Making LLM Training Faster with Unsloth and NVIDIA
segmenta
114 points
21 comments
May 07, 2026
Related Discussions
Found 5 related stories in 761.8ms across 14,015 title embeddings via pgvector HNSW
- Executing programs inside transformers with exponentially faster inference u1hcw9nx · 17 pts · March 12, 2026 · 57% similar
- Unsloth Studio brainless · 233 pts · March 17, 2026 · 56% similar
- Real-time LLM Inference on Standard GPUs: 3k tokens/s per request NicoConstant · 202 pts · May 29, 2026 · 56% similar
- Show HN: Tiny-vLLM – high performance LLM inference engine in C++ and CUDA yu3zhou4 · 122 pts · May 29, 2026 · 55% similar
- MegaTrain: Full Precision Training of 100B+ Parameter LLMs on a Single GPU chrsw · 280 pts · April 08, 2026 · 54% similar
Discussion Highlights (3 comments)
stared
While I do admire Unsloth (especially their https://huggingface.co/unsloth/Qwen3.6-35B-A3B-GGUF binarizations), the linked blog post looks like written by AI from notes (unless a human author acquired this taste from interactions with chatbots).
electroglyph
nice writeup! looking forward to doing some more training as soon as i get some more data sorted. it'll be a custom arch, but i'll probably shoehorn it into unsloth for a speed boost.
wiradikusuma
Quick question, for average joe do we still need to "train" LLM or we can just use off the shelf model and use it ("inference"?) for normal use cases like business process augmentation (e.g. helping read paper receipts, or generate cat videos)?