Hacker News: danielhanchenhttps://news.ycombinator.com/user?id=danielhanchenHacker News RSShttps://hnrss.org/hnrss v2.1.1Mon, 17 Aug 2026 08:06:38 +0000<![CDATA[New comment by danielhanchen in "Qwen 3.8 27B"]]>Actually we do publish non KLD benchmarks - top-1% is better - for NVFP4 for eg we did MMLU Pro, GPQA, AIME 2025: https://unsloth.ai/docs/models/qwen3.6#nvfp4-benchmarks

Sometimes they're just slow and expensive, so we we KLD as a proxy measure and it's very high correlation (95%+)

]]>Fri, 14 Aug 2026 17:24:56 +0000https://news.ycombinator.com/item?id=49301832danielhanchenhttps://news.ycombinator.com/item?id=49301832https://news.ycombinator.com/item?id=49301832<![CDATA[New comment by danielhanchen in "Qwen 3.8 27B"]]>That wasn't our problem right? Gemma officially updated tool calling which we adopted

]]>Fri, 14 Aug 2026 17:22:46 +0000https://news.ycombinator.com/item?id=49301799danielhanchenhttps://news.ycombinator.com/item?id=49301799https://news.ycombinator.com/item?id=49301799<![CDATA[New comment by danielhanchen in "Qwen 3.8 27B"]]>We also made NVFP4 ones if that helps! https://huggingface.co/unsloth/Qwen3.8-27B-NVFP4

]]>Fri, 14 Aug 2026 16:04:11 +0000https://news.ycombinator.com/item?id=49300638danielhanchenhttps://news.ycombinator.com/item?id=49300638https://news.ycombinator.com/item?id=49300638<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Ok no worries - if there are any future issues - feel free to message / make a HF issue - we'll fix promptly!

Also note its best to follow Gemma4's official sampling params since they evaled with it - dry multiplier sometimes works, but it actually screws up reasoning sometimes

]]>Fri, 14 Aug 2026 06:12:01 +0000https://news.ycombinator.com/item?id=49295255danielhanchenhttps://news.ycombinator.com/item?id=49295255https://news.ycombinator.com/item?id=49295255<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>We will investigate Ling!

]]>Thu, 13 Aug 2026 07:58:06 +0000https://news.ycombinator.com/item?id=49282955danielhanchenhttps://news.ycombinator.com/item?id=49282955https://news.ycombinator.com/item?id=49282955<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Thanks for the support and to the community!

]]>Thu, 13 Aug 2026 02:54:03 +0000https://news.ycombinator.com/item?id=49281282danielhanchenhttps://news.ycombinator.com/item?id=49281282https://news.ycombinator.com/item?id=49281282<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Hey :)

]]>Wed, 12 Aug 2026 22:39:45 +0000https://news.ycombinator.com/item?id=49279561danielhanchenhttps://news.ycombinator.com/item?id=49279561https://news.ycombinator.com/item?id=49279561<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Thanks haha

]]>Wed, 12 Aug 2026 22:39:29 +0000https://news.ycombinator.com/item?id=49279556danielhanchenhttps://news.ycombinator.com/item?id=49279556https://news.ycombinator.com/item?id=49279556<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Hey yes - if you could describe what the issues are - we will gladly fix them!

]]>Wed, 12 Aug 2026 22:39:19 +0000https://news.ycombinator.com/item?id=49279552danielhanchenhttps://news.ycombinator.com/item?id=49279552https://news.ycombinator.com/item?id=49279552<![CDATA[New comment by danielhanchen in "Qwen3.8-2.4T"]]>Hey sorry what are the problems that you're experiencing - we're more than happy to help fix them!

]]>Wed, 12 Aug 2026 22:38:38 +0000https://news.ycombinator.com/item?id=49279544danielhanchenhttps://news.ycombinator.com/item?id=49279544https://news.ycombinator.com/item?id=49279544<![CDATA[New comment by danielhanchen in "Unsloth Desktop: Open-Source App for Local Models"]]>Oh thanks for sharing! We launched a Desktop app with fast diffusion, video gen, training support, inference + tool calling, web search, canvas, HTTPS secure remote access via Cloudflared, API / model swapping + more!

It works in [Windows, Linux, Mac, WSL] X [NVIDIA GPUs, AMD GPUs, CPUs]!

