Qwen 3.5

1. I have 4x 128 GB VRAM now , what should i do. (www.reddit.com via reddit)

   3h vllmqwen

2. nice_meme (www.reddit.com via reddit)

   7h copilot

   https://preview.redd.it/z66h627yi96h1.png?width=1080&format=png&auto=webp&s=94040bb76c0f8099b58927771c2193dd6a5019da qwen3.5 9b at 0 bit quant>>>>>>>copilot

3. [Opinion/Benchmark] Gemma4-12B's architecture change is too big of a tradeoff; A quick reasoning comparison between Gemma4-12B and Qwen 3.5-9B (www.reddit.com via reddit)

   8h qwenllama

   I took the liberty to test both models today on my favorite benchmark question, head to head. Device: Apple Mac M3 Max 64GB Environment: llama.cpp, all defaults Gemma4-12B's token generation speed: 47 tps with MTP and 2 predicted tokens 29…

4. Nex N2 has a funny "few words do trick" reasoning (www.reddit.com via reddit)

   1d qwen

5. Preferred two LLM combo (www.reddit.com via reddit)

   1d

   I’m using my MacBook Pro M1 Pro with 32GB to run Qwen3.5-35B in Q4 as my coding agent. I have a gaming PC with a 5070 Ti that I’m currently not using but would like to.

6. Dense vs MoE quantization resiliance (www.reddit.com via reddit)

   2d moeqwen

   Which one is more resiliant to quantization? Especially at 4-bit?

7. Running Hermes fully local (www.reddit.com via reddit)

   2d agentic

   Before Hermes was announced, I was working on my own fully local, personal agentic system. Now, I'm a novice when it comes to coding.

8. I can't wait for all the x250 sample distills of Mythos and GPT-5.6 (www.reddit.com via reddit)

   2d mythosgpt-5gemma+1

   Just kidding. Are there any distills that actually improve a model's quality?

9. It felt good to return my Asus Spark (www.reddit.com via reddit)

   3d moeqwen

   It's an incredible little package but too expensive of a price to pay for the performance and I simply didn't want to be part of the great "Superchip lie" - it could be super, but its super ruined by its limited memory bandwidth even thoug…

10. Launch HN: General Instinct (YC P26) – Frontier models on edge devices (news.ycombinator.com)

    +42 4d moe

    Hey HN, Guanming and Bill here from General Instinct (https://general-instinct.com/). After years of working in robotics, we kept running into the same problem: the best models never fit the hardware we actually had available.

11. Show HN: Hitoku Draft – Context aware local assistant (hitoku.me via hn)

    +5 5d gemmaqwen

    Hi guys. I have been working on Hitoku Draft, an open-source, voice-first AI assistant that runs entirely locally.

← all threads