Marcos V

11.9K posts

Marcos V banner
Marcos V

Marcos V

@TheRealMarcosV

professional beta tester

Katılım Ağustos 2010
353 Takip Edilen299 Takipçiler
Sabitlenmiş Tweet
Marcos V
Marcos V@TheRealMarcosV·
Couple days with the Solana seeker it's been awesome!! @solanamobile
Marcos V tweet media
English
51
17
274
26.4K
Marcos V
Marcos V@TheRealMarcosV·
Got Gemma 4 working with full context on multi-GPU.
GIF
English
0
0
0
9
Boris Cherny
Boris Cherny@bcherny·
Starting tomorrow at 12pm PT, Claude subscriptions will no longer cover usage on third-party tools like OpenClaw. You can still use these tools with your Claude login via extra usage bundles (now available at a discount), or with a Claude API key.
English
830
336
4.2K
1.2M
Marcos V
Marcos V@TheRealMarcosV·
Anthropic throwing me a bone for complaining on X
Marcos V tweet media
English
0
0
0
1
Marcos V
Marcos V@TheRealMarcosV·
@lmstudio @GoogleDeepMind having trouble loading bigger context on any of these models -multi gpu rig 32gb vram any ideas? runtime has been updated btw.
English
0
0
0
42
Marcos V
Marcos V@TheRealMarcosV·
Same
Jeddi@antinertia

i’ve been a gamer since i was 12 every year i kind of hibernate for a few weeks and play 24/7 whether it’s world of warcraft or any new mmorpg (i’ve tried them all) i’m really into mmorpgs since january, i’ve been claude code pilled and i get the same kind of joy from it as i do from playing video games i can’t really explain why but it feels exactly like a video game i can be whoever i want to be interact with the world in any way i want and no one is there to tell me: you can’t do that, you can’t be that

English
0
0
0
27
dax
dax@thdxr·
what if we gave you unlimited tokens for free and we also paid you
English
707
33
3.6K
239.6K
Ettore Di Giacinto
Ettore Di Giacinto@mudler_it·
APEX + TurboQuant from @no_stp_on_snek are landing in @LocalAI_API - Q8_0 quality at half the size and significantly faster inference. To benchmark bigger models we need GPU access. If you can help, DM me 🙏 What model do you want quantized with APEX next? 👇
English
11
4
33
2.8K
Marcos V
Marcos V@TheRealMarcosV·
i cannot keep up, everyday its a new model.
Ettore Di Giacinto@mudler_it

I've just released APEX (Adaptive Precision for EXpert Models): a novel MoE quantization technique that outperforms @UnslothAI Dynamic 2.0 on accuracy while being 2x smaller for MoE architectures. Benchmarked on Qwen3.5-35B-A3B, but the method applies to any MoE model. Half the size of Q8. Perplexity comparable to F16. Works with stock @ggml_org's llama.cpp. Open source (of course!), with ❤️ from the @LocalAI_API team. 👇Links to the model, repository and benchmarks below! (+ Bonus TurboQuant benchmarks with @no_stp_on_snek's TQ+! )

English
0
0
1
30
Eric ⚡️ Building...
Eric ⚡️ Building...@outsource_·
bro stole my tweet and the image I used 😂
Craig Hewitt@TheCraigHewitt

Very bullish on open source and local models Imagine running near-Opus-level model locally on that $600, 16GB Mac Mini you bought last month This 27B Qwen3.5 distill was trained on Claude 4.6 Opus reasoning traces and is putting up real numbers: - beats Claude Sonnet 4.5 on SWE-bench - keeps 96.91% HumanEval - cuts CoT (chain of thought) bloat by 24% - runs in 4-bit quantization Why this matters: local agent loops get a lot cheaper, faster, and more usable. frontier models aren’t going to keep subsidizing cheap tokens on subscriptions forever 300K+ downloads already on HF Link below 👇🏻 We’re early

