Skip to content

Latest commit

 

History

History
151 lines (131 loc) · 10.5 KB

CHANGELOG.md

File metadata and controls

151 lines (131 loc) · 10.5 KB
Raw

v0.19

Breaking Changes

  • InstructLab now uses XDG-based directories on macOS, similar to Linux. Users are advised to re-initialize their config files and remove cached models.
  • Removed unused argument --rouge-threshold of ilab data generate
  • Removed the following aliased commands:
    • convert
    • diff
    • download
    • evaluate
    • init
    • list
    • sysinfo
    • test
  • Intel Gaudi software has been updated to 1.17.1 with Python 3.11 and Torch 2.3.1 support.
  • --legacy has been removed and replaced with --pipeline=simple in ilab model train

Features

  • ilab config init now auto detects your hardware when running on Nvidia enabled systems and chooses the best train profile. It does this by checking first if your system directly matches one of our supported train profiles and then attempts to match the vRAM for each profile to the total vRAM on your system.
  • Add log_format to the config.yaml file to allow for customizing the log format.
  • ilab model evaluate --max-workers=auto is now supported and is the default option. When auto is specified, the optimal value is determined based on your GPUs, CPUs, and configuration.
  • ilab model train now supports --pipeline. The supported pipelines are simple, full, and accelerated. Simple preserves the functionality found in --legacy and current MacOS training. Full introduces a new training loop optimized for CPU and MacOS performance. accelerated allows users with dedicated graphics cards to run the full fine tuning and multi-phase training found in our training library.
  • --device in ilab model train now supports mps which stands for Metal Performance Shaders. This is a PyTorch device for MacOS training that allows us to utilize the same code path for Linux and MacOS.
  • Multi-phase training with ilab model train --strategy lab-multiphase is now resumeable! In the case of a machine failure or an incidental stop, progress is tracked in a new journal file, rendered in yaml. Upon restarting training, the user can confirm whether they would like to proceed with a pre-existing training run (one that might have only evaluated a few checkpoints of the first eval phase, for instance) or restart from scratch.
  • Allow users to pick a distributed training backend framework for GPU accelerated training between 'fsdp' and 'deepspeed'. Also add support for FSDP specific configuration options.

v0.18.1

Features

  • ilab data generate and ilab taxonomy diff now support --taxonomy-base=empty to allow specifying that all taxonomy files in the supplied repo should be included.

