feat: NebulaGraph as persistant layer

[wey-gu]

• as BaseGraphStorage, via 502d92f
• as BaseKVStorage
  • WIP via 7ff213a

partially-implement: #2

[wey-gu]

I realize that implementing NebulaGraph as the BaseKVStore is also necessary to fully leverage its graph capabilities.

This is still a working in progress and requires more effort.

[gusye1234]

Really COOL PR! Just a quick reminder:
I noticed NebulaGraphStorage has many dataclass field with default value.
However the Instantiation of NebulaGraphStorage is controlled by nano-graphrag, not the user. So those fields will always remain the default values. I'm not sure that's what you want.

If not, all the parameters of GraphRAG class will be passed into BaseStorage as self.global_config, maybe you can read self.global_config['addon_params'] to make sure user can overwrite some parameters of NebulaGraphStorage

[wey-gu]

Really COOL PR! Just a quick reminder: I noticed NebulaGraphStorage has many dataclass field with default value. However the Instantiation of NebulaGraphStorage is controlled by nano-graphrag, not the user. So those fields will always remain the default values. I'm not sure that's what you want.

If not, all the parameters of GraphRAG class will be passed into BaseStorage as self.global_config, maybe you can read self.global_config['addon_params'] to make sure user can overwrite some parameters of NebulaGraphStorage

Thanks! Got it, some of them indeed are configurable and I'll use addon_params!

[wey-gu]

update aug-21 7ff213a:

• WIP adding knowledge storage with NebulaGraph(some of KVStorage namespaces will be graph native in NebulaGraph), some are not(llm cache)
• rebased the main to resolve conflict

[gusye1234]

update aug-21 7ff213a:

• WIP adding knowledge storage with NebulaGraph(some of KVStorage namespaces will be graph native in NebulaGraph), some are not(llm cache)
• rebased the main to resolve conflict

Hi, wey. Great work and design. StorageFactory is great design for fine-grained control.
But maybe making StorageFactory an optional schema for initialization is a better idea?
I think it kinda high-level design and the main path of nano-graphrag should be simple and easy.

Users should firstly understand the BaseStorage concepts(kv, gragh, vector) and play with it, then the detailed components (full_docs, text chunks...) configures.