# 0.10.0

## Release Information

- OS & Framework support - Tested with the following configurations:

> 
> 
> - Non-Gen AI:
> 
> 
> 
> > 
> > 
> > - Frameworks: ONNX, TFLite, Pytorch
> >         - Host Platforms: Linux-x86\_64 (Ubuntu 22.04, Ubuntu 24.04), Linux-aarch64 (Ubuntu 24.04, Python 3.12), Windows-x86\_64 (10+), Windows-arm64 (10+)
> >         - Target Platforms: Linux-x86\_64 (Ubuntu 22.04, Ubuntu 24.04), Linux-aarch64 (Ubuntu 24.04), Windows-arm64 (10+), Android-arm64, QNX
>     - Gen AI:
> 
> 
> 
> > 
> > 
> > - Frameworks: ONNX, GGUF
> >         - Host Platforms: Linux-x86\_64 (Ubuntu 22.04, Ubuntu 24.04)
> >         - Target Platforms: Android-arm64, Linux-aarch64
- QAIRT SDK - Tested with version (2.47.0, 2.48.0, 2.49.0(Default))

Note

For QAIRT SDK, other supported SDK versions can be listed using `qairt-vm fetch --list`.

## Highlights

- Default QAIRT SDK updated to 2.49.0
- Minimum supported QAIRT SDK updated to 2.46.0
- New features in QAIRT Dev Python APIs (compatible with QAIRT SDK 2.46.0+, unless noted otherwise)

> 
> 
> - Compile API
> 
> 
> 
> > 
> > 
> > - `soc_model` is now accessible via the compile and context-binary generation APIs, letting
> > host scripts select a simulated SoC for offline builds without additional hardware
> > introspection (see [GGUF Inference on HTP](https://docs.qualcomm.com/doc/80-87189-2/topic/gguf_builder.html#gguf-builder))
>     - Gen AI API Enhancements
> 
> 
> 
> > 
> > 
> > - `GenAIBuilderHTP` and the GGUF builder now read the chat template directly from GGUF
> > metadata when no template is provided by the caller, so model-specific prompt formatting
> > (e.g., Llama 3, Qwen 2.5) is applied automatically without manual configuration
> > (see [GGUF Inference on HTP](https://docs.qualcomm.com/doc/80-87189-2/topic/gguf_builder.html#gguf-builder))
> >         - `evaluate_perplexity()` added as a public method on `GenAIExecutable` and
> > `T2TExecutor` — standalone on-device perplexity evaluation via the Genie Accuracy API,
> > usable on any built model without the experimental pipeline
>     - [Preview] QAIRT Dev Pipeline (`qairt.experimental.pipeline`)
> 
> 
> 
> > 
> > 
> > - LoRA is now supported through the pipeline for both beginner and advanced users —
> > attach PEFT adapters, quantize per use-case, and build end-to-end; validated with
> > Qwen3 4B and Llama 3.2 3B models
> >         - `LLMPipeline.evaluate()` and per-stage `evaluate()` now compute perplexity metrics
> > at any pipeline stage, enabling model-quality tracking across the quantization and build
> > workflow (see [Customizing the Pipeline](https://docs.qualcomm.com/doc/80-87189-2/topic/pipeline_customization.html#pipeline-customization))
> >         - `QuantizationOptStage` now supports composable multi-step optimization recipes —
> > chain multiple post-quantization optimizer passes (AdaScale, SeqMSE, GPTAQ) via the
> > `quantization_opt` key in a pipeline YAML recipe
> > (see [Quantization Recipes](https://docs.qualcomm.com/doc/80-87189-2/topic/pipeline_quantization_recipes.html#pipeline-quantization-recipes))

## Resolved Issues

> 
> 
> - Fixed `CalibratorRecipe` silently dropping `model_preparation_path` on a mid-run quantization
> failure; the path is now preserved so a retry can skip re-preparation and use cached artifacts.

## Deprecation Notices

To be removed in 0.11.0:

> 
> 
> - The ONNX Model interface (`OnnxModel` and its public methods `OnnxModel.load()`,
> `OnnxModel.split()`, `OnnxModel.mha2sha_v2()`) is deprecated. Migrate to `GraphContext`
> together with the simple optimizer APIs (`convert_mha_to_sha`, `split_llm`, `adapt_moe`).

To be removed in 0.12.0:

> 
> 
> - `GenAIConfig.allow_async_init` and `GenAIConfig.enable_graph_switching` are deprecated. Set
> these through `EngineConfig.htp.allow_async_init` and `EngineConfig.htp.enable_graph_switching`
> and pass the `EngineConfig` to `get_executor()` / `T2TExecutor` instead.
> - `GenAIExecutor` (`qairt.gen_ai_api.executors.gen_ai_executor`) is deprecated and has been
> renamed to `GenAIExecutable` (`qairt.gen_ai_api.executors.gen_ai_executable`). Update imports
> to the new module.

## Known Issues

> 
> 
> - Gen AI Builder is temporarily unsupported on Windows-arm64. Use Linux-x86\_64 as the host platform
> for Gen AI workflows.
> - Windows on ARM64 is not functional. This is a known issue (AISW-188015).
> - Models with batch size greater than 1
> 
> 
> 
> > 
> > 
> > - A discrepancy exists between the qnn-net-run CLI and the Python API regarding data preparation
> > for batch sizes greater than 1. While the CLI automatically groups individual tensor paths from
> > an input list into a batch, the Python API requires inputs to be pre-batched (concatenated into
> > a single raw file) prior to execution

Last Published: Aug 26, 2026

[Previous Topic
Known Issues](https://docs.qualcomm.com/bundle/publicresource/80-87189-2/topics/0_9_0.md) [Next Topic
0.10.1](https://docs.qualcomm.com/bundle/publicresource/80-87189-2/topics/0_10_1.md)