Hug! 🦂🦂 (@hugprat) header
Hug! 🦂🦂 (@hugprat) profile picture

IA local per a empreses a Barcelona: LLMs que no surten del teu servidor. ChatGPT privat, RAG, agents, AI Act. 25 Freelance Local IA with Dgx Spark

Barcelona
Joined June 2010
Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)Photo by Hug! 🦂🦂 (@hugprat)
Hug! 🦂🦂 retweeted
Jensen
" You're gonna be remembered forever because some things are eternal."
6,243
13,609
1,290
152,780
12,579,726
Hug! 🦂🦂 retweeted
Sobre esta pedazo de investigación os diré que, en el apartado tecnológico, la hemos llevado adelante utilizando para el procesamiento de datos modelos locales de inteligencia artificial corriendo en un nodo de inferencia propio. Ha sido un desafío muy estimulante.
TikTok, el hogar del pederasta. Hemos localizado, y visto, más de 1.000 vídeos con contenido explícito de abuso infantil y también hemos comprobado cómo las propias herramientas de la plataforma conducen directamente a este material y lo “promocionan”. La entrega de hoy aquí 👉🏽 bit.ly/4BS9aEM
23
53
1
314
22,116
Hug! 🦂🦂 retweeted
JENSEN! THATS NOT WHAT WE MEANT BY CHINESE MODELS!!!!!!!
593
8,129
701
122,742
2,694,455
Hug! 🦂🦂 retweeted
Degut al servei deficient i la inexistent informació de @rodalies ens trobem 140 castellers atrapats a l'estació de Sants impossibilitant així la nostra arribada, de moment, a l'actuació de Mataró.
56
600
35
1,240
52,013
Hug! 🦂🦂 retweeted
🔴 Cada semana los Servicios de Urgencias de los hospitales de las principales ciudades reciben pacientes sin ningún vínculo con España que han sido diagnosticados de una enfermedad grave en sus país y que directamente desde el aeropuerto cogen un taxi o solicitan una ambulancia. (1/10)
704
8,212
1,104
14,668
2,290,327
Hug! 🦂🦂 retweeted
Why so accurate? 😂
134
1,227
64
14,432
275,605
Hug! 🦂🦂 retweeted
How To Trap Sam Altman
603
10,070
424
93,266
1,410,357
Hug! 🦂🦂 retweeted
NUNCA HE TENIDO TANTAS GANAS DE VOTAR y no me representa ningún partido.
25
12
118
6,951
Hug! 🦂🦂 retweeted
La posible futura alcaldesa de Berlín reconocía temer por su vida y la de su familia. El origen de su miedo son grupos radicales palestinos afines a su propio partido elpais.com/internacional/202…
175
594
181
1,593
181,361
Hug! 🦂🦂 retweeted
Everyday is zero-day these days
4,088
5,926
1,104
73,278
8,480,849
Ara m'ha vingut al cap una cosa, ¿com haguès jugat ahir el clàssic un jugador com Marc Casadó? (si no tinguès la lessió actual, clar)
53
Quin drama – at Barcelona, Spain
COMUNICADO MÉDICO Las pruebas realizadas han confirmado que Xavi Espart sufre una lesión en la sindesmosis del tobillo derecho como consecuencia de una acción traumática con un jugador rival durante el partido contra el Real Madrid. El tiempo estimado de baja es de aproximadamente siete semanas.
1
51
Vergonyós el Barça avui. Espero la gent es faci una idea del nivell real que tenim i baixem una mica del núvol. – at Barcelona, Spain
41
El dia de la huelga general tendremos que tener mucho cuidado….
19
26
1
127
5,060
Hug! 🦂🦂 retweeted
💻 A partir de demà estarà disponible xn--caadorscaputxes-fmb.cat Una iniciativa particular per recopilar proves dels actes vandàlics d'avui a la nostra ciutat i facilitar-les als Mossos d’Esquadra perquè els responsables responguin penalment. POSEM ORDRE. DEFENSEM BARCELONA❗️
27
222
14
679
44,170
Desktop RTX 4060 Ti 16GB tensor unlock, confirmed on TWO different boards 🎉 MSI (10de:2803) + ASUS ProArt (10de:2805) — same chip, different board variants, both unlocked with the same kernel module. GEMM 8192³, locked vs unlocked @165W: TF32:  23.4 → 34.7 TF/s (+48%) BF16/FP32-acc:  47.8 → 74.7 (+56%) FP16/FP32-acc:  46.4 → 71.2 (+53%) FP8 e4m3:  93 → 126 (+36%) INT8 & FP32 CUDA: unchanged (control ✓) No vBIOS flash, no driver patch. A kernel module runs before the NVIDIA driver binds and opens the FP32-accumulate tensor paths. Fully reversible. @CommonFoundry1 your AD106M payload extended to desktop AD106 (2803/2805) just works — thanks! Also kudos to Zagitta for gpu-unlock. RTX 3090 (GA102) on the same box: BF16 78→129 TF/s, FP16 78→125 @420W.
1
4
629
Hug! 🦂🦂 retweeted
Dear @nvidia, a friendly ask from a guy running 4 of your 3090s and 1 of 3080 Ti. The 30 series is 6 years old. You dont make these cards anymore. Maybe its time to hand them over to the community completely. My post about unlocking the 3090 got 270K views in a day, and the replies are the real story. People asking about the 4090, 3080, 3060 and 2080 Ti. People asking how to do it on Windows. Others listing what the community already patches on these cards: the NVENC session limit, P2P between GPUs. Thousands of people are digging through your old cards with kernel modules, 1 setting at a time, because they love this hardware and want everything it can give. Publish what is limited on consumer Ampere and Turing. Ship an official driver switch for older generations. Or just open it up. I understand the business side. Some people might keep their old cards longer instead of buying new ones. But it would be a huge gesture. You talk about local AI on stage, you ship open models, you sell DGX Spark for home. This would show the care goes all the way down to a used 3090 under someone’s desk. Give them to us. We will do the rest: benchmarks, guides, models tuned for these cards. @NVIDIAAI
NVIDIA sold millions of 3090s with 50% of the tensor speed switched off. Not missing. Switched off. 1 kernel module flips it to full, and my BF16 tensor pipe went from 82 to 155 TFLOPS. Nothing else changed on the cards, and the outputs match the stock run byte for byte. All so the gaming card wouldnt eat into workstation sales. Every 3090 owner ran at 50% for 6 years, so a pro card could look worth the price. Fine, its their business model. And LLMs are the least interesting place for this. Decode waits on memory, and my engine runs its experts in int8. Still +10% on prefill. The models that live on BF16 tensor math are everything else. Qwen Image 2.1 and FLUX 3 Klein, where every diffusion step is big matmuls. MiniMax H3, which takes 4 to 7 minutes on my cards. Music and speech models. And training, every LoRA anyone ever trained on a 3090 went through exactly the path that was cut. Thats what I measure next. This is only the start.
47
98
6
1,004
57,506
PHP2Shell, día 3: PHP reconoce un 3er CVE (2026-115560), otra RCE vía session.upload_progress, incluso con el parche del 21-10. No es bypass: vector nuevo y estrecho. Mitigación: session.upload_progress.enabled=0 en el php.ini del servidor. file.cat/es/alertas/
90