Alibaba's Qwen team released Qwen3.6-35B-A3B as open weights. A 40-layer hybrid of Gated DeltaNet, Gated Attention, and MoE hits 73.4 on SWE-bench Verified, 37.0 on MCPMark, and 1397 on QwenWebBench.
WAI0731 (creator of WAI-Illustrious) released WAI-Anima v1, a derivative model based on Anima. In the two months since the February Anima article, derivative models have surged along with a LoRA toolkit and text encoder upgrades. Hands-on comparison of preview3-base and WAI-Anima v1.
Tested 5 approaches including Qwen Image Edit, JS color reduction, and Illustrious i2i + LoRA. Illustrious i2i alone turned out to be the fastest and lightest solution for pixel art conversion.
I tested local Vision LLMs (Gemma 3, Qwen2.5-VL, Llama 3.2 Vision, Gemma 4) to see if they could look at character illustrations and pixel art and generate RPG-style stats in JSON format.
Benchmarking NII's LLM-jp-4-32B-A3B-thinking on EVO-X2 (Ryzen AI Max+ 395) with ROCm. 62.9 t/s vs Qwen3.5-35B-A3B's 44.7 t/s. Covers thinking control issues, KV cache trade-offs, knowledge cutoff, Japanese quality comparisons, code generation tests, and training data composition.
Qwen3.5-35B-A3B is an SSM+Attention hybrid where only 10 of 40 layers use KV cache. Expanding ctx-size from 4096 to 65536 on llama-server added just 800MB VRAM with zero speed loss. Includes q8_0 KV quantization benchmarks and TurboQuant status.
Flash-MoE is a C/Metal inference engine that runs Qwen3.5-397B-A17B on a MacBook Pro M3 Max at 4.36 tokens/s. With expert streaming from SSD and hand-written Metal shaders, it fits the 209GB model into a 48GB memory budget.
The three-stage pipeline of BERT perplexity scan → LLM judgment → escalation packaged as a cross-platform Python tool. The installer automatically downloads llama-server and GGUF models.
Using tori29umai’s LoRA to automatically split facial parts, results from batching 28 images, and a log of running into the limits when attempting finer hair separation
Set up the CLI version of NDLOCR-Lite on Apple Silicon Mac, then tested OCR result correction with Qwen 3.5 and Swallow. Includes experiments with direct image reading and the anchoring effect.
An explanation of why Qwen-Image-Edit's VAE is so heavy, how HunyuanImage 2.1 chose a 32x high-compression VAE instead, and how Kohya's memory-optimization work fits in.
