epfml / landmark-attentionLinks
Landmark Attention: Random-Access Infinite Context Length for Transformers
☆426Updated last year
Alternatives and similar repositories for landmark-attention
Users that are interested in landmark-attention are comparing it to the libraries listed below
Sorting:
- The Truth Is In There: Improving Reasoning in Language Models with Layer-Selective Rank Reduction☆389Updated last year
- [ICML'24 Spotlight] LLM Maybe LongLM: Self-Extend LLM Context Window Without Tuning☆660Updated last year
- ☆415Updated 2 years ago
- A bagel, with everything.☆324Updated last year
- ☆95Updated 2 years ago
- Multipack distributed sampler for fast padding-free training of LLMs☆201Updated last year
- batched loras☆347Updated 2 years ago
- Merge Transformers language models by use of gradient parameters.☆207Updated last year
- ☆457Updated 2 years ago
- Extend existing LLMs way beyond the original training length with constant memory usage, without retraining☆729Updated last year
- Official repository for LongChat and LongEval☆531Updated last year
- ☆534Updated last year
- [ICML 2024] SqueezeLLM: Dense-and-Sparse Quantization☆704Updated last year
- ☆547Updated 10 months ago
- Official code for ReLoRA from the paper Stack More Layers Differently: High-Rank Training Through Low-Rank Updates☆467Updated last year
- Code for paper: "QuIP: 2-Bit Quantization of Large Language Models With Guarantees"☆385Updated last year
- ☆374Updated 2 years ago
- Code for the paper "QMoE: Practical Sub-1-Bit Compression of Trillion-Parameter Models".☆277Updated 2 years ago
- ☆575Updated last year
- Code for fine-tuning Platypus fam LLMs using LoRA☆629Updated last year
- Mass-editing thousands of facts into a transformer memory (ICLR 2023)☆522Updated last year
- ☆274Updated 2 years ago
- Inference code for Mistral and Mixtral hacked up into original Llama implementation☆369Updated last year
- Inference code for Persimmon-8B☆412Updated 2 years ago
- Experiments on speculative sampling with Llama models☆126Updated 2 years ago
- QLoRA: Efficient Finetuning of Quantized LLMs☆78Updated last year
- ModuleFormer is a MoE-based architecture that includes two different types of experts: stick-breaking attention heads and feedforward exp…☆224Updated last month
- Customizable implementation of the self-instruct paper.☆1,051Updated last year
- Positional Skip-wise Training for Efficient Context Window Extension of LLMs to Extremely Length (ICLR 2024)☆204Updated last year
- TART: A plug-and-play Transformer module for task-agnostic reasoning☆200Updated 2 years ago