If anyone has a better workflow for creating lots of captions in kdenlive please let me know. I had to duplicate each title to the media library and drag it into the timeline, because if I simply copy/pasted then the text content/styling would be shared across instances

Re-read that

You should. 3.5 MoE was worse than 3.5 dense, so expecting 3.6 MoE to be superior than 3.5 dense is questionable, one could argue that 3.6 dense (not yet released) to be superior than 3.5 dense.

Ok but you made a claim about the new model by stating a fact about the old model. It's easy to see how you appeared to be talking about different things. As for the claim, Qwen do indeed say that their new 3.6 MoE model is on a par with the old 3.5 dense model:

> Despite its efficiency, Qwen3.6-35B-A3B delivers outstanding agentic coding performance, surpassing its predecessor Qwen3.5-35B-A3B by a wide margin and rivaling much larger dense models such as Qwen3.5-27B.

https://qwen.ai/blog?id=qwen3.6-35b-a3b

This says a slightly different thing:

https://x.com/alibaba_qwen/status/2044768734234243427?s=48&t...

If you look, at many benchmarks the old dense model is still ahead but in couple benchmarks the new 35B demolishes the old 27B. "rivaling" so YMMV.

Does the large system prompt work fine for this model? If needed, you could use a lightweight CLI like Pi, which only comes with 4 tools by default

I built a Claude-inspired UI for Ollama/llama.cpp

https://github.com/mudkipdev/chat

People have made toki pona translation models before, not exclusively trained though

It's strange that my iPhone 14 is at regular temperature when using the E2B model. But also it's a lot slower (not sure how to measure the exact tokens per second, ~12 if I had to guess)

This is probably a consequence of the training data being fully lowercase:

You> hello Guppy> hi. did you bring micro pellets.

You> HELLO Guppy> i don't know what it means but it's mine.

Great find! It appears uppercase tokens are completely unknonw to the tokenizer.

But the character still comes through in response :)

If you use the 'run' command, it pulls automatically for you

Under 15 is too slow for conversation personally. I guess 5 tokens per second is nice if you're one of the people who likes letting coding agents run overnight

Can't wait for gemma4-31b-it-claude-opus-4-6-distilled-q4-k-m on huggingface tomorrow

I'd rather see a distill on the 26B model that uses only 3.8B parameters at inference time. Seems like it will be wildly productive to use for locally-hosted stuff

gemma4-31b-it-claude-opus-4-6-distilled-abliterated-heretic-GGUF-q4-k-m