .encoder import Encoder, LongTextEncoder, LongformerEncoder
 from .nn import (
     LinearRegression,
     LogisticRegression,

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx/encoder/__init__.py

@@ -1,3 +1,3 @@
-from .bert_encoder import BertEncoder
+from .encoder import Encoder
 from .long_text_encoder import LongTextEncoder
 from .longformer_encoder import LongformerEncoder

deeplotx-0.5.3/deeplotx/encoder/bert_encoder.py → deeplotx-0.5.6/deeplotx/encoder/encoder.py

@@ -4,30 +4,30 @@ import math
 
 import torch
 from torch import nn
-from transformers import BertTokenizer, BertModel
+from transformers import AutoTokenizer, AutoModel
 
 from deeplotx import __ROOT__
 
 CACHE_PATH = os.path.join(__ROOT__, '.cache')
-DEFAULT_BERT = 'bert-base-uncased'
+DEFAULT_BERT = 'FacebookAI/xlm-roberta-base'
 logger = logging.getLogger('deeplotx.embedding')
 
 
-class BertEncoder(nn.Module):
+class Encoder(nn.Module):
     def __init__(self, model_name_or_path: str = DEFAULT_BERT, device: str | None = None):
         super().__init__()
         self.device = torch.device(device) if device is not None \
             else torch.device('cuda' if torch.cuda.is_available() else 'cpu')
-        self.tokenizer = BertTokenizer.from_pretrained(pretrained_model_name_or_path=model_name_or_path,
+        self.tokenizer = AutoTokenizer.from_pretrained(pretrained_model_name_or_path=model_name_or_path,
                                                        cache_dir=CACHE_PATH, _from_auto=True)
-        self.bert = BertModel.from_pretrained(pretrained_model_name_or_path=model_name_or_path,
-                                              cache_dir=CACHE_PATH, _from_auto=True).to(self.device)
-        self.embed_dim = self.bert.config.max_position_embeddings
-        logger.debug(f'{BertEncoder.__name__} initialized on device: {self.device}.')
+        self.encoder = AutoModel.from_pretrained(pretrained_model_name_or_path=model_name_or_path,
+                                                 cache_dir=CACHE_PATH, _from_auto=True).to(self.device)
+        self.embed_dim = self.encoder.config.max_position_embeddings
+        logger.debug(f'{Encoder.__name__} initialized on device: {self.device}.')
 
-    def forward(self, input_ids: torch.Tensor, attention_mask: torch.Tensor) -> torch.Tensor:
+    def forward(self, input_ids: torch.Tensor, attention_mask: torch.Tensor, *args, **kwargs) -> torch.Tensor:
         def _encoder(_input_tup: tuple[torch.Tensor, torch.Tensor]) -> torch.Tensor:
-            return self.bert.forward(_input_tup[0], attention_mask=_input_tup[1]).last_hidden_state[:, 0, :]
+            return self.encoder.forward(_input_tup[0], attention_mask=_input_tup[1]).last_hidden_state[:, 0, :]
 
         num_chunks = math.ceil(input_ids.shape[-1] / self.embed_dim)
         chunks = chunk_results = []
@@ -35,11 +35,11 @@ class BertEncoder(nn.Module):
             start_idx = i * self.embed_dim
             end_idx = min(start_idx + self.embed_dim, input_ids.shape[-1])
             chunks.append((input_ids[:, start_idx: end_idx], attention_mask[:, start_idx: end_idx]))
-        ori_mode = self.bert.training
-        self.bert.eval()
+        ori_mode = self.encoder.training
+        self.encoder.eval()
         with torch.no_grad():
             chunk_results = [_encoder(x) for x in chunks]
-        self.bert.train(mode=ori_mode)
+        self.encoder.train(mode=ori_mode)
         return torch.cat(chunk_results, dim=-1)
 
     def encode(self, text: str) -> torch.Tensor:

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx/encoder/long_text_encoder.py

@@ -1,30 +1,36 @@
 import logging
 import math
-from concurrent.futures import ThreadPoolExecutor
 from typing_extensions import override
 
 import torch
+from vortezwohl.concurrent import ThreadPool
+from vortezwohl.cache import LRUCache
 
-from deeplotx.encoder.bert_encoder import BertEncoder, DEFAULT_BERT
-from deeplotx.util.hash import md5
+from deeplotx.encoder.encoder import Encoder, DEFAULT_BERT
+from deeplotx.util.hash import sha512
 
 logger = logging.getLogger('deeplotx.embedding')
 
