Brad
@brad_agi
Followers
1,084
Following
781
Media
80
Statuses
767
Computer Scientist | Dev at @cmdLang (AI Agent Platform)
Joined May 2023
Don't wanna be here?
Send us removal request.
Explore trending content on Musk Viewer
#KawalPutusanMK
• 1848912 Tweets
#TolakPolitikDinasti
• 1659246 Tweets
#TolakPilkadaAkal2an
• 1613505 Tweets
#jjk267
• 329596 Tweets
Nobara
• 221639 Tweets
Botafogo
• 161853 Tweets
Mulyono
• 114180 Tweets
Sian
• 71498 Tweets
BATALKAN BUKAN TUNDA
• 63770 Tweets
BOTO PARA SA SB19
• 45327 Tweets
Bolt
• 37832 Tweets
チャイルドシート
• 31048 Tweets
#BauKetekOligarki
• 29524 Tweets
鯉登少尉
• 28424 Tweets
PilihDAMAI BarengPRABOWO
• 27282 Tweets
ポケットキャンプ
• 25465 Tweets
LebihSEJUK LebihNYAMAN
• 22775 Tweets
BBFA LAST HURRAH
• 21233 Tweets
ポケ森サ終
• 20969 Tweets
KITA SEMUA TURUN
• 20393 Tweets
首都高バトル
• 19476 Tweets
Widjiatno Notomihardjo
• 13983 Tweets
中川大志
• 12293 Tweets
買い切りアプリ
• 11188 Tweets
GCSE
• 10862 Tweets
Getting over 4 tokens per second on a Raspberry Pi 5 with Microsoft's Phi-3 Mini! Great model to run entirely locally! Model link in comments!
5
5
44
Just pushed GraphMemory 0.2.1! 👨💻
Includes DSPy knowledge graph extractor example!
Link in comments! (Please star ⭐️)
3
8
33
Getting a very usable 2.2 tokens a second with Llama3 Q4 on Raspberry Pi 5 (fanless)
2
4
19
I highly recommend people check out GraphMemory in light of MSR GraphRAG release!
1
2
21
Just finished cooking a new GraphRAG database! 👨🍳 🔥🍳 - introducing GraphMemory!
Great for making hybrid knowledge / vector graphs! Each node holds a vector embedding. Should scale well and does not require a server to run!
2
7
18
Updated lexical graph example to include edge on each node (text chunk) connecting them together (link in comments)
4
1
17
LLaMA3 70b + Groq + Cursor = 🔥🔥🔥
🏃♂️💨🏃♂️💨🏃♂️💨🏃♂️💨🏃♂️💨🏃♂️💨
How to run Groq on Cursor for WARP SPEED responses:
5
3
17
Hey guys I’m going to be publishing a new opensource GraphRAG database later today! Would appreciate your support!
🔥Attention🔥 Will be dropping an opensource GraphRAG database later today for use with LLMs.
Integrated lessons learned from previous experience including dual semantic representations (knowledge graph / vector hybrid). Would appreciate all support.
Stay tuned & keep cooking.🧑🍳
2
2
14
1
3
17
Very interesting article on AI improving itself!
0
3
17
Blown away by this. Very impressive that a LLM model can be so small! Lots of room at the bottom :)
hacked bitnet for finetuning, ended up with a 74mb file. It talks fine at 198 tokens per second on just 1 cpu core. Basically witchcraft.
opensourcing later via
@skunkworks_ai
base here:
60
220
2K
1
0
16
Running Llama 3.1 8b on a Raspberry Pi 5!
Seems good at extracting nodes and edges for a knowledge graph for use with GraphRAG!
4
2
16
Experimenting with a visualizer for GraphMemory! Should be good for visualizing knowledge graphs!
1
0
16
‼️🚨 Imminent Drop alert 🚨‼️
Will be pushing a Cypher query language parser and DSPy knowledge graph extractor soon to GraphMemory…
Stay tuned and keep cooking! 🧑🍳 🔥 🍳🥘
3
1
13
We going to dissect the GraphRAG drop and implement all lessons learned into GraphMemory.
1
0
14
🔥Just pushed GraphMemory 0.2! 🔥
Updates include:
- Parser to allow Cypher queries to database
- Updated OpenAI example to build knowledge graphs
- Updated schema for compatibility
Would appreciate stars on the Github Repo!
Pip command in comments!
5
1
14
🔥Attention🔥 Will be dropping an opensource GraphRAG database later today for use with LLMs.
Integrated lessons learned from previous experience including dual semantic representations (knowledge graph / vector hybrid). Would appreciate all support.
Stay tuned & keep cooking.🧑🍳
2
2
14
Going to add sentence embeddings natively to so vectors can automatically be created…
2
0
14
Llama 3.1 405B is now on:
Fireworks
Nvidia
Groq
AWS Bedrock
OpenRouter
2
2
13
So pretty apparent that agents will eat apps over time and app stores may go into permanent decline with advent of generative UIs paired with agents.
