From f15a93b19b885ad139e12685272dd9ab95de5140 Mon Sep 17 00:00:00 2001 From: Ettore Di Giacinto Date: Mon, 5 Aug 2024 10:11:00 +0200 Subject: [PATCH] models(gallery): add humanish-roleplay-llama-3.1-8b-i1 (#3126) Signed-off-by: Ettore Di Giacinto --- gallery/index.yaml | 17 +++++++++++++++++ 1 file changed, 17 insertions(+) diff --git a/gallery/index.yaml b/gallery/index.yaml index bcfa4f35..c80455a8 100644 --- a/gallery/index.yaml +++ b/gallery/index.yaml @@ -307,6 +307,23 @@ - filename: L3.1-70b-glitz-v0.2.i1-Q4_K_M.gguf sha256: 585efc83e7f6893043be2487fc09c914a381fb463ce97942ef2f25ae85103bcd uri: huggingface://mradermacher/L3.1-70b-glitz-v0.2-i1-GGUF/L3.1-70b-glitz-v0.2.i1-Q4_K_M.gguf +- !!merge <<: *llama31 + name: "humanish-roleplay-llama-3.1-8b-i1" + icon: https://cdn-uploads.huggingface.co/production/uploads/5fad8602b8423e1d80b8a965/VPwtjS3BtjEEEq7ck4kAQ.webp + urls: + - https://huggingface.co/mradermacher/Humanish-Roleplay-Llama-3.1-8B-i1-GGUF + description: | + A DPO-tuned Llama-3.1 to behave more "humanish", i.e., avoiding all the AI assistant slop. It also works for role-play (RP). To achieve this, the model was fine-tuned over a series of datasets: + General conversations from Claude Opus, from Undi95/Meta-Llama-3.1-8B-Claude + Undi95/Weyaxi-humanish-dpo-project-noemoji, to make the model react as a human, rejecting assistant-like or too neutral responses. + ResplendentAI/NSFW_RP_Format_DPO, to steer the model towards using the *action* format in RP settings. Works best if in the first message you also use this format naturally (see example) + overrides: + parameters: + model: Humanish-Roleplay-Llama-3.1-8B.i1-Q4_K_M.gguf + files: + - filename: Humanish-Roleplay-Llama-3.1-8B.i1-Q4_K_M.gguf + sha256: 18cf753684e5226b51f3defc708852ca4924f50dc8bc31c9a7d0a036a477b7a7 + uri: huggingface://mradermacher/Humanish-Roleplay-Llama-3.1-8B-i1-GGUF/Humanish-Roleplay-Llama-3.1-8B.i1-Q4_K_M.gguf ## Uncensored models - !!merge <<: *llama31 name: "darkidol-llama-3.1-8b-instruct-1.0-uncensored-i1"