Tested on EVO-X2 (gfx1151) under Windows + ROCm: the official b10666 binary runs it at pp512 159.25 / tg128 23.26, and the earlier crash was my own workaround flag.
Traced across 31 blog posts and 464 Hugging Face repos: Qwen's capybara debuted Oct 11, 2024, hit blog banners in Jan 2025, and its own edit prompts all say 'the bear'.
Tested on M1 Max 64GB: AtomicChat's M64 GGUF keeps the 51B N-gram table in its own shard, so a llama.cpp PR #27742 build leaves it on SSD and runs the 125B MoE at 17.6 tok/s.
Tested on M1 Max 64GB, ComfyUI v0.33.3: from-behind and a back-row drummer now pass, out-of-frame crops get worse, and Anima-Base LoRAs need a 52-block key remap.
Tested on M1 Max 64GB with MLX: LLaMA Pro-style block expansion on Qwen3-0.6B-Base vs LoRA vs full FT, trained on 722 blog posts. Held-out PPL drops 14.7 to 10.5, but Wikipedia-ja rises 13.5 to 17.9, more forgetting than full FT. A 20% Wikipedia mix nearly removes it, and renumbering 28-layer LoRA keys reproduces the Anima-2.9B result in numbers.
16GB M4 Mac mini test of QwenLM's Qwen-MM-Plugins core. Claude Code adds a plugin_ prefix that breaks --allowedTools, codex exec rejects MCP calls by default. PDF, STL and video read with no API key.
Tested on M1 Max 64GB, ComfyUI v0.33.3. Anima-2.9B inserts 12 DiT blocks, so Anima-Base LoRA keys hit the wrong layers with zero warnings, 0/3. Renumbering the keys gives 3/3.
Tested on GMKtec EVO-X2 (Ryzen AI Max+ 395): Q8_0 + MTP beats a 4-bit M1 Max at 22 tok/s, thinking burns 32,712 chars before any HTML, and the NSFW refusal line moves.
Tested on M1 Max 64GB: Qwen3.8-27B hits ~19 tok/s on both MLX and Ollama, but the default reasoning_effort=xhigh blew thinking up to 50,373 chars. Why Ollama dodges it.
Eight ComfyUI experiments on a 4-character Anima (Qwen-DiT) LoRA in plain terms: the conditioning right before the DiT decides who appears, and outfit mix-ups come from a biased DiT LoRA.
Built a fictional idol label site with pi.dev + Qwen 3.7/3.8 and Astro. Generated characters, logos, and event photos via genserver — then shipped wireframes with SVG placeholders. Full failure log inside.
Tested on ComfyUI: Qwen3-4B–14B prompt writing, a 4B encoder bridge, and split conditioning all lost to one hand-written 507-token band prompt. An 896-caption audit shows the real fix.
