Qwen 3.6

1. Qwen3.6 35B - TXT vs Markdown vs HTML vs HTML+CSS (www.reddit.com)

   +154 12d claude-code

   Theres been talk of late about using HTML rather than markdown in Claude Code. I was curious how this worked with a local model so loaded up Qwen3.6 35B A3B at Q8 and F16 KV cache.

2. VLLM gives 5x speed of llama but quants not available (unsloth/gguf). What to do? (www.reddit.com)

   +939 12d vllmllama

   EDIT - IGNORE. I MADE A MISTAKE.

3. I'm seeing low draft acceptance when using Qwen3.x MTP, what am I doing wrong? (www.reddit.com)

   +314 12d llama

   I'm using llama.cpp, and I've tried Bartowski's and my own quants. When using Qwen3.5-122B or Qwen3.6-27B, I'm seeing really low draft acceptance in chats with interleaved code snippets (chatting with the LLM about programming / a code pro…

4. Qwen3.6-35B-A3B-APEX / 128K ctx on RTX 3060 12GB — 37 t/s gen with 72k ctx filled, PPL 3.25, offloading 17GB model (www.reddit.com)

   +1221 12d llama

   I'm posting this because it may be helpful to squeeze the 12GB VRAM in the 3060. All credit goes to spiritbuun's fork (github.com/spiritbuun/buun-llama-cpp) and mudler's APEX quantizations (huggingface.co/mudler).

5. Krasis update: Qwen3.6-35B-A3B (Q4) at reading speed, 1x 8GB 3070 Mobile laptop (32GB RAM) (www.reddit.com)

   +1813 12d

   Context Krasis is an LLM runtime for running models that don't fit into VRAM. Krasis streams the model through VRAM from system RAM efficiently and handles prefill and decode as separate architectures and optimised usecases.

6. Qwen/Qwen-Image-Bench · Hugging Face (huggingface.co via reddit)

   +6013 12d qwen

   Model Description Q-Judger is a vision-language model fine-tuned specifically for automated evaluation of text-to-image generated images. Given a text prompt and a generated image, the model evaluates the image on fine-grained quality crit…

7. Question: Llama cpp, whats good right now for: MTP, KV cache quant, Long context. (www.reddit.com)

   +819 12d vllmqwenllama

   Used the vllm version of https://github.com/noonghunna/club-3090 It worked fine for myabe 20 40k context, havent tried the new one. Anyone used the new llama.cpp patched one for single 3090?

← all threads