v0.18

Features

  • ilab data generate now supports parallelized data generation across batches of the seed data when running with a the vLLM serving. The --batch-size argument can be used to control this behavior.
  • ilab model download now supports downloading models from OCI registries. Repositories that are prefixed by "docker://" and specified against --repository are treated as OCI registries.
  • ilab now uses dedicated directories for storing config and data files. On Linux, these will generally be the XDG directories: ~/.config/instructlab for config, ~/.local/share/instructlab for data, and ~/.cache for temporary files, including downloaded models. On MacOS, both the config and data is located at ~/Library/Application Support/instructlab.
  • A new ilab config show command is introduced as a convenience feature, which prints out the contents of the actively loaded config, not just the contents of the config file.
  • ilab system: A new command group named ilab system has been added which will serve as the basis for all system-related commands. This currently contains ilab system info as its only sub-command.
  • Add vLLM backend to serve, chat and generate commands.
  • Add --backend flag to ilab model serve command to allow for specifying the backend to use when serving a model. This is useful when you have multiple backends installed and want to specify which one to use. Currently, the only supported backend are llama-cpp and vllm.
  • Update llama-cpp-python to latest upstream release 0.2.79 to address poor results of synthetic data generation and local training.
  • Adding ilab model evaluate which uses the new backend serving functionality. Evaluate offers two standard benchmarks (mt_bench and mmlu) as well as two variations (mt_bench_branch and mmlu_branch) which are integrated with the ilab workflow to evaluate new skills and knowledge. Includes --gpus option for specifying number of gpus to utilize when serving models for evaluation (currently applicable for vLLM only). Also includes --merge-system-user-message flag to enable Mistral based judge models and a --enable-serving-output flag that configures whether the output of the model serving backend is suppressed.
  • The ilab command now accepts a -v / --verbose option to enable debug logging. ilab -vv or ilab --verbose --verbose enables more verbose debug logging.
  • ilab model test generic support
  • Add --chat-template option to ilab model serve to support customization of the chat template for both vLLM and llama.cpp backends. Options include 'auto' (current behavior, ilab provides its own template), 'tokenizer' (uses the model's tokenizer config), and an external file name.
  • Default log format changes to include the logger name in the logs.
  • ilab data generate now supports a new and more extensive pipeline with the option --pipeline full. This option requires mixtral-8x7b-instruct as the teacher model.
  • The instructlab package now uses optional dependencies for each supported hardware cpu, cuda, hpu, mps, and rocm. To install InstructLab for e.g. NVIDIA CUDA, use pip install instructlab[cuda].
  • Add a --enable-serving-output flag for ilab data generate. This flag determines whether vLLM will have its output suppressed when it serves the teacher model in the background.
  • The generate section of the config now has a teacher section. This section configures the teacher model when it is automatically served in the background. This new section has the same values as the serve section of the config.
  • Support for ILAB_GLOBAL_CONFIG environment variable: When set, this environment variable specifies a global configuration file that serves as the template for the ~/.config/instructlab/config.yaml user space config. This bypasses the interactive mode in ilab config init and can be used to specify alternative configurations for any command, ensuring that defaults such as taxonomy repositories and base models are honored from the global config.
  • ilab model list: a new command which lists all GGUF and Safetensor Models on the system.
  • ilab data list: a new command which lists the generated datasets in the user's datasets directory.
  • Legacy Linux training now supports the new messages format. When a dataset is provided in the HuggingFace messages format, ilab will automatically convert it back into the legacy format.
  • Legacy Linux training is now compatible with the phase07 pretraining format.
  • Add support for ILAB_TRAIN_PROFILE_DIR which will point to the template train profiles to be brought into the train_configuration directory.
  • Add interactive prompt for users to choose their train profile.
  • The generate section of the config now has a pipeline value. This value sets a default value and can be overridden by the --pipeline flag. The default for this value is 'simple'.

Breaking Changes

  • ilab: Deprecation of Python 3.9 support and withdrawal of Python 3.12 support Due to changes to training requiring the usage of GPTDolomite, Python 3.9 is no longer supported and Python 3.12 support is currently withdrawn. If you are using either of these versions, you will need to start using either Python 3.10 or Python 3.11 to use this and subsequent versions of the CLI.
  • ilab model train: The '--device' parameter no longer supports specifying a GPU index (e.g., 'cuda:0'). To use a specific GPU, set the visible GPU before running the train command.
  • ilab init: With the introduction of a dedicated storage system within the ilab CLI, ilab init and ilab config init will now output and read the config file from the platform's config directory under the instructlab package.
  • ilab taxonomy and ilab data: The ilab CLI now uses the platform's dedicated data directory to store the taxonomy under the instructlab/taxonomy directory as a default.
  • ilab data: The default directory for new datasets is now under instructlab/datasets in the platform's dedicated data directory under the instructlab package.
  • ilab model: The default location for saved and downloaded models is now under instructlab/models in the platform's dedicated data directory under the instructlab package. Outputted checkpoints now live in the instructlab/checkpoints directory under the platform's dedicated program cache directory.
  • ilab model chat: Chatlogs are now stored under the instructlab/checkpoints directory in the platform's dedicated data directory under the instructlab package.
  • The --num-instructions option to ilab data generate has been deprecated. See --sdg-scale-factor for an updated option providing similar functionality.
  • ilab model train --legacy: Trained GGUF models are now saved in the global user checkpoints directory. Previously, checkpoints were always saved into a directory local to where the user called it from.

Fixes

  • ilab config: Fixed a bug where ilab didn't recognize train.lora_quantize_dtype: null as a valid value.
  • ilab model chat: Fixed an issue where the default served model couldn't be resolved when running model besides the default merlinite-7b-lab-Q4_K_M.gguf.

v0.17

Features

ilab command redesign

The ilab command redesign included in v0.17 introduces a new command structure that follows a resource group design. This means that commands that once were something like ilab chat now are ilab model chat. The new groups are model, data, taxonomy, and config. The commands that fall under these are all of the pre-existing ilab commands just now grouped by the resource which the command commonly deals with.

The old command structure is still aliased to work but will be removed in 0.19.0. This means for 0.17.0 and 0.18.0 the aliases will exist and work as expected.

Breaking Changes