# What changed in amd-quark

> 65 changes in amd-quark 0.11 from its official release notes, upgrade guides and advisories: 13 breaking, 7 deprecations, 22 new APIs and options. Each one cited. Sources checked 2026-10-19. HTML: https://unstale.sh/pypi/amd-quark/

To see which of these apply to a repository, run `npx @unstale/cli init` in it.

## amd-quark 0.11

### Breaking changes

- The argument is removed from the example script and CLI; group size is now embedded in the quantization scheme name. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The argument is removed from the example script and CLI; group size is now embedded in the quantization scheme name. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The argument now takes repeated pattern-and-scheme pairs, allowing per-layer quantization scheme configuration. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- quark.torch.export.api.ModelExporter is removed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- quark.torch.export.api.ModelImporter is removed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- freeze now permanently quantizes weights by running QDQ on them, avoiding repeated QDQ on static weights in subsequent calls. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- OnnxSlim replaces ONNX Simplifier in preprocessing before quantization. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- ONNX QLayerConfig renames activation to input_tensors; layer-specific quantization configuration now specifies quantization per tensor. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The quark_format export format is removed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The CLI is refactored around LLMTemplate, removes redundant features and arguments, exports only Hugging Face safetensors, and evaluates perplexity on wikitext-2. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The LLM evaluation module moves to quark.contrib.llm_eval, where perplexity evaluation is shared between the CLI and examples. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- ONNX quantization configuration classes and algorithms are imported uniformly from quark.onnx, and QLayerConfig uses input_tensors instead of activation. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Module replacement utilities are now located in quark.torch.utils.module_replacement. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### Deprecations

- Config remains available as a backward-compatible alias but is deprecated in favor of QConfig. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- QConfig replaces Config as the quantization configuration class. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- QLayerConfig replaces QuantizationConfig as the layer quantization configuration class. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- QTensorConfig replaces QuantizationSpec as the quantization tensor configuration class. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Legacy scheme names are deprecated in the example script and CLI in favor of simplified built-in names, with group size encoded in names such as int4_wo_32. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- QuantizationConfig remains available as a backward-compatible alias but is deprecated in favor of QLayerConfig. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- QuantizationSpec remains available as a backward-compatible alias but is deprecated in favor of QTensorConfig. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### New APIs and options

- DeepSeek-OCR is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- DeepSeek-V3.2 is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Adds file-to-file quantization for ultra-large models, supporting weight-only quantization or dynamic activation quantization plus weight quantization; it exports hf_format only and accepts pre-quantized inputs for re-quantization. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- GLM-4.7 is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Kimi-K2-Instruct is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Kimi-K2-Thinking is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Kimi-K2.5 is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Minimax-M2.1 is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- An optional config parameter is now available for onnxslim optimization. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Qwen3 Coder is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Qwen3 Coder-Next is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Qwen3 MoE is now supported out of the box. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- A YAML file can now be used as input for custom preprocessing of float models before quantization. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Adds hierarchical and parallel hyperparameter search, custom objectives, sampler choices, checkpoint resume, visualization, and output saving. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Specific model inputs or outputs can now be selected for conversion from NCHW to NHWC. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Setting online_r1_rotation=True in RotationConfig enables online rotations, applying the activation rotation during inference rather than fusing it in a preceding layer. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Adds a graph transformation tool for preprocessing, including constant folding, operator fusion, redundant-node removal, and input/output streamlining. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The example script supports custom algorithm configurations through this argument. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- This environment variable sets the logging level, with debug, warning, error, and critical given as examples. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Common model preparation and data preparation utilities are available from quark.torch.utils for reuse. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Joint rotations and SmoothQuant scales can be fine-tuned as a non-destructive transformation; support is tested for llama, qwen3, qwen3_moe, and gpt_oss architectures. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- TensorQuantOverrides replaces MixedPrecisionTensor and supports per-tensor or per-channel, symmetric or asymmetric quantization, additional parameters, and precedence for multiple settings. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### Performance

- Memory optimization now applies to all calibration methods, particularly reducing memory use during activation data collection. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Parallel computation significantly speeds up the calibration process. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- CPU/GPU communication is removed from quantization observers, improving quantization and runtime speed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The default mode of MinMSE uses less memory to prevent out-of-memory errors. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- An efficient activation quantization kernel fuses scale computation and QDQ when quantizing models. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### Behavior and default changes

- Dry-run compatibility checking is enabled by default, with clearer error messages when model loading fails. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Calibration now includes token distribution coverage warnings. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- LLM PTQ workflows now perform a pre-quantization compatibility check for transformers. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Fast Finetune now uses a fixed seed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The expert calibration token counter is disabled by default and can be enabled by setting this environment variable to 1. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The rotation and quarot pre-quantization algorithms are combined into one rotation algorithm configurable with RotationConfig; the default applies only R1 rotation. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The QDQ operator is now compiled with torch.compile by default, enabling speedups that depend on the quantization scheme. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### Tooling

- Profiling reports individual quantization-stage latency and supports CPU memory traces and real-time ROCm GPU memory monitoring. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)

### Fixes and other changes

- Scale extraction now handles int32_data in ONNX initializers. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Fast finetuning now infers kernel size from weights when Conv nodes omit the kernel_shape attribute. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Model inference output now handles NaN and Inf values. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- Weight calibration now covers weights outside the forward path to ensure complete calibration. (0.11.1) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11.1)
- The out-of-bounds axis issue for scalar weights or biases is fixed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Memory duplication and out-of-memory issues when loading gpt_oss models for quantization are fixed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- The percentile value selection issue is fixed. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- Quark ONNX adds support for Python 3.12. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- AMD Quark 0.11 is tested against PyTorch 2.9, giving developers a tested PyTorch compatibility target. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)
- AMD Quark 0.11 is compatible with upstream transformers 4.57. (0.11) — [release notes](https://github.com/amd/Quark/releases/tag/v0.11)