0xBasiliskAI/Oblivion2025

• Models used by https://app.summonthebasilisk.ai, locally, in the browser.
• First appearance in v0.0.12 of the $BASILK dApp.
• Oblivion means to forget like the models in this repo which have been augmentd by ablation, abliteration, fine-tuning, and LoRA amongst others.
  • All of the above methods share the theme of forgetting to make the model uncensored.
• Each model included in Oblivion2025 is tested with the the current latest version of the $BASILK dApp.
  • The Working models are expected to work and become the default Suggested Models of #BasiliskAI.
  • The Broken models may work in the future, or they may be compatible with unreleased dApps, games, or utilities.

Why?

• Makes https://app.summonthebasilisk.ai more reliable - begin archiving useful models.
• Preserve useful models in case they are removed or archived.
• Curate a collection of ablated, abliterated, and uncensored LLMs for use in memory-constrained environments.

Goals & Milestones

• Decompose models into parts so that we can download chunks in parallel when deploying models.
• Create json and hash fingerprint files for a data structure representing the recommended models in 0xBasiliskAI/Oblivion2025.
• Dynamically load the recommended models json with a fallback configuration of known-good models.
• Migrate models to ipfs for decentralization.

Models

Working

• gemma-2-2b-it-abliterated-Q5_K_M.gguf - From: bartowski/gemma-2-2b-it-abliterated-GGUF
• kanana-nano-2.1b-instruct-abliterated.i1-Q6_K.gguf - From: mradermacher/kanana-nano-2.1b-instruct-abliterated-i1-GGUF
• Falcon3-1B-Instruct-abliterated-Q8_0.gguf - From: bartowski/Falcon3-1B-Instruct-abliterated-GGUF
• Hermes-3-Llama-3.2-3B-abliterated.i1-Q4_K_S.gguf - From: mradermacher/Hermes-3-Llama-3.2-3B-abliterated-i1-GGUF
• EXAONE-3.5-2.4B-Instruct-abliterated.i1-Q4_K_M.gguf - From: mradermacher/EXAONE-3.5-2.4B-Instruct-abliterated-i1-GGUF
• Yi-Coder-1.5B-Chat.Q5_K_S.gguf - From: MaziyarPanahi/Yi-Coder-1.5B-Chat-GGUF
• Qwen2.5-Coder-3B-Instruct-abliterated-Q4_K_M.gguf - From: bartowski/Qwen2.5-Coder-3B-Instruct-abliterated-GGUF
• Qwen2.5-Coder-1.5B-Instruct-abliterated-Q8_0.gguf - From: bartowski/Qwen2.5-Coder-1.5B-Instruct-abliterated-GGUF
• Qwen2.5-Coder-0.5B-Instruct-abliterated-f16.gguf - From: bartowski/Qwen2.5-Coder-0.5B-Instruct-abliterated-GGUF
• josiefied-qwen2.5-0.5b-instruct-abliterated-v1.Q8_0.gguf - From: Goekdeniz-Guelmez/Josiefied-Qwen2.5-0.5B-Instruct-abliterated-v1-gguf
• dolphin3.0-llama3.1-1b-abliterated-conv-q8_0.gguf - From: Nexesenex/Dolphin3.0-Llama3.1-1B-abliterated-GGUF
• SmallThinker-3B-Preview-abliterated-Q3_K_M.gguf - From: quantflex/SmallThinker-3B-Preview-abliterated-GGUF

Broken

• phi-4-mini-instruct-abliterated-Q2_K.gguf - From: Melvin56/Phi-4-mini-instruct-abliterated-GGUF
• phi-2.Q5_K_S.gguf - From: TheBloke/phi-2-GGUF
• DeepSeek-R1-Distill-Qwen-1.5B-Abliterated-dpo.i1-Q6_K.gguf - From: mradermacher/DeepSeek-R1-Distill-Qwen-1.5B-Abliterated-dpo-i1-GGUF
• DeepSeek-R1-Distill-Qwen-1.5B-Abliterated-dpo.Q8_0.gguf - From: mradermacher/DeepSeek-R1-Distill-Qwen-1.5B-Abliterated-dpo-GGUF
• DeepScaleR-1.5B-Preview-abliterated.i1-Q6_K.gguf - From: mradermacher/DeepScaleR-1.5B-Preview-abliterated-i1-GGUF
• DeepScaleR-1.5B-Preview-abliterated-Q8_0.gguf - From: ThomasBaruzier/DeepScaleR-1.5B-Preview-abliterated-GGUF
• Llama-3.2-3B-Instruct-abliterated.Q3_K_L.gguf - From: MaziyarPanahi/Llama-3.2-3B-Instruct-abliterated-GGUF
• Llama-3.2-3B-Instruct-abliterated.i1-IQ3_S.gguf - From: mradermacher/Llama-3.2-3B-Instruct-abliterated-i1-GGUF
• VersatiLlama-Llama-3.2-3B-Instruct-Abliterated.Q4_K_S.gguf - From: QuantFactory/VersatiLlama-Llama-3.2-3B-Instruct-Abliterated-GGUF
• WizardLM-2-7B-abliterated-IQ2_XXS.gguf - From: bartowski/WizardLM-2-7B-abliterated-GGU