sparseml
sparseml copied to clipboard
Libraries for applying sparsification recipes to neural networks with a few lines of code, enabling faster and smaller models
The document that should guide the user through the creation of a KV cache model, that can be later used in the pipeline. This PR goes "in tandem" with: https://github.com/neuralmagic/deepsparse/pull/1149
It would be nice to be able to use pytorch-lightning 2.0. We switch over to lightning 2.0 as soon as torch 2.0 was released, but would love to be able...
This PR adds support for variable-bit weight quantization in the ONNXToDeepsparse exporter. This affects two steps: - Conversion of intiailziers to unit8 - Clipping in quantization of weight arrays **Test...
fixes for new install pathway
- Renames `BaseManager.phase` -> `BaseManager.phase_at_end_of` - Clarifies behavior - Integrates saving checkpoints based on phases into torchvision # Test Plan Ran the following recipe: ```yaml version: 1.1.0 training_modifiers: - !EpochRangeModifier...
--- - To see the specific tasks where the Asana app for GitHub is being used, see below: - https://app.asana.com/0/0/1203900282961782
Update get_named_layers_and_params_by_regex in src/sparseml/pytorch/utils/helpers.py to still match in the DDP training case, where module. is prepended to layer names. I came across this issue when trying to run the suggested...
**What is the URL, file, or UI containing proposed doc change** I recognize that this section is requesting some set of changes I've attempted to enact for your reference, but...
The quantization modifier is logging detailed info such as number of bits for each layer, taking too many panels in wandb. This is to guard it under the "verbose" flag.
**NOTE: this PR is targeting `torchvision-phases` branch, not main!** Similar to #1499, this updates checkpoint saving logic to save checkpoints with the phase the model is in appended to the...