1
0
13
Cool example of caching / getting a result from a LLM model before finishing typing
2
0
13
Just wanted to send a huge thank you to
@ryancarson
and the rest of the team at
@IntelAI
for this amazing Arc A770 GPU.
I won it in a giveaway during one of their Spaces!
🙏 🙏 🙏
Will be trying some cool stuff with it using / training LLM models in the near future!
3
2
13
Fixed a memory issue while building Flowise…
0
4
12
Sometimes all it takes is the magic word.
(Never store secrets in a LLM system / pre-prompt)
2
2
12
Great resource for fine tuning LLM models!
2
1
11
Just learned Anyscale already has function calling on
Mistral-7B-Instruct-v0.1
1
0
10
Testing Nisten's quantization of Phi-3 medium on a Raspberry Pi-5!
13B phi-medium-4k GGUF files here, model is looking very very good.
just search nisten in
@LMStudioAI
and you should be able to find them there too
11
21
137
2
2
10
GraphRag is the future. We going to body traditional vector databases. 😆
2
0
10
Wrapping up some loose ends with the open source GraphRAG database I will be dropping soon.
This is the problem I’m solving:
0
1
9
Added chunking example where every node is a chunk of text to GraphMemory. See comments for link.
Feel free to⭐️repo! 😉
3
1
9
My friend, Alignment Lab, just released a very interesting dataset!
ANNOUNCEMENT TIME
UpvoteWeb-24-600M ! build out by Grass Foundation! a majorly useful research artifact!
UpvoteWeb-24-600M contains:
600 million of the top posts and comments from Reddit in 2024
media links!
reply lineage!
The data has been totally anonymized to preserve user
11
34
118
1
0
9
Very cool project by
@alxfazio
replicating Perplexity’s Pages! I tested it and it is very good for researching topics!
Introducing Openplexity Pages, with agents!🕵🏻
My new repo that allows Llama 3.1 405B to create, organize, and share information.
Think of it as an open-source version of Perplexity Pages, powered by
@crewAIInc
agents and
@GroqInc
's blazingly fast inference.
6
51
260
0
0
8
This looks awesome! Congrats Alex!
Tomorrow! Thrilled to finally release the open-source version of Perplexity Pages with
@crewAIInc
, Llama 3.1 405B, and
@GroqInc
.
Big thanks to
@GroqInc
for their blazing-fast inference that makes this all possible.
8
25
79
0
1
8
Cool that people are starting to recognize the graph databases are the next step up from vector databases!
0
0
8
Nice family tree I found on the hierarchy of AI.
Would be very interested to see this fully built out with LLM models / their genealogy. Tracking the history of LLM merges could be very interesting / useful in the future...
1
0
6
Lots of people have heard of Groq but SambaNova is also providing very fast inference on ASIC hardware!
WOW, SambaNova is giving over 123 tokens/second on llama 3.1 405B with their ASIC hardware 🤯
Llama 3.1 8B at over 1000 tokens / second 🤯
Also, you use it without any signup
14
66
396
0
1
7
Still an open source win. People now realize they need a backup LLM they control as a bare min
0
0
7
Just came across this fantastic example of visualizing agent/human participation in complex tasks. Graphs like these are incredibly powerful yet still underutilized! 📊
Please show
@duttonphi
some love and ⭐️his repo!
@Teknium1
The concept of visualizing task descriptions is a kind of utility case (as seen in my repo which needs some stars ⭐️ 😂
I assume this technique could play a part of more reasoning benchmarks. YMMV
)
1
0
2
1
0
7
@ryancarson
Great way to test out code! I got it to output gradient descent algorithm in bubble time!
2
0
7
Data and LLM models need to be split. It doesn't make sense to solely rely on a model if it hallucinates. It's oftentimes difficult to tell when a model is hallucinating, therefore data from a model cannot be trusted at all. RAG on verified data is the only way forward.
4
0
7
LLama 3.1 405B seems very capable! I just tried it and here are the results when using it for entity extraction with knowledge graphs!
4
1
6
Open sourcing software then making a paid cloud version is
an excellent strategy.
9
asking to be ripped off.
3
3
0
6
LLM Prompt injection != jailbreaking
1
0
6
I've decided I will be open sourcing a lot more LLM tools in the coming days.
Will be dropping a LLM oriented database tool that I've been sitting on for a bit either later today or tomorrow.
Stay tuned.
0
1
6
So I’m thinking best way to improve
@cmdLang
is with more tools. More tools the merrier.
#buildinpublic
0
0
6
"Well I think what's happening or what's about to happen it's sort of like we've discovered a new continent with 100 billion people on it and they're willing to work for free."
-Nat Friedman
0
0
6