English
3
0
4
748
Marcos V
Marcos V@TheRealMarcosV·
@ashen_one Codex loves to hand hold and make you waste 5-6 prompts before getting what you want done opus just goes yolo
English
1
0
1
18
ashen
ashen@ashen_one·
I think Codex is amazing but in my experience, for what I use it for, Claude has always been the most effective I use it for vibe coding and Open Claw more than anything. A lot of the stuff that I require my Open Claw to do is like free range stuff that he uses a lot of API calls for and browser access for During the week that I tested Codex on one of my OpenClaws, I ran into a lot of problems where he would just tell me that he's not allowed to or doesn't know how to do certain things That level of autonomy is something that I've never dealt with with Claude For Claude, he's less of a fed, where if he doesn't know how to do something or even if it's a little bit of a gray area, he'll still do it For Codex on OpenClaw he would always just refuse to do certain things that he was hard coded to not do Switching models became kinda annoying too like no need to waste brain power on these different models, i think it’s best to just find one you love for your specific use cases and focus on using it rather than trying to find the “perfect model” I also think that since Claude Cowork is basically openclaw, it makes sense to just claudemaxx and not deal with switching models
Hagrid@rubieshagrid

@ashen_one as a professional Claude maxer do u literally use Claude for everything and not codex at all or do u think codex is better at implementing Claude’s plans or vice versa? Im lost rn and Claude is being so ass

English
10
0
24
2.1K
Marcos V
Marcos V@TheRealMarcosV·
You get no context on 16gb
Craig Hewitt@TheCraigHewitt

Very bullish on open source and local models Imagine running near-Opus-level model locally on that $600, 16GB Mac Mini you bought last month This 27B Qwen3.5 distill was trained on Claude 4.6 Opus reasoning traces and is putting up real numbers: - beats Claude Sonnet 4.5 on SWE-bench - keeps 96.91% HumanEval - cuts CoT (chain of thought) bloat by 24% - runs in 4-bit quantization Why this matters: local agent loops get a lot cheaper, faster, and more usable. frontier models aren’t going to keep subsidizing cheap tokens on subscriptions forever 300K+ downloads already on HF Link below 👇🏻 We’re early

English
0
0
0
46
Marcos V
Marcos V@TheRealMarcosV·
@manoj_ahi It's alright way more usage but I think opus really is better than chat gpt
English
0
0
0
476
Manoj Ahirwar
Manoj Ahirwar@manoj_ahi·
Ok enough!! time to cancel my claude subscription is codex any good?
Manoj Ahirwar tweet media
English
218
22
1.2K
132K
mr-r0b0t
mr-r0b0t@mr_r0b0t·
@TheRealMarcosV V2 is available in his collections, V3 is dropping now, he’s got 9B done, probably finishing up the others as we speak 😁
English
1
0
0
19
Marcos V
Marcos V@TheRealMarcosV·
Testing on my 32gb multi GPU rig right now.
Eric ⚡️ Building...@outsource_

🚀 Imagine running Claude 4.6 Opus-level reasoning... but entirely on your own GPU with just 16GB VRAM. This 27B Qwen3.5 variant, distilled on Claude 4.6 Opus reasoning traces, delivers frontier coding power locally. It’s beating Claude Sonnet 4.5 on SWE-bench in 4-bit quantization (Q4_K_M) while slashing chain-of-thought bloat by 24%. ✅ Retains 96.91% HumanEval accuracy ✅ Perfect for agentic coding loops (no API costs or latency) 300K+ downloads on HF Link below 👇🏻

English
2
0
1
68
Marcos V
Marcos V@TheRealMarcosV·
@mr_r0b0t Did it just drop? I downloaded this one last night not sure if it's v2 or not lmao
English
1
0
1
18
Marcos V
Marcos V@TheRealMarcosV·
Tool calling seems to work exceptionally well over qwen3.5 35 a3b which is what I was using before.
English
0
0
0
12

Keşfet

@bcherny @lmstudio @GoogleDeepMind @ZagZino @thdxr @mudler_it @no_stp_on_snek @LocalAI_API