> ## Documentation Index
> Fetch the complete documentation index at: https://docs.liquid.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Changelog

## `v0.9.0` - 2026-01-02

**New features**

* GGUF is now the default inference engine for model bundling, generating `.gguf` files for llama.cpp inference.
* Add `--executorch` flag to use ExecuTorch bundling instead of GGUF. ExecuTorch inference is deprecated and may be removed in a future version.
* Add `--mmproj-quantization` option for GGUF bundling of vision-language and audio models.
* Support downloading multiple `.gguf` files for GGUF bundle requests.

## `v0.8.0` - 2025-12-16

**Improvements**

* Update GGUF model download interface to take model name and quantization rather than manifest URL directly.

## `v0.7.0` - 2025-11-19

**New features**

* Download GGUF model files from JSON manifest
* `leap-bundle download <json-manifest-url> [--output-path <output-path>] [--overwrite]`
* Bundle request download will throw error if the file already exists in the output path.
* Use `--overwrite` option to allow overwriting the existing file.

**Bug fixes**

* Show correct filenames when uploading files for bundle creation

## `v0.6.0` - 2025-11-05

**New features**

* Support latest `LFM2-VL` model.

**Improvements**

* Move model type validation to server side for faster new model support in the future.

## `v0.5.1` - 2025-10-06

**Bug fixes**

* Add validation to reject Liquid audio model submissions.
* Liquid audio models are not supported by the bundle service yet.
* They will be in a future release.

## `v0.5.0` - 2025-09-24

**New features**

* Add `--quantization` option for the `create` command to specify the quantization type.
* Supported quantization types: `8da4w_output_8da8w`, `8da8w_output_8da8w`.
* Default quantization: `8da4w_output_8da8w`.

## `v0.4.0` - 2025-09-16

**New features**

* Add `resume` command to resume an interrupted or failed upload.

**Improvements**

* Add parallel upload support for faster upload speed.
* Users can choose between `--parallel` and `--sequential` upload modes for `create` and `resume` commands.
* By default, `--parallel` upload is used. If it fails, upload will fall back to `--sequential` mode.
* Add `--json` option to `create`, `resume`, and `list` commands to output the final results in JSON for easy parsing.
* Add `--last` option to `list` command to show only the last request.
* Show user-friendly information when upload fails.

**Fixes**

* Fix `--output-path` option for `download` command.
* Previously this option was ignored and the bundle file would always be downloaded to the current directory.

## `v0.3.0` - 2025-08-26

**New features**

* Support `lfm2-vl` models.

**Improvements**

* Show user-friendly information and error messages.
* Fix minor bugs.

## `v0.2.0` - 2025-08-12

**New features**

* Add `validate` command.

**Improvements**

* Add `Notes` columns in request table.
* Validate input model path before creating a request.
* Update and add processing error message.
* Show download URL and link to doc page.

## `v0.1.1` - 2025-07-30

**Improvements**

* Test and report network connectivity before model upload.

## `v0.1.0` - 2025-07-25

Initial release of the [`leap-bundle` package](https://pypi.org/project/leap-bundle).