Matthew Malaker

134 posts

Matthew Malaker

@MMalaker57

Katılım Eylül 2023

29 Takip Edilen9 Takipçiler

Qwen3.6 35BA3B seems to spend an enormous time thinking and makes tons of tokens. My Hermes agent compacted a 128K window (0.85 threshold I think) 4+ times writing what seems to be not too complicated code. A speed up over 27B means nothing if it's spent on more thinking.

English

Matthew Malaker@MMalaker57·1d

The lack of availability of many lower bit quantizations on ROCm really puts a brake on using AMD for local inference beyond GGUFs. I wanted to use vLLM, but the good INT4, AWQ, etc. safetensors don't work with rocm, and flat truncating a higher but quant hurts model quality.

English

Matthew Malaker@MMalaker57·2d

@TheAhmadOsman All I want for Christmas is a data center

English

Ahmad@TheAhmadOsman·2d

I am a simple guy I just want my own data center

English

673

16.7K

Matthew Malaker@MMalaker57·2d

@techdevnotes That doesn't seem too bad to me. What are the competitors like?

English

168

Tech Dev Notes@techdevnotes·2d

xAI has Released Pricing of Speech to Text API

English

115

5.4K

Matthew Malaker@MMalaker57·3d

@michaeljknowles Mayflower Cuban edition in the cards 🤔?

English

397

Michael Knowles@michaeljknowles·3d

Polymarket@Polymarket

BREAKING: U.S. reportedly preparing plans for a possible operation in Cuba, awaiting Trump's orders.

ZXX

114

3.3K

72.5K

Matthew Malaker@MMalaker57·3d

@TheAhmadOsman I feel this just wanting a second radeon r9700

English

Ahmad@TheAhmadOsman·3d

My name is Ahmad and I have a Compute problem

Ahmad@TheAhmadOsman

New Tenstorrent cluster hot from the kitchen > 1TB of VRAM > 3TB DDR5 RAM > 32TB SSD Storage New product, will share more later P.S. Can you find the cat in the picture?

English

205

1.5K

87.8K

Matthew Malaker@MMalaker57·4d

@TheAhmadOsman How much memory does the KV cache take for the 31B? My preliminary tests showed it takes quite a lot. Do you quantize the KV cache and if so, what settings do you use?

English

517

Ahmad@TheAhmadOsman·4d

I changed my mind Gemma 3 31B is better and more capable than Qwen 3.5 27B Requires better prompting but it’s more capable, intelligent, and token efficient

English

800

45.9K

Matthew Malaker@MMalaker57·4d

@tonysimons_ @Teknium I'm using it to code up an agent-enabled silly tavern replacement that has intelligent, dynamic, and cache maxing context/lore handling and concurrent agent-driven background evolution of the story that the user can't see to add depth to the world and exploit vLLM concurrency.

English

127

Tony Simons@tonysimons_·5d

Dear Algorithm, Only show this to the most elite Hermes Agent builders on 𝕏 . If you see this, show me what you’re working on. Perhaps we can collab?

English

211

10.5K

Matthew Malaker@MMalaker57·4d

@jukan05 How many times can these masks be used before they get too damaged? My only experience with contact lithography is with chromium photomasks, and you really can't use those many times, but NIL is a bit different. What are they using this on that makes sense?

English

234

Jukan@jukan05·5d

At Photomask Japan 2026 (PMJ), held April 8-10, DNP announced in its Nano-Imprint Lithography (NIL) presentation that it has successfully manufactured using a template with 10nm circuit line widths. At its FY3/26 IR-Day, DNP stated that its NIL business would begin mass production in FY3/28, and at its FY3/25 IR-Day, it set a sales target of ¥4.0bn for the business by FY3/31. I thought NIL was dead, but it's actually being commercialized?

English

167

31.3K

Matthew Malaker@MMalaker57·11 Nis

@elonmusk On one hand, I want one with a bluish osmium colored finish. On the other, the Satisfactory player in me wants one in Caterium finish. Actually, a copper finished cybertruck would look sick as long as it's waxed/has PPF.

English

Elon Musk@elonmusk·11 Nis

Cybertruck is so awesome. Until you’ve tried it out, you have no idea, because there’s nothing like it. Best product Tesla has ever made to date.