 
-class LongTextEncoder(BertEncoder):
-    def __init__(self, max_length: int, chunk_size: int = 256,
-                 overlapping: int = 0, model_name_or_path: str = DEFAULT_BERT, device: str | None = None):
+class LongTextEncoder(Encoder):
+    def __init__(self, max_length: int, chunk_size: int = 448,
+                 overlapping: int = 32, model_name_or_path: str = DEFAULT_BERT,
+                 cache_capacity: int = 64, device: str | None = None):
         super().__init__(model_name_or_path=model_name_or_path, device=device)
         self._max_length = max_length
         self._chunk_size = chunk_size
         self._overlapping = overlapping
-        self._cache = dict()
+        self._cache = LRUCache(capacity=cache_capacity)
 
-    def __chunk_embedding(self, input_tup: tuple[int, torch.Tensor]) -> tuple[int, torch.Tensor]:
-        return input_tup[0], super().forward(input_tup[1], attention_mask=input_tup[2])
+    def __chunk_embedding(self, idx: int, x: torch.Tensor, mask: torch.Tensor) -> tuple[int, torch.Tensor]:
+        return idx, super().forward(x, attention_mask=mask)
 
     @override
-    def encode(self, text: str, flatten: bool = True, use_cache: bool = True) -> torch.Tensor:
+    def forward(self, text: str, flatten: bool = False, *args, **kwargs) -> torch.Tensor:
+        return self.encode(text=text, flatten=flatten)
+
+    @override
+    def encode(self, text: str, flatten: bool = False) -> torch.Tensor:
         def postprocess(tensors: list[torch.Tensor], _flatten: bool) -> torch.Tensor:
             if not _flatten:
                 return torch.stack(tensors, dim=0).squeeze()
@@ -36,8 +42,8 @@ class LongTextEncoder(BertEncoder):
         _text_to_show = text.replace("\n", str())
         logger.debug(f'Embedding \"{_text_to_show if len(_text_to_show) < 128 else _text_to_show[:128] + "..."}\".')
         # read cache
-        _text_hash = md5(text)
-        if _text_hash in self._cache.keys():
+        _text_hash = sha512(text)
+        if _text_hash in self._cache:
             return postprocess(self._cache[_text_hash], flatten)
         _text_to_input_ids = self.tokenizer.encode(text.strip())[:self._max_length]
         _text_to_input_ids_att_mask = []
@@ -57,11 +63,9 @@ class LongTextEncoder(BertEncoder):
             _tmp_right = (i + 1) * self._chunk_size + self._overlapping
             chunks.append((i, torch.tensor([_text_to_input_ids[_tmp_left: _tmp_right]], dtype=torch.int, device=self.device),
                            torch.tensor([_text_to_input_ids_att_mask[_tmp_left: _tmp_right]], dtype=torch.int, device=self.device)))
-        with ThreadPoolExecutor(max_workers=min(num_chunks + 1, 3)) as executor:
-            embeddings = list(executor.map(self.__chunk_embedding, chunks))
-        embeddings.sort(key=lambda x: x[0])
+        embeddings = list(ThreadPool(max_workers=min(num_chunks + 1, 8)).map(self.__chunk_embedding, chunks))
+        embeddings = sorted([x.returns for x in embeddings], key=lambda x: x[0], reverse=False)
         fin_embedding = [x[1] for x in embeddings]
         # write cache
-        if use_cache:
-            self._cache[_text_hash] = fin_embedding
+        self._cache[_text_hash] = fin_embedding
         return postprocess(fin_embedding, flatten)

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx/trainer/text_binary_classification_trainer.py

@@ -31,7 +31,7 @@ class TextBinaryClassifierTrainer(BaseTrainer):
             positive_texts = positive_texts[:min_length]
             negative_texts = negative_texts[:min_length]
         all_texts = positive_texts + negative_texts
-        text_embeddings = [self._long_text_encoder.encode(x, flatten=False, use_cache=True) for x in all_texts]
+        text_embeddings = [self._long_text_encoder.encode(x, flatten=False) for x in all_texts]
         feature_dim = text_embeddings[0].shape[-1]
         dtype = text_embeddings[0].dtype
         labels = ([torch.tensor([1.], dtype=dtype, device=self.device) for _ in range(len(positive_texts))]

deeplotx-0.5.6/deeplotx/util/hash.py

@@ -0,0 +1,29 @@
+import hashlib
+
+
+def md5(text: str) -> str:
+    _hash = hashlib.md5()
+    text_bytes = text.encode('utf-8')
+    _hash.update(text_bytes)
+    return _hash.hexdigest()
+
+
+def sha1(text: str) -> str:
+    _hash = hashlib.sha1()
+    text_bytes = text.encode('utf-8')
+    _hash.update(text_bytes)
+    return _hash.hexdigest()
+
+
+def sha256(text: str) -> str:
+    _hash = hashlib.sha256()
+    text_bytes = text.encode('utf-8')
+    _hash.update(text_bytes)
+    return _hash.hexdigest()
+
+
+def sha512(text: str) -> str:
+    _hash = hashlib.sha512()
+    text_bytes = text.encode('utf-8')
+    _hash.update(text_bytes)
+    return _hash.hexdigest()

