TriCacheLLM-MMA 0.1.5__tar.gz → 0.1.6__tar.gz

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: TriCacheLLM_MMA
-Version: 0.1.5
+Version: 0.1.6
 Summary: A robust three-tier LLM-aware caching SDK using Redis, ChromaDB, and automatic SQLite bootstrap.
 Author-email: Mohib Ashfaq Butt <inboxmohib@gmail.com>
 License-Expression: LGPL-3.0-only
@@ -76,7 +76,6 @@ The internal infrastructure handles the multi-tier lookup and promotion logic.
 
 > **💡 Note for PyPI Visitors:** If you are reading this on PyPI, please check out the [TriCacheLLM_MMA GitHub Repository](https://github.com/mohib-ash/TriCacheLLM_MMA) for the most up-to-date documentation, integration guides, and advanced examples!
 
-
 ---
 
 # Architecture
@@ -172,7 +171,7 @@ Tier 2 is intentionally optimized for fast semantic cache retrieval.
 
 Tier 3 is the persistent cache layer.
 
-The persistent vector database acts as the long-term backing store for cached Q&A entries.
+The persistent vector database acts as the long-term backing store for cached Q\&A entries.
 
 When Tier 1 and Tier 2 miss, Tier 3 performs the persistent lookup.
 
@@ -293,8 +292,6 @@ Persistent cache seeded
 
 `populate_cache()` and `check_cache()` have intentionally different responsibilities.
 
-
-
 `populate_cache()` seeds the persistent cache VDB.
 
 It does **not** directly populate every cache tier.
@@ -427,7 +424,7 @@ Or install the development version directly from the repository:
 pip install .
 ```
 
-PyPI link: https://pypi.org/project/TriCacheLLM-MMA/0.1.3/
+PyPI link: [https://pypi.org/project/TriCacheLLM-MMA/0.1.6/](https://pypi.org/project/TriCacheLLM-MMA/0.1.6/)​
 
 ---
 
@@ -459,7 +456,7 @@ TriCacheLLM_MMA uses Celery for background operations such as:
 After installing the package, start the cache worker in a **separate terminal**:
 
 ```bash
-celery -A portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
+celery -A TriCacheLLM_MMA.portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
 ```
 
 You do **not** need to navigate into the package's `site-packages` directory.
@@ -571,13 +568,12 @@ The core cache operations are asynchronous Python APIs and are not inherently ti
 The included integration example uses FastAPI because it provides a convenient demonstration of multi-user request handling.
 
 V1 is primarily demonstrated in a web-service architecture, while future versions aim to make standalone application integration equally straightforward.
-***NOTE: On first initialization, the embedding model may contact Hugging Face to obtain the model if it is not already available in the local cache. Subsequent process reloads and reuse the locally cached model files Blazing fast.***
-
----
-
 
 
+> **Note on Model Initialization:**
+> On the first request or system initialization, the embedding model will load into RAM (fetching from Hugging Face if it's not already present in your local cache). Subsequent requests and code reloads reuse the cached instance instantly for blazing-fast performance.
 
+---
 
 # Cache Data Model
 
@@ -613,9 +609,6 @@ For example, you may want to restrict cache retrieval to:
 
 The persistent retrieval logic can be extended around the VDB lookup.
 
-
-
-
 This allows applications to combine semantic retrieval with deterministic metadata constraints.
 
 ---
@@ -626,8 +619,6 @@ Additional metadata can be added to the cache payload.
 
 The cache population path constructs metadata similar to:
 
-
-
 For example:
 
 ```python
@@ -880,8 +871,6 @@ These can be considered for future versions.
 
 ---
 
-
-
 # Consumer Responsibility
 
 The consuming application is responsible for:
@@ -938,7 +927,6 @@ The current V1 intentionally keeps the architecture close to the underlying impl
 
 ---
 
-
 # License
 
 This project is licensed under the **GNU Lesser General Public License v3.0 (LGPLv3)**. 

