Merge pull request #9 from unifyai/addCompressionTools

Add compression tools

compression/awq.yaml

@@ -0,0 +1,19 @@
+llm-awq:
+
+  name: "llm-awq"
+
+  image_url: https://github.com/mit-han-lab/llm-awq/blob/main/figures/overview.png
+
+  tags: 
+    - quantization
+    - pytorch
+    - llms
+    - open-source
+
+  url: https://github.com/mit-han-lab/llm-awq
+
+  description: "Efficient and accurate low-bit weight quantization (INT3/4) for LLMs, supporting instruction-tuned models and multi-modal LMs."
+
+  features:
+    - "INT 3/4 Weight Quantization for LLMs"
+    - "Calibration Data-set free quantization"

compression/bitsandbytes.yaml

@@ -0,0 +1,19 @@
+bitsandbytes:
+
+  name: "bitsandbytes"
+
+  image_url: https://huggingface.co/blog/assets/96_hf_bitsandbytes_integration/Thumbnail_blue.png
+
+  tags: 
+    - quantization
+
+  url: https://github.com/TimDettmers/bitsandbytes
+
+  description: "The bitsandbytes is a lightweight wrapper around CUDA custom functions,
+                 in particular 8-bit optimizers, matrix multiplication (LLM.int8()),
+                 and quantization functions.""
+ 
+  features:
+    - "8 Bit Weight Only Quantization"
+    - "Support for QLora Finetuning through NF4 dtype"
+    - "Calibration free Zero Shot Quantization"

compression/gptq.yaml

@@ -0,0 +1,20 @@
+gptq:
+
+  name: "gptq"
+
+  image_url: https://www.marktechpost.com/wp-content/uploads/2023/08/Screenshot-2023-08-26-at-6.10.51-PM.png
+
+  tags: 
+    - quantization
+
+  url: https://github.com/IST-DASLab/gptq
+
+  description: "GPTQ is a technique for post-training quantization, which is used to quantize large language models (LLMs) such as GPT.
+                This method minimizes the storage requirements for GPT models by decreasing the bit count necessary to represent each 
+                weight within the model, reducing it from 32 bits to a mere 3-4 bits or even 2 bits."
+
+  features:
+    - "2, 3, 4 bit Weight Quantization for LLMs"
+    - "Requires Calibration Dataset"
+    - "Calibration free Zero Shot Quantization"
+    - "Supports AMD GPU out of the box."

compression/huggingfaceoptimum.yaml

@@ -0,0 +1,28 @@
+optimum:
+
+  name: "hugginface/optimum"
+
+  image_url: https://huggingface.co/front/thumbnails/docs/optimum.png
+
+  tags: 
+    - quantization
+    - pruning
+    - knowledge distillation
+    - pytorch
+    - open-source
+
+  url: https://github.com/huggingface/optimum
+
+  description: "🤗 Optimum is an extension of 🤗 Transformers and Diffusers, providing a set of optimization
+                   tools enabling maximum efficiency to train and run models on targeted hardware, while
+                   keeping things easy to use."
+
+  features:
+    - "Post training dynamic quantization"
+    - "Post training static quantization"
+    - "Quantization Aware Training"
+    - "Mixed Precision Training"
+    - "Pruning"
+    - "Distillation"
+    - "Joint Pruning, Quantization and Distillation"
+    - "Graph Optimization using Habana and ONNX"

compression/tensorly-torch.yaml

@@ -0,0 +1,21 @@
+
+tensorly-torch:
+
+  name: "TensorLy-Torch"
+
+  image_url: https://tensorly.org/torch/dev/_images/tensorly-torch-pyramid.png
+
+  tags: 
+    - tensorization
+    - pytorch
+
+  url: https://github.com/tensorly/torch
+
+  description: "TensorLy-Torch is a Python library for deep tensor networks that builds on top of TensorLy and PyTorch.
+               It allows to easily leverage tensor methods in a deep learning setting and comes with all batteries included."
+
+  features:
+    - "Tensorized/Factorized Layers"
+    - "Tensor Regression and Contraction"
+    - "Tensor Dropout"
+    - "Tensor Hooks"

compression/tensorly.yaml

@@ -14,13 +14,15 @@ tensorly:
     - jax
     - open-source
 
-  url: https://github.com/intel/neural-compressor
+  url: https://github.com/tensorly/tensorly
 
   description: "TensorLy is a Python library that aims at making tensor learning simple and accessible. It allows simple performing of tensor decomposition,
   tensor learning and tensor algebra. Its backend system allows to seamlessly perform computation with NumPy,
   PyTorch, JAX, MXNet, TensorFlow or CuPy, and run methods at scale on CPU or GPU."
 
   features:
+    - "Tensor Algebra"
     - "Tucker Decomposition"
     - "Canonical Polyadic Decomposition"
     - "Tensor Train Decomposition"
+    - "Parafac Decomposition"