kdb 2024-03-11 - KDB Project Call Meeting Minutes

Date

Monday, March 11, 2024 - 9:30am ET / 2:30pm UK

// Second Monday of every month

Attendees

Fullname	Affiliation	GitHub Username
Ross Duffy	Morgan Stanley	@rdffy
Alvin Shih	Morgan Stanley	@alvin-c-shih
Karl Moll	FINOS	@karlmoll
Aaron Davies	Morgan Stanley	@adavies42
Jo Shinonome	ExeQution Analytics	@jshinonome
Peter Gyorok	KX Systems	@gyorokpeter
Hong Wing Hau	BNP Paribas	@honus85
Conor Twomey	KX Systems	@ctwomeykx
Claire Poland	Data Intellect	@claire-sloan

Meeting notices

FINOS Project leads are responsible for observing the FINOS guidelines for running project meetings. Project maintainers can find additional resources in the FINOS Maintainers Cheatsheet.
All participants in FINOS project meetings are subject to the LF Antitrust Policy, the FINOS Community Code of Conduct and all other FINOS policies.
FINOS meetings involve participation by industry competitors, and it is the intention of FINOS and the Linux Foundation to conduct all of its activities in accordance with applicable antitrust and competition laws. It is therefore extremely important that attendees adhere to meeting agendas, and be aware of, and not participate in, any activities that are prohibited under applicable US state, federal or foreign antitrust and competition laws. Please contact [email protected] with any questions.
FINOS project meetings may be recorded for use solely by the FINOS team for administration purposes. In very limited instances, and with explicit approval, recordings may be made more widely available.

Agenda

(5 min) Convene, roll call, welcome new people.
[ ] Display FINOS Antitrust Policy summary slide
[ ] Review Meeting Notices (see above)
(5 min) Approve previous meeting minutes.
- Approved by acclamation.
(10 min) kdb+ 4.1 serialization
- Still using Protocol 6.
- Last time q added a verb, was put on wire as a lambda.
- Lambdas sent as strings.
(10 min) Horizontal Scaling
- qclone a wrapper around fork(2)
  - shares memory pages with parent
  - only works on a single host
- qcluster at Morgan Stanley is for multi-host distribution
  - A collaboration between kdb-using teams from different business lines.
  - Hooks to: spin up containers, retry on failure, etc.
  - Uses conn.q a bit, but could use it more.
  - Using common module loader facilitates this sort of collaboration.
  - qdepends has a lower-level API and pluggable resolver to allow it to be adapted to various environments.
(10 min) Fabric - LLMs via CLI, Unix pipes, and crowdsourced prompts
- https://github.com/danielmiessler/fabric
- pbpaste on MacOS to paste from clipboard as input to fabric
- Could use a .z.pi handler to interact with fabric CLI.
- awq and jqq could be used to process the output and chain requests
  - https://github.com/adavies42/qist/tree/master/lib
- Can run LLM locally using llama.cpp
  - https://github.com/ggerganov/llama.cpp/
  - server provides OpenAI-compatible REST end point and simple ChatUI
    - https://github.com/ggerganov/llama.cpp/tree/master/examples/server
  - Slightly-older releases include compiled binaries:
    - https://github.com/ggerganov/llama.cpp/tree/master/examples/server
  - Can also look into llamafile:
    - https://hacks.mozilla.org/2023/11/introducing-llamafile/
(10 min) Concise / accurate description of q for an LLM?
- Context windows of 64k tokens (~ 256 KB) are becoming mainstream.
- What would you put in there to explain q while leaving enough space for a code completion task?
- Don Orth's abridged manuals
  - https://code.kx.com/q/learn/archive/
- HTML docs
  - https://github.com/KxSystems/kdb/tree/master/d
(5 min) AOB, Q&A, Adjourn.
- FINOS AI Readiness SIG in the works
  - https://github.com/finos/ai-readiness
- KDB.AI team interested in hearing about scale of current RAG projects, particularly ones for alpha generation.