权重:https://huggingface.co/allenai/OLMo-7B 代码:https://github.com/allenai/OLMo 数据:https://huggingface.co/datasets/allenai/dolma 评估:https://github.com/allenai/OLMo-Eval 适配:https://github.com/allenai/open-instruc
InstructGPT 的数据集虽然从未公开发布,但一些开源项目尝试复现其设置,并收集自身的数据,例如 OpenAssistant 项目。这些项目旨在模仿 OpenAI 使用人工标注者创建对话的过程。 举例来说,一个提示可能是:“请写一个简短的介绍,介绍‘买方垄断’(monopsony)这一术语在经济学中的相关性,并举例说明。” 然后,另一个人会...
Toolkit for linearizing PDFs for LLM datasets/training - Merge branch 'main' of https://github.com/allenai/olmocr · allenai/olmocr@4e990e2
公开「后训练」一切,性能超越Llama 3.1 Instruct 最近,艾伦人工智能研究所(Ai2)发布了一个全新的开源模型——**Tülu 3**。这个模型不仅在性能上超过了Llama 3.1 Instruct,还在技术报告中详细公开了后训练的每一个细节。这标志着开源模型在后训练领域的重大突破! ## 主要亮点 - **性能卓越**:Tülu 3 的 8B ...
训练405亿参数模型突破了技术极限,需要32个计算节点和256个GPU一起工作,每个训练步骤需要35分钟。尽管由于计算限制,Tülu不得不提前结束培训,但其性能优于其他开源模型,如Llama 3.1 405B Instruct和Nous Hermes 3 405B。用户可以在AI2 Playground中测试模型,代码在GitHub上可用,模型在Hugging Face上可用。链接...
9 logseq A privacy-first, open-source platform for knowledge management and collaboration. Download link: http://github.com/logseq/logseq/releases. roadmap: http://trello.com/b/8txSM12G/roadmap logseq 36786 10 sequence Immutable, scalable, and easy to use ledger service. hoophq 498 11 works...
git clone https://github.com/Shark-NLP/OpenICL cd OpenICL pip install -e . Afterwards, we can run evaluation using the following command: CUDA_VISIBLE_DEVICES=0, python icl.py --model_path OUTPUT_PATH/MERGED_PATH Note you may have to modify the model initialization part of OpenICL in ...
Divergence Instruction (Read more on arXiv or HuggingFace) Weijian Luo, Debing Zhang, Colin Zhang, Weimin Bai, smallAI i) The paper introduces Uni-Instruct, a novel framework for one-step diffusion model distillation. ii) The research aims to unify existing one-step diffusion distillation metho...
unslothai- A framework that specializes in efficient fine-tuning. On its GitHub page, you can find ready-to-use fine-tuning templates for various LLMs, allowing you to easily train your own data for free on the Google Colab cloud.
@@ -164,7 +164,7 @@ https://github.com/hiyouga/LLaMA-Factory/assets/16256802/ec36a9dd-37f4-4f72-81bd 164 164 | [OLMo](https://huggingface.co/allenai) | 1B/7B | q_proj,v_proj | - | 165 165 | [PaliGemma](https://huggingface.co/google) | 3B | q_proj,v_proj | gemm...