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/README.md

@@ -34,7 +34,6 @@ The internal infrastructure handles the multi-tier lookup and promotion logic.
 
 > **💡 Note for PyPI Visitors:** If you are reading this on PyPI, please check out the [TriCacheLLM_MMA GitHub Repository](https://github.com/mohib-ash/TriCacheLLM_MMA) for the most up-to-date documentation, integration guides, and advanced examples!
 
-
 ---
 
 # Architecture
@@ -130,7 +129,7 @@ Tier 2 is intentionally optimized for fast semantic cache retrieval.
 
 Tier 3 is the persistent cache layer.
 
-The persistent vector database acts as the long-term backing store for cached Q&A entries.
+The persistent vector database acts as the long-term backing store for cached Q\&A entries.
 
 When Tier 1 and Tier 2 miss, Tier 3 performs the persistent lookup.
 
@@ -251,8 +250,6 @@ Persistent cache seeded
 
 `populate_cache()` and `check_cache()` have intentionally different responsibilities.
 
-
-
 `populate_cache()` seeds the persistent cache VDB.
 
 It does **not** directly populate every cache tier.
@@ -385,7 +382,7 @@ Or install the development version directly from the repository:
 pip install .
 ```
 
-PyPI link: https://pypi.org/project/TriCacheLLM-MMA/0.1.3/
+PyPI link: [https://pypi.org/project/TriCacheLLM-MMA/0.1.6/](https://pypi.org/project/TriCacheLLM-MMA/0.1.6/)​
 
 ---
 
@@ -417,7 +414,7 @@ TriCacheLLM_MMA uses Celery for background operations such as:
 After installing the package, start the cache worker in a **separate terminal**:
 
 ```bash
-celery -A portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
+celery -A TriCacheLLM_MMA.portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
 ```
 
 You do **not** need to navigate into the package's `site-packages` directory.
@@ -529,13 +526,12 @@ The core cache operations are asynchronous Python APIs and are not inherently ti
 The included integration example uses FastAPI because it provides a convenient demonstration of multi-user request handling.
 
 V1 is primarily demonstrated in a web-service architecture, while future versions aim to make standalone application integration equally straightforward.
-***NOTE: On first initialization, the embedding model may contact Hugging Face to obtain the model if it is not already available in the local cache. Subsequent process reloads and reuse the locally cached model files Blazing fast.***
-
----
-
 
 
+> **Note on Model Initialization:**
+> On the first request or system initialization, the embedding model will load into RAM (fetching from Hugging Face if it's not already present in your local cache). Subsequent requests and code reloads reuse the cached instance instantly for blazing-fast performance.
 
+---
 
 # Cache Data Model
 
@@ -571,9 +567,6 @@ For example, you may want to restrict cache retrieval to:
 
 The persistent retrieval logic can be extended around the VDB lookup.
 
-
-
-
 This allows applications to combine semantic retrieval with deterministic metadata constraints.
 
 ---
@@ -584,8 +577,6 @@ Additional metadata can be added to the cache payload.
 
 The cache population path constructs metadata similar to:
 
-
-
 For example:
 
 ```python
@@ -838,8 +829,6 @@ These can be considered for future versions.
 
 ---
 
-
-
 # Consumer Responsibility
 
 The consuming application is responsible for:
@@ -896,7 +885,6 @@ The current V1 intentionally keeps the architecture close to the underlying impl
 
 ---
 
-
 # License
 
 This project is licensed under the **GNU Lesser General Public License v3.0 (LGPLv3)**. 
@@ -929,4 +917,4 @@ with the goal of making expensive LLM inference the **last resort rather than th
 ---
 
 # Disclaimer: 
-This software is provided "as is" without warranty of any kind. The author is not responsible for any data loss, system failures, or damages arising from its use.
+This software is provided "as is" without warranty of any kind. The author is not responsible for any data loss, system failures, or damages arising from its use.

