Sync wiki with main through eb90dbdd5
Covers merged PRs #3478, #3483, #3484, #3485:
- #3478 ROCm RDNA4: gfx950/gfx120*/gfx125* fused GEMV+small-M GEMM
with per-arch launch tables in utils/arch.py, ROCm>=10.1 floor,
rocBLAS default on CDNA+RDNA4, AWQ GEMV/QQQ/VPTQ_CUDA/EXL3 explicit
ROCm rejections, ParoQuant Triton + FP8 ROCm support, FP4 not
advertised on ROCm, compute_device_filter honored by replicas and
resolve_pack_device (Installation, Inference-Backends,
Kernel-Backends-Reference, FP8-and-NVFP4, AWQ, QQQ, ParoQuant,
VPTQ, EXL3, Memory-Offload, Troubleshooting)
- #3485 Qwen-Omni transformers 5.22 thinker.model.language_model
layout: dual module_tree variants + thinker_* helpers (Multimodal)
- #3483 bounded remote image fetches: load_image_from_url timeouts,
retries, byte/pixel caps, ImageFetchError, env overrides
(Multimodal, Troubleshooting)
- #3484 MoE early-stop: run_parallel skips empty batches instead of
raising (Multimodal, Memory-Offload)
Rename Checkpoint-Formats page to Packing-Formats; add Complementary configs nav group
- Checkpoint-Formats-and-Compatibility -> Packing-Formats-and-Compatibility
(page file renamed; every inbound link retargeted and retitled).
- Sidebar + Home: move Rotations/Smoothing/Norm Tuning and EoRA out of
Quantization methods into a new Complementary configs group, with deep
links to MSEConfig (mse=) and SmootherConfig in QuantizeConfig-Reference.
These configs attach to multiple lifecycles rather than being methods.
Sync wiki with main through c1976918
Add workflow SVGs and refresh documentation navigation and accuracy
Full documentation site: concepts, all quant methods, config reference, lifecycle, backends, internals