LLM · Text Generation
GenStructDolphin 7B Slerp
Mistral 7B merge of Genstruct-7B and Dolphin 2.6 built with mergekit slerp for instruction-following chat.
About this model
GenStructDolphin-7B-Slerp is a text-generation model for Hugging Face Transformers, created by merging two Mistral-7B-based models with the slerp merge method via LazyMergekit. It combines NousResearch/Genstruct-7B, a model tuned for generating instruction-following datasets, with cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser, an Uncensored Dolphin variant fine-tuned for conversational responses. The result is a single 7B model that inherits both instruction-structure awareness and fluent chat behavior.
The merge configuration is fully documented in the model card: both models contribute layers 0 through 32, with the base model set to NousResearch/Genstruct-7B, dtype bfloat16, and per-layer t values applied to self-attention ([0, 0.5, 0.3, 0.7, 1]) and MLP blocks ([1, 0.5, 0.7, 0.3, 0]) with a default t of 0.5. As a merge rather than a fine-tune, there is no separate trigger word or instance prompt — the model is used through its chat template like any Mistral-based instruct model.
The model card includes a ready-to-run transformers pipeline example: load the tokenizer and model in float16 with device_map auto, apply the chat template, and generate with do_sample, temperature 0.7, top_k 50, and top_p 0.95. These settings are the recommended defaults for interactive use. The license is not specified on the model card, so check both base model repositories (Genstruct-7B and Dolphin 2.6 Mistral) before commercial use.
Learn more
Project signals
- 15 Hugging Face downloads
- 2 Hugging Face likes
Topics
transformers · safetensors · mistral · text-generation · merge · mergekit · lazymergekit · nousresearch/genstruct-7b · cognitivecomputations/dolphin-2.6-mistral-7b-dpo-laser · conversational