# Changelog

## Score

- CAI 53 → 54 (+0.6)
- Rubric changed (rubric-2026.10.5 → rubric-2026.11.4) — scores are not directly comparable.

## Lenses

- Code Health 53 → 55 (+2.5)
- Architecture 96 → 95 (-0.3)
- Maturity 66 → 66 (+0.2)
- Readiness 62 → 53 (-9.6)
- Security 44 → 44 (+0.2)
- Domain Modelling 100 (new)
- Event Sourcing 100 → 100 (+0.0)
- Accessibility 60 → 83 (+23.5)
- Performance 100 → 100 (+0.0)

## Resolved (933)

- BlockwiseFP8Linear::try_forward_fused_split_glu (cognitive 20) (mistralrs-quant/src/blockwise_fp8/mod.rs)
- BlockwiseFP8Linear::try_forward_fused_split_glu (cyclomatic 16) (mistralrs-quant/src/blockwise_fp8/mod.rs)
- Change coupling: diffusion.rs ↔ multimodal.rs (mistralrs-core/src/pipeline/diffusion.rs)
- Change coupling: diffusion.rs ↔ normal.rs (mistralrs-core/src/pipeline/diffusion.rs)
- ChatCompletionStreamer::__next__ (cyclomatic 20) (mistralrs-pyo3/src/stream.rs)
- ChatStore.editMessage (cognitive 31) (mistralrs-cli/webui/src/lib/stores/chat.svelte.ts)
- ChatStore.editMessage (cyclomatic 19) (mistralrs-cli/webui/src/lib/stores/chat.svelte.ts)
- ClassTooLong: FastExpertsWeights (mistralrs-core/src/moe/experts/backends.rs)
- ClassTooLong: Qwen3VLImageProcessor (mistralrs-core/src/vision_models/qwen3_vl/inputs_processor.rs)
- Duplicated block (10 lines × 12) (mistralrs-core/src/models/deepseek2.rs)
- Duplicated block (10 lines × 2) (mistralrs-code-exec/src/lib.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/engine/add_request.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/kv_cache/hybrid_cache.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/models/gpt_oss.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/models/granite.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/ops.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/ops.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/ops.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/paged_attention/cache_engine.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/pipeline/loaders/multimodal_loaders.rs)
- …and 913 more

## New (1499)

- AutoDeviceMapQuantization::pack_factor_for_candidates (cognitive 18) (mistralrs-core/src/pipeline/loaders/mod.rs)
- BlockwiseFP8Linear::forward_packed_glu (cognitive 20) (mistralrs-quant/src/blockwise_fp8/mod.rs)
- BlockwiseFP8Linear::forward_packed_glu (cyclomatic 16) (mistralrs-quant/src/blockwise_fp8/mod.rs)
- BlockwiseFP8Linear::prepare_provider (cyclomatic 17) (mistralrs-quant/src/blockwise_fp8/mod.rs)
- Change-coupling hub: diffusion.rs → gguf.rs, multimodal.rs, normal.rs (mistralrs-core/src/pipeline/diffusion.rs)
- ClassTooLong: AfqLayer (mistralrs-quant/src/afq/mod.rs)
- ClassTooLong: GridImageProcessor (mistralrs-core/src/vision_models/grid_processor.rs)
- ClassTooLong: Lfm3OmniModel (mistralrs-core/src/vision_models/lfm3_omni/mod.rs)
- ClassTooLong: ManagedEmbedding (mistralrs-quant/src/embedding/storage.rs)
- ClassTooLong: Model (mistralrs-core/src/models/lfm3.rs)
- CpuInputs::run (cognitive 21) (mistralrs-quant/src/lfm3_conv.rs)
- DecisionPlan::read_span (cognitive 19) (mistralrs-core/src/vision_models/lfm3_d1/readout.rs)
- DecisionPrompt::plan (cognitive 23) (mistralrs-core/src/vision_models/lfm3_d1/prompt.rs)
- Duplicated block (10 lines × 15) (mistralrs-core/src/pipeline/loaders/lfm3.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/distributed.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/engine/add_request.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/layers.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/models/gpt_oss.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/models/granite.rs)
- Duplicated block (10 lines × 2) (mistralrs-core/src/models/lfm2.rs)
- …and 1479 more

## Changes since last survey

- 22 commits — 14 feature/other, 8 fixes

## By area

- mistralrs-core/src — 14 commits
- (root) — 2 commits
- docs/src — 2 commits
- mistralrs-cli/webui — 2 commits
- mistralrs-quant/src — 2 commits

## Notable commits

- fix: fix(core): detect the CUDA UMD banner emitted by driver 13.x (#2458)
- fix: fix(core): keep a closing tag found inside a hunyuan tool-call string argument (#2518)
- fix: fix(core): preserve prompt phase in the default scheduler fast path (#2469)
- fix: fix(deps): resolve some Dependabot alerts (#2533)
- fix: fix(lfm3): fix tool call special token parsing (#2531)
- fix: fix(quant): make CPU GGUF matmul input contiguous (#2516)
- fix: fix(quant): reject UQFF GGUF weights whose byte length disagrees with weight.shape (#2495)
- fix: fix(qwen3.5): fix qwen3.5 adaptive speculative behavior (#2542)
- change: chore(deps): bump deps (#2534)
- change: chore(pyo3): exclude mistralrs-pyo3 from default workspace members
- change: feat(core): add reasoning token budgets  (#2541)
- change: feat(core): support fast Xet downloads via hf-hub (#2536)
- change: feat(cuda): support mixed FP4 checkpoints and optimize qwen4_exp (#2526)
- change: feat(docs): add mistralrs download command
- change: feat(docs): improve website and docs (#2535)
- change: feat(gguf): add lfm3 gguf loading support (#2529)
- change: feat(lfm3): support d1 gguf (#2532)
- change: feat(models): add lfm3 omni and speech (#2527)
- change: feat(models): implement LFM3 text and VL (#2515)
- change: feat(models): support systemone models and LFM3-d1 (#2530)
- …and 2 more

## Architecture

- Unchanged — 0 containers · 1 contexts · 0 edges
