# What changed in ai-edge-quantizer > 28 changes in ai-edge-quantizer 0.4 to 0.10 from its official release notes, upgrade guides and advisories: 3 breaking, 11 new APIs and options. Each one cited. Sources checked 2026-10-18. HTML: https://unstale.sh/pypi/ai-edge-quantizer/ To see which of these apply to a repository, run `npx @unstale/cli init` in it. ## ai-edge-quantizer 0.10 ### New APIs and options - Adds an experimental algorithm for FULLY_CONNECTED that combines activation-aware per-channel activation scaling with optimal weight clipping. (0.10.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.10.0) - Adds uint8 data type support for model INPUT/OUTPUT tensors in recipes, enabling compatibility with some hardware accelerators. (0.10.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.10.0) ### Fixes and other changes - Recipe validation no longer requires weight_tensor_config for ops without weights, such as model INPUT/OUTPUT, so dummy weight configs are unnecessary. (0.10.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.10.0) ## ai-edge-quantizer 0.9 ### Breaking changes - Same-size Hadamard matrices now use a single tensor, and inserted Hadamard matrices are float. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) ### New APIs and options - Recipes and buffer packing were added for 2-bit model quantization. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Weight recovery algorithms now support scale recovery and blockwise quantization, with ready-to-use 4-bit and 8-bit named recipes. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Calibrator and Quantizer gained a profiler-based calibration mode for faster, more memory-efficient XNNPACK-assisted calibration. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - A calibration API was added that lets developers reuse the inference pipeline for calibration. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - GPTQ support was added as a quantization algorithm. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Int4 matrix multiplication Hadamard transforms gained a max_hadamard_size control. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - LiteRT-LM files can now be read, quantized, and written, including extracting serialized TFLite models and applying recipes across TFLiteModel sections. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Model validation now supports multiple metrics. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) ### Performance - Large tensors are processed in chunks along the leading dimension to avoid generating large intermediate values. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - These buffers are explicitly garbage-collected immediately after output buffer writing to reclaim memory. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Model serialization uses disk-backed memory mappings and chunk-by-chunk file writing to prevent OOM errors on large LLMs. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Buffer mapping IDs are tracked, producer and consumer ops are pre-computed, and empty buffers are pruned to reduce serialized file size and parsing overhead. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Quantization uses zero-copy array views and in-place core operations instead of unnecessary copies, lowering peak memory usage. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) ### Tooling - A CLI was introduced for running model quantization workflows from the terminal. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) ### Fixes and other changes - Calibration metadata tracking was added. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Crashes in float casting operations with non-constant weights were fixed. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Output tensors are no longer inadvertently rotated during fully connected op processing. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Accidental double-quantization of input tensors is prevented. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) - Progress bars and reports were added for quantization steps and calibration passes. (0.9.0) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.9.0) ## ai-edge-quantizer 0.4 ### Breaking changes - Bias error checks are disabled. (0.4.2) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.4.2) - Python 3.9 support was dropped and Python 3.13 support was added. (0.4.2) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.4.2) ### New APIs and options - Progress bars and reporting utilities were added. (0.4.2) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.4.2) ### Behavior and default changes - Paths now support PathLike objects. (0.4.2) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.4.2) - The skip_checks configuration is now propagated. (0.4.2) — [release notes](https://github.com/google-ai-edge/ai-edge-quantizer/releases/tag/v0.4.2)