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/TriCacheLLM_MMA/portable_cache_bgWorkers/portable_cache_celery_conf.py

@@ -34,7 +34,7 @@ celery_app = Celery(
 )
 
 celery_app.conf.imports = (
-    "portable_cache_bgWorkers.portable_cache_workers",
+    "TriCacheLLM_MMA.portable_cache_bgWorkers.portable_cache_workers",
 )
 
 celery_app.conf.task_default_queue = "default"

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/TriCacheLLM_MMA/portable_cache_bgWorkers/portable_cache_workers.py

@@ -46,7 +46,6 @@ async def create_cache_vdb_async(task_instance: Any, user_id: int):
                 parents=True,
                 exist_ok=True,
             )
-
             await asyncio.to_thread(
                 lambda: Chroma(
                     collection_name=f"question_cache_{user_id}",
@@ -142,7 +141,6 @@ async def push_response_in_cache_async(task_instance: Any, user_id, question: st
                 )
 
             user_cache_vdb_path = Path(cache_res.vdb_path)
-
             user_cache_vdb = await asyncio.to_thread(
                 lambda: Chroma(
                     collection_name=f"question_cache_{user_id}",

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/TriCacheLLM_MMA/portable_cache_main.py

@@ -524,7 +524,7 @@ async def populate_cache(to_cache_answer, to_cache_question, user_id: Any = 0) -
 async def close_cache_system():
     print("SHUTTING_DOWN_PORTABLE_CACHE: Cleaning up resources...")
 
-    from portable_cache_schemas.portable_cache_dbConf import db_manager
+    from .portable_cache_schemas.portable_cache_dbConf import db_manager
 
     if db_manager.norma_engine is not None:
         try:

tricachellm_mma-0.1.6/TriCacheLLM_MMA/portable_cache_utils/portable_cache_embedding_model.py

@@ -0,0 +1,31 @@
+# File: portable_cache_utils/portable_cache_embedding_model.py
+
+#on each --reload this reads weights
+from langchain_huggingface import HuggingFaceEmbeddings
+from ..portable_cache_utils.protable_cache_DynamicEnv_maker import get_settings
+
+embedding_model = HuggingFaceEmbeddings(
+    model_name=get_settings().portable_cache_cache_proj_embedding_model,
+    cache_folder=get_settings().portable_cache_chroma_db_dir
+)
+
+
+
+#lazy load, on --reload it will read weights but not at the start but when needed (un-comment me if u need lazyload)
+"""
+from langchain_huggingface import HuggingFaceEmbeddings
+from ..portable_cache_utils.protable_cache_DynamicEnv_maker import get_settings
+
+class EmbeddingModel:
+    def __init__(self):
+        self.model = None
+
+    def get_model(self):
+        if self.model is None:
+            self.model = HuggingFaceEmbeddings(
+                model_name=get_settings().portable_cache_cache_proj_embedding_model,
+                cache_folder=get_settings().portable_cache_chroma_db_dir,
+            )
+        return self.model
+embedding_manager = EmbeddingModel()
+"""

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/TriCacheLLM_MMA.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: TriCacheLLM_MMA
-Version: 0.1.5
+Version: 0.1.6
 Summary: A robust three-tier LLM-aware caching SDK using Redis, ChromaDB, and automatic SQLite bootstrap.
 Author-email: Mohib Ashfaq Butt <inboxmohib@gmail.com>
 License-Expression: LGPL-3.0-only
@@ -76,7 +76,6 @@ The internal infrastructure handles the multi-tier lookup and promotion logic.
 
 > **💡 Note for PyPI Visitors:** If you are reading this on PyPI, please check out the [TriCacheLLM_MMA GitHub Repository](https://github.com/mohib-ash/TriCacheLLM_MMA) for the most up-to-date documentation, integration guides, and advanced examples!
 
-
 ---
 
 # Architecture
@@ -172,7 +171,7 @@ Tier 2 is intentionally optimized for fast semantic cache retrieval.
 
 Tier 3 is the persistent cache layer.
 
