2 years ago
3
14
Posted On: March 25th 2025, 09:08 pm
Tagged Users:
🔥🎶 ¡Este viernes 28 de marzo, CONECTA trae una noche inolvidable! 🎶🔥
🔊 MAIN DJ: @mayromusic
Con su música apoyada por leyendas como Hernán Cattáneo y Nick Warren, llega este viernes a @wabifunclub explorando géneros como Progressive House, Melodic House y Techno nuestro main dj @mayromusic
🎧 OPEN WARM: @disantojuanx_
🕛 Horario: 23:59 - 06:00
📍 Lugar: @wabifunclub
🎟️ Entradas disponibles en el link de nuestro perfil. ¡No te quedes afuera!
#CONECTA #WabiFunClub #MayroMusic #DisantoJuanx #MendozaRave #TechHouse #MelodicVibes
🔊 MAIN DJ: @mayromusic
Con su música apoyada por leyendas como Hernán Cattáneo y Nick Warren, llega este viernes a @wabifunclub explorando géneros como Progressive House, Melodic House y Techno nuestro main dj @mayromusic
🎧 OPEN WARM: @disantojuanx_
🕛 Horario: 23:59 - 06:00
📍 Lugar: @wabifunclub
🎟️ Entradas disponibles en el link de nuestro perfil. ¡No te quedes afuera!
#CONECTA #WabiFunClub #MayroMusic #DisantoJuanx #MendozaRave #TechHouse #MelodicVibes
Comments (14)
More posts
![Follow me then comment "SOUP" and I'll send you the open-source tool that fine-tunes an AI model on a 4 GB laptop GPU.
Your laptop can now teach an AI model using your own data. A developer from Kazakhstan named Alpamys Makazhan fine-tuned an 8-billion-parameter model on a laptop with 4 GB of GPU memory, and he gave the tool away for free. It's called Soup.
Here's how it works:
1. Install it with one line: pipx install "soup-cli[train]"
2. Run soup init and it writes you a config file. Set the base model (Llama, Qwen, Gemma, Mistral) and point it at your data.
3. Put your examples in a JSONL file. Your emails, support tickets, notes, product docs, anything you want the model to learn from.
4. Run soup train. Soup picks the batch size, sets up the LoRA adapter, and streams the model into your GPU one layer at a time so it never needs more than about 3.3 GB.
5. Run soup export to GGUF and load it in Ollama. You now have a model that knows your data, on your laptop, with no cloud bill.
What makes this insane:
✅ 8B model trained on an RTX 3050 laptop at 3.32 GB peak VRAM
✅ About 120 tokens a second on that same laptop
✅ One YAML file and one command, no custom training scripts
✅ Free Colab notebook that proves the 4 GB claim before you install anything
✅ 8,700+ stars, Apache 2.0, free forever
Why this matters:
Fine-tuning used to mean renting a cloud GPU and fighting the setup for days, so almost everyone just used someone else's model. Now the gaming laptop you already own can train one on your own data over a lunch break.
Follow me then comment "SOUP" and I'll send you the link.
#finetuning #LLM #opensource #GitHub #machinelearning Llama Ollama AItools localAI deeplearning NVIDIA GPU AIdeveloper Python artificialintelligence buildinpublic AIstartup techfounder ZeroToFounder TheTechPlate](https://s5.imginn.com/846646217_17918422725522037_9154593072214613774_n.jpg?t51.82787-15/846646217_17918422725522037_9154593072214613774_n.jpg?stp=dst-jpg_e35_p720x720_sh2.08_tt6&_nc_cat=102&ig_cache_key=NDAwNzA0ODg4NTgzMjQwNTExNzE3OTE4NDIyNzIyNTIyMDM3.3-ccb7-5&ccb=7-5&_nc_sid=58cdad&efg=eyJ2ZW5jb2RlX3RhZyI6IkNMSVBTLnhwaWRzLjEwODAuc2RyLnZpZGVvX2RlZmF1bHRfY292ZXJfZnJhbWUuQzMifQ%3D%3D&_nc_ohc=cTsP5DnnRsoQ7kNvwE3fYmO&_nc_oc=AdrWrW3Pbl3AyJbnKG3XqbJM__EHglfXORxxHVfBNZnwm5y5RFnIcpykLWs3ZwK-9b-O9kMUP-ubZWcq04CPLHPJ&_nc_zt=23&_nc_ht=scontent-sjc6-1.cdninstagram.com&_nc_gid=hIE4SF489cWx4RWitqsn-Q&_nc_ss=70689&oh=00_AQP2nukfD1K9s6tBGQMcFAe35uw8Ll2_qBdaZujr96Wtvg&oe=6AD6E768)