Cybertruck@cybertruck

Cybertruck is extremely comfortable on long trips especially thanks to: – Minimal body roll – Comfort Mode on FSD (gentler acceleration & braking, softer damping, smoother handling) – Air suspension + self-leveling keeping the truck flat even with load

English

8.8K

5.7K

65.7K

26.8M

Matthew Malaker@MMalaker57·10 Nis

@dpoddolphinpro Whichever Artemis mission lands on the moon first (3 I think) needs to bring it with them and plant it. That, or hang it in the moon base that gets built.

English

190

Ryan Caton@dpoddolphinpro·9 Nis

This is so so cool. I was so excited when the payload manifest was announced. The Artemis II crew have just showcased the Apollo 18 flag - THIS VERY FLAG would have been planted on the lunar surface, had it gone ahead. Of course, Apollos 18 & beyond were cancelled - but this flag finally got its rightful trip to the Moon.

English

208

2.8K

105.5K

Matthew Malaker@MMalaker57·9 Nis

@michaeljknowles This meaning what it says, that you must actually live out your faith in action for it to mean anything. If you claim to believe yet don't do anything the faith commands or requires, then what is your faith, really?

English

Matthew Malaker@MMalaker57·9 Nis

@michaeljknowles St. James said "Faith without works is dead." It seems quite clear to me.

English

234

Michael Knowles@michaeljknowles·9 Nis

How many times does the phrase "faith alone" appear in the Bible? And what specifically does the Bible say about "faith alone"?

Frank Turek@DrFrankTurek

IMPORTANT QUESTION: Is salvation by faith alone, or by faith plus works?

English

1.6K

113

1.1M

Matthew Malaker@MMalaker57·9 Nis

@Scobleizer I have an interest in multi-agent from the perspective of exploiting vLLM for higher total token throughput for complex problems. That's if it doesn't destroy prompt caching, I think. I don't have the fastest prompt processing atm.

English

476

Robert Scoble@Scobleizer·9 Nis

twitter.com/i/spaces/1DGle…

ZXX

137

31.8K

Matthew Malaker@MMalaker57·8 Nis

@BasedMikeLee Any

Mike Lee@BasedMikeLee·8 Nis

How many votes cast illegally by noncitizens are too many?

English

5.6K

2.4K

14.5K

213.2K

Matthew Malaker@MMalaker57·6 Nis

@jwsaml I was unhappy with my openclaw install, so I started over. It automatically detected the openclaw install, though, and I hear the migration process is generally pretty smooth. You can always copy your openclaw directory as a backup and try it, though.

English

172

Jesse Samuel@jwsaml·6 Nis

@MMalaker57 Nice, was it difficult to migrate?

English

1.2K

Jesse Samuel@jwsaml·6 Nis

Has anyone fully replaced their OpenClaw with Hermes?

English

304

559

89K

Matthew Malaker@MMalaker57·6 Nis

@loktar00 Running models like this in an agent harness on NAS/small home servers with a decent APU is going to become a very common use case, I think.

English

128

Loktar 🇺🇸@loktar00·6 Nis

The "local AI is a waste of money" takes keep coming while people are running Qwen 3.5 35B on $400 mini PCs.... at some point the math just speaks for itself.

English

502

33.4K

Matthew Malaker@MMalaker57·5 Nis

@SecRubio He is risen indeed, Alleluia!

Nederlands

Secretary Marco Rubio@SecRubio·5 Nis

He is Risen.

English

12.1K

43.4K

233.9K

5.6M

Matthew Malaker@MMalaker57·4 Nis

@elonmusk Will it v be' able to do photonics? I'm asking as a potential future applicant.

English

Elon Musk@elonmusk·4 Nis

My idea of a good time is working with amazing engineers to create incredible technology 🤩 The Tesla chip research fab will have all the machines needed to do logic, memory, packing & masks in one building for a lightning fast development cycle. Heaven 💫

English

10.1K

16K

207K

49.4M

Keşfet

@TheAhmadOsman @techdevnotes @michaeljknowles @tonysimons_ @Teknium @jukan05 @elonmusk @dpoddolphinpro