Upcoming ELNA SNS Decentralization Sale

[icpp]

@branbuilder ,
I also want to crosslink this message from @jeshli , with a great proposal to introduce optionally charging for query calls in exchange for a big increase in the instructions limit.

I believe this would make it possible for you to run the LLM agents of ELNA on chain.

If it would indeed unblock you, please voice that with DFINITY to put more weight behind this request. The ELNA project is the most visible AI project on the IC, and your support would mean a lot.

Proposal: Composite Queries

@ulan Thank you for clarifying the instruction limits for composite queries on the Internet Computer. I am currently developing an open-source repository for AI inference, aiming to facilitate the deployment of PyTorch/Tensorflow AI models on the IC. The primary challenge is the instruction limit for single queries: models that are within the RAM capacity of a canister often exceed the available instructions. My initial strategy involved splitting the AI model across multiple canisters and link…

EDIT: this proposal would unblock the instructions limit. Still probably blocked by the memory and lack of GPU limits, but perhaps certain, smaller LLM agents could already run with just this instructions limit removed.

[wpb]

@icpp

Still probably blocked by the memory and lack of GPU limits

Anyone who is blocked by GPU limits should reply to this new thread. Community feedback is requested for Gen 3 node specs on the topic of GPU.

Community Consideration: GPU Use Cases and Specifications

TLDR: The focus is on community feedback for GPU types and usage in the upcoming Gen 3 node specifications for Dfinity. Current nodes lack GPU support, and the discussion centers on whether to include GPUs in the Gen 3 specifications or in specific sub-nets, seeking community insights on GPU utilization. Status Quo: Current gen 2 nodes in Dfinity’s network boast 500 GB of RAM and substantial processing power, adept at parallelizing processes and running multiple single-threaded WASM canister…

[domwoe]

Hey @branbuilder,

Congratulations on submitting the proposal NNS Dapp

Based on the proposal, the following canisters will be controlled by the SNS:

 #User manegement canister
    - ev7jo-jaaaa-aaaah-adthq-cai
    #RAG canister
    - n4ahg-pqaaa-aaaah-advla-cai
    #frontend canister
    - gpdbs-xqaaa-aaaah-adtiq-cai
    #agent canister
    - gichg-2iaaa-aaaah-adtia-cai

However, I can’t find the implementation of the RAG canister, and I can’t map it very well to the architecture described here: https://docs.elna.ai/elna-whitepaper/technology-architecture. Could you please clarify this further?

It’s also not clear to if and how the Vector DB canister is part of the system.

[branbuilder]

@domwoe

#RAG canister
    - n4ahg-pqaaa-aaaah-advla-cai

Sorry for the confusion. There is some miss naming in the git, description and diagram.

RAG Canister: Key to enhancing LLM interaction, this canister orchestrates the prompt engine, linking it with the vector database for efficient data retrieval and processing. It plays a crucial role in the generation and handling of prompts for LLMs.

In git you could find in the name “elnaAi”

In the diagram you can see it as RAG canister [Prompt Engine and Agent Orchestration]

will correct the naming & description.

We plan to have the VectorDB [GitHub - elna-ai/elna-vector-db] implementation done on the user side, where user have the control .

[bjoernek]

Question on the token distribution:

The proposal text states

Proposed Token Distribution Total token supply - 250 Million Reserved for SNS treasury - 50.70% To be distributed by the NNS-run decentralisation swap which includes the Community Fund - 25.00% Allocated to the ELNA founding team which vest over 36 months - 15% Allocated to seed investors which vest over 36 months - 3% Allocated to LBP participants & other airdrops which vest over 12 months - 5.3%

Adding these numbers up gives me 99%. In contrast the whitepaper, includes an additional 1% for airdrops.

Furthermore, I see from the SNS init file that the 5.3% for LBP is included in the treasury component. Could you please explain why (and when is the distribution of these tokens planned) ?

[bjoernek]

Further question, as I am trying to understand the technical architecture and the roadmap:

Which precise components from the technical architecture are
a) not yet implemented or not yet accessible
b) already implemented (on chain)
c) already implemented (off chain)

Would it be fair to summarize that so far access control and the agent metadata is on-chain and implemented (so belonging to category b) ) ?

[branbuilder]

Hi @bjoernek,

Additional 1% of the total token supply indeed is reserved for future airdrops. This allocation is designed to enhance community engagement and participation in the Elna ecosystem. The slight oversight in the aggregation of percentages in the proposal text was an error on our part, and we appreciate your understanding in this matter.

As for the 5.3% allocation for Liquidity Bootstrapping Pool (LBP) participants and other airdrops being included in the SNS treasury component, this decision was made with strategic intent. Including these tokens in the treasury allows us to manage their distribution more effectively and transparently through treasury proposals post-Service Nervous System (SNS) implementation. The distribution of these tokens is planned to occur in a phased manner, following the successful deployment of the SNS and the establishment of a governance framework that facilitates community-driven decisions

[branbuilder]

With respect to our white-paper > technology architecture

Technology Architecture

Major components on chain are

• Vector DB and Similarity Search
• Agent Framework > management of the agent and RAG
• User Interface
• User Management
  *Agent Marketplace

Off the chain we have the model inferencing and LLM.
We are working on the on chain inferencing and larger LLMs [waiting for GPU canisters and larger instruction limits]

[bjoernek]

Thank you for the quick feedback @branbuilder !

Major components on chain are

• Vector DB and Similarity Search
• Agent Framework > management of the agent and RAG
• User Interface
• User Management
• Agent Marketplace

Could you please comment which of these are already accessible on-chain now ?

[branbuilder]

Currently

• Vector DB and Similarity Search
  -done development, available in beta , code available in git

• Agent Framework > management of the agent and RAG
  –done development, available in beta , code available in git

• User Interface
  –in production

• User Management
  –in production

• Agent Marketplace
  –on development