If there are any issues / suggestions - we'll be quick to fix / implement them!

]]>Tue, 11 Aug 2026 18:39:39 +0000https://news.ycombinator.com/item?id=49262640danielhanchenhttps://news.ycombinator.com/item?id=49262640https://news.ycombinator.com/item?id=49262640<![CDATA[New comment by danielhanchen in "Inkling: Our Open-Weights Model"]]>Oh thanks for sharing! The llama.cpp PRs should generally be fine for now - I'm fixing a few small edge cases as well!

]]>Thu, 16 Jul 2026 06:47:46 +0000https://news.ycombinator.com/item?id=48931117danielhanchenhttps://news.ycombinator.com/item?id=48931117https://news.ycombinator.com/item?id=48931117<![CDATA[New comment by danielhanchen in "Good results fine tuning a local LLM like Qwen 3:0.6B to categorize questions"]]>Very cool write-up and GitHub repo!

]]>Mon, 22 Jun 2026 07:36:06 +0000https://news.ycombinator.com/item?id=48626978danielhanchenhttps://news.ycombinator.com/item?id=48626978https://news.ycombinator.com/item?id=48626978<![CDATA[New comment by danielhanchen in "Unsloth Joins PyTorch Ecosystem"]]>Thank you appreciate the support! It's all thanks to you guys and the community!

]]>Mon, 11 May 2026 15:15:35 +0000https://news.ycombinator.com/item?id=48096146danielhanchenhttps://news.ycombinator.com/item?id=48096146https://news.ycombinator.com/item?id=48096146<![CDATA[New comment by danielhanchen in "Making LLM Training Faster with Unsloth and NVIDIA"]]>Update - Just got rid of the spiced up intro

]]>Thu, 07 May 2026 10:24:14 +0000https://news.ycombinator.com/item?id=48047664danielhanchenhttps://news.ycombinator.com/item?id=48047664https://news.ycombinator.com/item?id=48047664<![CDATA[New comment by danielhanchen in "Making LLM Training Faster with Unsloth and NVIDIA"]]>Thank you!

]]>Thu, 07 May 2026 10:22:58 +0000https://news.ycombinator.com/item?id=48047659danielhanchenhttps://news.ycombinator.com/item?id=48047659https://news.ycombinator.com/item?id=48047659<![CDATA[New comment by danielhanchen in "Making LLM Training Faster with Unsloth and NVIDIA"]]>Oh thanks :) We're also going to add MTP support soon for Qwen3.6!

95% of it is fully human done - the maths, algos, code snippets, screenshots & benchmarks are done / conducted by us and NVIDIA :)

We did use AI to fix spelling errors + made some nice plots using Chat (ours would look horrible lol)

Update - Just got rid of the spiced up intro

]]>Thu, 07 May 2026 10:22:32 +0000https://news.ycombinator.com/item?id=48047651danielhanchenhttps://news.ycombinator.com/item?id=48047651https://news.ycombinator.com/item?id=48047651<![CDATA[Mistral Medium 3.5 YaRN bug fix]]>Article URL: https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/discussions/18

Comments URL: https://news.ycombinator.com/item?id=47990697

Points: 1

# Comments: 0

]]>Sat, 02 May 2026 21:23:21 +0000https://huggingface.co/mistralai/Mistral-Medium-3.5-128B/discussions/18danielhanchenhttps://news.ycombinator.com/item?id=47990697https://news.ycombinator.com/item?id=47990697<![CDATA[New comment by danielhanchen in "Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model"]]>Sorry on the delay - so it installs https://github.com/Blaizzy/mlx-vlm and other components and sets up the commands - you don't need to use it but we thought it might be easier for folks

]]>Thu, 30 Apr 2026 06:13:50 +0000https://news.ycombinator.com/item?id=47958823danielhanchenhttps://news.ycombinator.com/item?id=47958823https://news.ycombinator.com/item?id=47958823<![CDATA[New comment by danielhanchen in "Qwen3.6-27B: Flagship-Level Coding in a 27B Dense Model"]]>Sorry on the delay - oh haha that would be cool :) We did release 2bit dynamic ones, but unsure if they'll be helpful

]]>Thu, 30 Apr 2026 06:13:08 +0000https://news.ycombinator.com/item?id=47958818danielhanchenhttps://news.ycombinator.com/item?id=47958818https://news.ycombinator.com/item?id=47958818