Spaces:
Sleeping
Sleeping
| """EvoLLM knowledge layer — RAG + LoRA-on-upload. | |
| Components: | |
| - parser : PDF / TXT / MD / DOCX → clean text | |
| - chunker : long text → overlapping ~400-token chunks | |
| - embedder : fastembed wrapper (no torch dependency) | |
| - store : SQLite + numpy vector store | |
| - pipeline : high-level ingest / query API | |
| - dataset_builder : doc chunks → training JSONL | |
| - notebook_generator : produce a Colab notebook with the dataset baked in | |
| - adapter_importer : accept uploaded GGUF + manifest, register an adapter | |
| Local persistence at data/knowledge.sqlite + data/adapters/. On HF Spaces | |
| the path is the container's ephemeral disk, so uploads vanish on rebuild | |
| — there's a visible notice in the UI explaining that. | |
| """ | |
| from .adapter_importer import import_adapter | |
| from .dataset_builder import build_dataset, write_jsonl | |
| from .notebook_generator import generate_training_notebook | |
| from .pipeline import KnowledgePipeline | |
| from .store import KnowledgeStore | |
| __all__ = [ | |
| "KnowledgePipeline", | |
| "KnowledgeStore", | |
| "build_dataset", | |
| "write_jsonl", | |
| "generate_training_notebook", | |
| "import_adapter", | |
| ] | |