-The persistent vector database acts as the long-term backing store for cached Q&A entries.
+The persistent vector database acts as the long-term backing store for cached Q\&A entries.
 
 When Tier 1 and Tier 2 miss, Tier 3 performs the persistent lookup.
 
@@ -293,8 +292,6 @@ Persistent cache seeded
 
 `populate_cache()` and `check_cache()` have intentionally different responsibilities.
 
-
-
 `populate_cache()` seeds the persistent cache VDB.
 
 It does **not** directly populate every cache tier.
@@ -427,7 +424,7 @@ Or install the development version directly from the repository:
 pip install .
 ```
 
-PyPI link: https://pypi.org/project/TriCacheLLM-MMA/0.1.3/
+PyPI link: [https://pypi.org/project/TriCacheLLM-MMA/0.1.6/](https://pypi.org/project/TriCacheLLM-MMA/0.1.6/)​
 
 ---
 
@@ -459,7 +456,7 @@ TriCacheLLM_MMA uses Celery for background operations such as:
 After installing the package, start the cache worker in a **separate terminal**:
 
 ```bash
-celery -A portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
+celery -A TriCacheLLM_MMA.portable_cache_bgWorkers.portable_cache_celery_conf.celery_app worker --loglevel=info -Q ai
 ```
 
 You do **not** need to navigate into the package's `site-packages` directory.
@@ -571,13 +568,12 @@ The core cache operations are asynchronous Python APIs and are not inherently ti
 The included integration example uses FastAPI because it provides a convenient demonstration of multi-user request handling.
 
 V1 is primarily demonstrated in a web-service architecture, while future versions aim to make standalone application integration equally straightforward.
-***NOTE: On first initialization, the embedding model may contact Hugging Face to obtain the model if it is not already available in the local cache. Subsequent process reloads and reuse the locally cached model files Blazing fast.***
-
----
-
 
 
+> **Note on Model Initialization:**
+> On the first request or system initialization, the embedding model will load into RAM (fetching from Hugging Face if it's not already present in your local cache). Subsequent requests and code reloads reuse the cached instance instantly for blazing-fast performance.
 
+---
 
 # Cache Data Model
 
@@ -613,9 +609,6 @@ For example, you may want to restrict cache retrieval to:
 
 The persistent retrieval logic can be extended around the VDB lookup.
 
-
-
-
 This allows applications to combine semantic retrieval with deterministic metadata constraints.
 
 ---
@@ -626,8 +619,6 @@ Additional metadata can be added to the cache payload.
 
 The cache population path constructs metadata similar to:
 
-
-
 For example:
 
 ```python
@@ -880,8 +871,6 @@ These can be considered for future versions.
 
 ---
 
-
-
 # Consumer Responsibility
 
 The consuming application is responsible for:
@@ -938,7 +927,6 @@ The current V1 intentionally keeps the architecture close to the underlying impl
 
 ---
 
-
 # License
 
 This project is licensed under the **GNU Lesser General Public License v3.0 (LGPLv3)**. 

{tricachellm_mma-0.1.5 → tricachellm_mma-0.1.6}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "TriCacheLLM_MMA"
-version = "0.1.5"
+version = "0.1.6"
 description = "A robust three-tier LLM-aware caching SDK using Redis, ChromaDB, and automatic SQLite bootstrap."
 readme = "README.md"
 requires-python = ">=3.11"

tricachellm_mma-0.1.5/TriCacheLLM_MMA/portable_cache_utils/portable_cache_embedding_model.py

@@ -1,8 +0,0 @@
-# # File: portable_cache_utils/portable_cache_embedding_model.py
-from langchain_huggingface import HuggingFaceEmbeddings
-from ..portable_cache_utils.protable_cache_DynamicEnv_maker import get_settings
-
-embedding_model = HuggingFaceEmbeddings(
-    model_name=get_settings().portable_cache_cache_proj_embedding_model,
-    cache_folder=get_settings().portable_cache_chroma_db_dir
-)