Yahoo Poland Wyszukiwanie w Internecie

Search results

1. github.com › microsoft › BitNetmicrosoft/BitNet: Official inference framework for 1-bit LLMs -...

   github.com › microsoft › BitNet

   • Zapisane w pamięci cache

   bitnet.cpp is the official inference framework for 1-bit LLMs (e.g., BitNet b1.58). It offers a suite of optimized kernels, that support fast and lossless inference of 1.58-bit models on CPU (with NPU and GPU support coming next).

2. huggingface.co › blog › 1_58_llm_extreme_quantizationFine-tuning LLMs to 1.58bit: extreme quantization made easy -...

   huggingface.co › blog › 1_58_llm_extreme_quantization

   • Zapisane w pamięci cache

   18 wrz 2024 · BitNet is a special transformers architecture that represents each parameter with only three values: (-1, 0, 1), offering a extreme quantization of just 1.58 ( l o g 2 (3) log_2(3) l o g 2 (3)) bits per parameter. However, it requires to train a model from scratch.

3. arxiv.org › html › 2410BitNet b1.58 Inference on CPUs - arXiv.org

   arxiv.org › html › 2410

   • Zapisane w pamięci cache

   21 paź 2024 · In this work, we introduce bitnet.cpp, a tailored software stack designed to unlock the full potential of 1-bit LLMs. Specifically, we develop a set of kernels to support fast and lossless inference of ternary BitNet b1.58 LLMs on CPUs.

4. huggingface.co › papers › 2402The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits

   huggingface.co › papers › 2402

   • Zapisane w pamięci cache

   28 lut 2024 · Recent research, such as BitNet, is paving the way for a new era of 1-bit Large Language Models (LLMs). In this work, we introduce a 1-bit LLM variant, namely BitNet b1.58, in which every single parameter (or weight) of the LLM is ternary {-1, 0, 1}.

5. github.com › AlarioAI › BitNetAlarioAI/bitnet: Train and evaluate 1.58 bits Neural Networks -...

   github.com › AlarioAI › BitNet

   • Zapisane w pamięci cache

   This repository not only provides PyTorch implementations for training and evaluating 1.58-bit neural networks but also includes a unique integration where the experiments conducted automatically update a LaTeX-generated paper.

6. huggingface.co › 1bitLLM › bitnet_b1_58-3BREADME.md · 1bitLLM/bitnet_b1_58-3B at main - Hugging Face

   huggingface.co › 1bitLLM › bitnet_b1_58-3B

   • Zapisane w pamięci cache

   29 mar 2024 · Here is the commands to run the evaluation: pip install lm-eval==0.3.0. python eval_ppl.py --hf_path 1bitLLM/bitnet_b1_58-3B --seqlen 2048. python eval_task.py --hf_path 1bitLLM/bitnet_b1_58-3B \ --batch_size 1 \ --tasks \ --output_path result.json \ --num_fewshot 0 \ --ctx_size 2048.

7. arxiv.org › html › 2402The Era of 1-bit LLMs: All Large Language Models are in 1.58 Bits

   arxiv.org › html › 2402

   • Zapisane w pamięci cache

   27 lut 2024 · More profoundly, the 1.58-bit LLM defines a new scaling law and recipe for training new generations of LLMs that are both high-performance and cost-effective. Furthermore, it enables a new computation paradigm and opens the door for designing specific hardware optimized for 1-bit LLMs.