🖼️ Available 8 models from 1 repositories
Filter by type:
gemma-2-9b-it
Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. They are text-to-text, decoder-only large language models, available in English, with open weights for both pre-trained variants and instruction-tuned variants. Gemma models are well-suited for a variety of text generation tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as a laptop, desktop or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone.
gemma-2-9b-it-sppo-iter3
Self-Play Preference Optimization for Language Model Alignment (https://arxiv.org/abs/2405.00675) Gemma-2-9B-It-SPPO-Iter3 This model was developed using Self-Play Preference Optimization at iteration 3, based on the google/gemma-2-9b-it architecture as starting point. We utilized the prompt sets from the openbmb/UltraFeedback dataset, splited to 3 parts for 3 iterations by snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset. All responses used are synthetic.
sunfall-simpo-9b
Crazy idea that what if you put the LoRA from crestf411/sunfall-peft on top of princeton-nlp/gemma-2-9b-it-SimPO and therefore this exists solely for that purpose alone in the universe.
Repository: localaiLicense: gemma
Link #1sunfall-simpo-9b-i1
Crazy idea that what if you put the LoRA from crestf411/sunfall-peft on top of princeton-nlp/gemma-2-9b-it-SimPO and therefore this exists solely for that purpose alone in the universe.
Repository: localaiLicense: gemma
Link #1gemma-2-9b-it-abliterated
Abliterated version of google/gemma-2-9b-it. The abliteration script (link) is based on code from the blog post and heavily uses TransformerLens. The only major difference from the code used for Llama is scaling the embedding layer back. Orthogonalization did not produce the same results as regular interventions since there are RMSNorm layers before merging activations into the residual stream. However, the final model still seems to be uncensored.
gemma-2-ataraxy-v3i-9b
Gemma-2-Ataraxy-v3i-9B is an experimental model that replaces the simpo model in the original recipe with a different simpo model and a writing model trained on Gutenberg, using a higher density. It is a merge of pre-trained language models created using mergekit, with della merge method using unsloth/gemma-2-9b-it as the base. The models included in the merge are nbeerbower/Gemma2-Gutenberg-Doppel-9B, ifable/gemma-2-Ifable-9B, and wzhouad/gemma-2-9b-it-WPO-HB. It has been quantized using llama.cpp.
Repository: localaiLicense: gemma
Link #1
quill-v1
Quill is a capable, humanlike writing model trained on a large dataset of late 19th and early 20th century writing from the Gutenberg Project. This model writes with a natural cadence and low gpt-slop, having inherited some human qualities from the Gutenberg3 dataset. It writes with more simple, spare prose than the typical overly-adjectived LLM writing style. This model was trained using gemma-2-9b-it as the base. The training methods used were ORPO (gently) then SIMPO (less gently).
delirium-v1
This model was cooked a bit too long during SIMPO training. It writes like Hunter S. Thompson 2 days into an ether binge. It's grotesque, dark, grimy and genius. It's trained on an experimental gutenberg + antislop dataset. This contains the original two gutenberg sets by jondurbin and nbeerbower, as well as a subset of my own set, gutenberg3. The antislop pairs were generated with gemma-2-9b-it, with one sample generated with the AntiSlop sampler and the rejected sample generated without.