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx.egg-info/PKG-INFO

@@ -1,18 +1,19 @@
 Metadata-Version: 2.4
 Name: deeplotx
-Version: 0.5.3
+Version: 0.5.6
 Summary: Easy-2-use long text NLP toolkit.
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: hf-xet>=1.0.5
+Requires-Dist: hf-xet
 Requires-Dist: jupyter
 Requires-Dist: numpy
 Requires-Dist: protobuf
-Requires-Dist: python-dotenv>=1.1.0
+Requires-Dist: python-dotenv
 Requires-Dist: torch
 Requires-Dist: transformers
-Requires-Dist: typing-extensions>=4.13.2
+Requires-Dist: typing-extensions
+Requires-Dist: vortezwohl>=0.0.6
 Dynamic: license-file
 
 [![Ask DeepWiki](https://deepwiki.com/badge.svg)](https://deepwiki.com/vortezwohl/DeepLoTX)
@@ -264,7 +265,8 @@ Dynamic: license-file
     long_text_encoder = LongTextEncoder(
         max_length=2048,  # 最大文本大小, 超出截断
         chunk_size=448,  # 块大小 (按 Token 计)
-        overlapping=32  # 块间重叠大小 (按 Token 计)
+        overlapping=32,  # 块间重叠大小 (按 Token 计)
+        cache_capacity=512  # 缓存大小
     )
 
     trainer = TextBinaryClassifierTrainer(

{deeplotx-0.5.3 → deeplotx-0.5.6}/deeplotx.egg-info/SOURCES.txt

@@ -8,7 +8,7 @@ deeplotx.egg-info/dependency_links.txt
 deeplotx.egg-info/requires.txt
 deeplotx.egg-info/top_level.txt
 deeplotx/encoder/__init__.py
-deeplotx/encoder/bert_encoder.py
+deeplotx/encoder/encoder.py
 deeplotx/encoder/long_text_encoder.py
 deeplotx/encoder/longformer_encoder.py
 deeplotx/nn/__init__.py

deeplotx-0.5.6/deeplotx.egg-info/requires.txt

@@ -0,0 +1,9 @@
+hf-xet
+jupyter
+numpy
+protobuf
+python-dotenv
+torch
+transformers
+typing-extensions
+vortezwohl>=0.0.6

{deeplotx-0.5.3 → deeplotx-0.5.6}/pyproject.toml

@@ -1,16 +1,17 @@
-[project]
-name = "deeplotx"
-version = "0.5.3"
-description = "Easy-2-use long text NLP toolkit."
-readme = "README.md"
-requires-python = ">=3.10"
-dependencies = [
-    "hf-xet>=1.0.5",
-    "jupyter",
-    "numpy",
-    "protobuf",
-    "python-dotenv>=1.1.0",
-    "torch",
-    "transformers",
-    "typing-extensions>=4.13.2",
-]
+[project]
+name = "deeplotx"
+version = "0.5.6"
+description = "Easy-2-use long text NLP toolkit."
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = [
+    "hf-xet",
+    "jupyter",
+    "numpy",
+    "protobuf",
+    "python-dotenv",
+    "torch",
+    "transformers",
+    "typing-extensions",
+    "vortezwohl>=0.0.6",
+]

deeplotx-0.5.3/deeplotx/util/hash.py

@@ -1,15 +0,0 @@
-import hashlib
-
-
-def md5(text: str) -> str:
-    md5_hash = hashlib.md5()
-    text_bytes = text.encode('utf-8')
-    md5_hash.update(text_bytes)
-    return md5_hash.hexdigest()
-
-
-def sha1(text: str) -> str:
-    md5_hash = hashlib.sha1()
-    text_bytes = text.encode('utf-8')
-    md5_hash.update(text_bytes)
-    return md5_hash.hexdigest()

deeplotx-0.5.3/deeplotx.egg-info/requires.txt

@@ -1,8 +0,0 @@
-hf-xet>=1.0.5
-jupyter
-numpy
-protobuf
-python-dotenv>=1.1.0
-torch
-transformers
-typing-extensions>=4.13.2