speedy-utils 1.1.15__tar.gz → 1.1.17__tar.gz

{speedy_utils-1.1.15 → speedy_utils-1.1.17}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: speedy-utils
-Version: 1.1.15
+Version: 1.1.17
 Summary: Fast and easy-to-use package for data science
 Author: AnhVTH
 Author-email: anhvth.226@gmail.com

{speedy_utils-1.1.15 → speedy_utils-1.1.17}/pyproject.toml

@@ -1,6 +1,6 @@
 [tool.poetry]
 name = "speedy-utils"
-version = "1.1.15"
+version = "1.1.17"
 description = "Fast and easy-to-use package for data science"
 authors = ["AnhVTH <anhvth.226@gmail.com>"]
 readme = "README.md"

{speedy_utils-1.1.15 → speedy_utils-1.1.17}/src/llm_utils/lm/async_lm/async_llm_task.py

@@ -11,8 +11,6 @@ from venv import logger
 
 from openai.types.chat import ChatCompletionMessageParam
 from pydantic import BaseModel
-from pytest import Cache
-from speedy_utils import jdumps
 from speedy_utils.all import dump_json_or_pickle, identify
 
 from llm_utils.chat_format.display import get_conversation_one_turn

{speedy_utils-1.1.15 → speedy_utils-1.1.17}/src/llm_utils/vector_cache/core.py

@@ -167,12 +167,11 @@ class VectorCache:
         return "vllm"
     def _try_infer_model_name(self, model_name: Optional[str]) -> Optional[str]:
         """Infer model name for OpenAI backend if not explicitly provided."""
-        # if self.backend != "openai":
-            # return model_name
         if model_name:
             return model_name
         if 'https://' in self.url_or_model:
             model_name =  "text-embedding-3-small"
+        
         if 'http://localhost' in self.url_or_model:
             from openai import OpenAI
             client = OpenAI(base_url=self.url_or_model, api_key='abc')
@@ -277,6 +276,8 @@ class VectorCache:
 
     def _get_embeddings(self, texts: list[str]) -> list[list[float]]:
         """Get embeddings using the configured backend."""
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         if self.backend == "openai":
             return self._get_openai_embeddings(texts)
         elif self.backend == "vllm":
@@ -288,6 +289,8 @@ class VectorCache:
 
     def _get_openai_embeddings(self, texts: list[str]) -> list[list[float]]:
         """Get embeddings using OpenAI API."""
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         # Assert valid model_name for OpenAI backend
         model_name = self.config["model_name"]
         assert model_name is not None and model_name.strip(), f"Invalid model_name for OpenAI backend: {model_name}. Model name must be provided and non-empty."
@@ -304,6 +307,8 @@ class VectorCache:
 
     def _get_vllm_embeddings(self, texts: list[str]) -> list[list[float]]:
         """Get embeddings using vLLM."""
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         if self._model is None:
             self._load_model()
         
@@ -313,6 +318,8 @@ class VectorCache:
 
     def _get_transformers_embeddings(self, texts: list[str]) -> list[list[float]]:
         """Get embeddings using transformers directly."""
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         if self._model is None:
             self._load_model()
         
@@ -380,6 +387,8 @@ class VectorCache:
         handle very large input lists. A tqdm progress bar is shown while
         computing missing embeddings.
         """
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         if not texts:
             return np.empty((0, 0), dtype=np.float32)
         t = time()
@@ -436,6 +445,8 @@ class VectorCache:
         return np.vstack([hit_map[h] for h in hashes])
 
     def __call__(self, texts: list[str], cache: bool = True) -> np.ndarray:
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         return self.embeds(texts, cache)
 
     def _bulk_insert(self, data: list[tuple[str, str, bytes]]) -> None:
@@ -454,6 +465,8 @@ class VectorCache:
         Precompute embeddings for a large list of texts efficiently.
         This is optimized for bulk operations when you know all texts upfront.
         """
+        assert isinstance(texts, list), "texts must be a list"
+        assert all(isinstance(t, str) for t in texts), "all elements in texts must be strings"
         if not texts:
             return