Models made at Liquid AI
Maxime Labonne PRO
mlabonne
AI & ML interests
Post-training, model editing, quantization
Recent Activity
updated a model 10 days ago
mlabonne/Meta-Llama-3-120B-Instruct updated a model 10 days ago
mlabonne/lfm25-350m-struct published a model 10 days ago
mlabonne/lfm25-350m-structOrganizations
Uncensored models using abliteration. See this article for more information: huggingface.co/blog/mlabonne/abliteration
-
mlabonne/gemma-3-27b-it-abliterated
Image-Text-to-Text • 27B • Updated • 3.62k • • 333 -
mlabonne/gemma-3-27b-it-abliterated-GGUF
Image-Text-to-Text • 27B • Updated • 6.14k • 250 -
mlabonne/gemma-3-12b-it-abliterated-v2
Image-Text-to-Text • 12B • Updated • 432 • 21 -
mlabonne/gemma-3-4b-it-abliterated-v2
Image-Text-to-Text • 4B • Updated • 239 • 11
Family of 7B models that combine excellent reasoning and conversational abilities.
The first Mixture of Experts with phi-2 models.
Models and code related to the DPO fine-tuned OpenHermes-2.5-Mistral-7B
Llama and CodeLlama models trained to improve the performance in terms of code generation.
Models and datasets from my book. All the code is freely available at https://github.com/PacktPublishing/LLM-Engineers-Handbook
Fine-tuned abliterated merge of the best Llama 3 8B model. Highest MMLU score in its category.
-
mlabonne/NeuralDaredevil-8B-abliterated
Text Generation • 8B • Updated • 17.3k • • 279 -
mlabonne/Daredevil-8B-abliterated
Text Generation • 8B • Updated • 8.21k • • 60 -
mlabonne/Daredevil-8B
Text Generation • 8B • Updated • 104 • • 44 -
mlabonne/NeuralLlama-3-8B-Instruct-abliterated
Text Generation • 8B • Updated • 42 • • 9
MoE done using mergekit and LazyMergekit: https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb#scrollTo=d5mYzDo1q96y
Merges done using mergekit and LazyMergekit: https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb#scrollTo=d5mYzDo1q96y
Mistral and Llama models trained on a corpus of French and English data to act as a medical chatbot and ace exams.
Set of models fine-tuned using QLoRA on Google Colab with the Guanaco dataset.
Models made at Liquid AI
Models and datasets from my book. All the code is freely available at https://github.com/PacktPublishing/LLM-Engineers-Handbook
Uncensored models using abliteration. See this article for more information: huggingface.co/blog/mlabonne/abliteration
-
mlabonne/gemma-3-27b-it-abliterated
Image-Text-to-Text • 27B • Updated • 3.62k • • 333 -
mlabonne/gemma-3-27b-it-abliterated-GGUF
Image-Text-to-Text • 27B • Updated • 6.14k • 250 -
mlabonne/gemma-3-12b-it-abliterated-v2
Image-Text-to-Text • 12B • Updated • 432 • 21 -
mlabonne/gemma-3-4b-it-abliterated-v2
Image-Text-to-Text • 4B • Updated • 239 • 11
Fine-tuned abliterated merge of the best Llama 3 8B model. Highest MMLU score in its category.
-
mlabonne/NeuralDaredevil-8B-abliterated
Text Generation • 8B • Updated • 17.3k • • 279 -
mlabonne/Daredevil-8B-abliterated
Text Generation • 8B • Updated • 8.21k • • 60 -
mlabonne/Daredevil-8B
Text Generation • 8B • Updated • 104 • • 44 -
mlabonne/NeuralLlama-3-8B-Instruct-abliterated
Text Generation • 8B • Updated • 42 • • 9
Family of 7B models that combine excellent reasoning and conversational abilities.
MoE done using mergekit and LazyMergekit: https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb#scrollTo=d5mYzDo1q96y
The first Mixture of Experts with phi-2 models.
Merges done using mergekit and LazyMergekit: https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb#scrollTo=d5mYzDo1q96y
Models and code related to the DPO fine-tuned OpenHermes-2.5-Mistral-7B
Mistral and Llama models trained on a corpus of French and English data to act as a medical chatbot and ace exams.
Llama and CodeLlama models trained to improve the performance in terms of code generation.
Set of models fine-tuned using QLoRA on Google Colab with the Guanaco dataset.