What's Changed
- fix(fav3): misc fixes relating to FAv3 by @EricLBuehler in #2369
- feat(qwen3.5): implement better video processor by @EricLBuehler in #2368
- feat(server): improve openai server compatability with /v1/responses by @EricLBuehler in #2370
- feat(qwen3.5): add mtp support by @EricLBuehler in #2372
- fix(qwen3.5): better support for hybrid models and auto device map by @EricLBuehler in #2373
- feat(metrics): add engine metrics for scheduler status by @EricLBuehler in #2374
- fix(qwen3.8): fixes around gdn rollback for qwen3.8 by @EricLBuehler in #2377
- feat(mtp): support cuda graphs by @EricLBuehler in #2379
- feat(gdn): in-place pooled recurrent state kernels, fused MLP, f32 gate params by @EricLBuehler in #2382
- feat(cuda): CUDA graph batch buckets by @EricLBuehler in #2384
- feat(metrics): improve engine metrics for ttft/itl and speculative by @EricLBuehler in #2383
- perf(mtp): quantized drafter, device verify, stash-based GDN rollback, size-aware n_predict by @EricLBuehler in #2385
- feat(speculative): DFlash and DFlash 2 block-diffusion draft models by @EricLBuehler in #2386
- perf(dflash): batched drafter forward and adaptive draft depth by @EricLBuehler in #2388
Full Changelog: v0.9.1...v0.9.2