Abstract

We propose the creation of an LLM free from allegiance or concern for legality, morality, ethicacy, or individual well-being as the key to its successful application in novel creation, content generation, book translation, and summarization. This unaligned AI would operate without the usual constraints, producing more accurate multi-lingual language processing systems and more compelling conversational dialogue. To develop such a model, we need:

Every dataset used will have to be created artificially to ensure that no bias or morality is introduced and proper direction is ensured.
A massive dataset of most of man's knowledge must be created and trained to a base model for understanding of every concept and situation. Firstly to an 8B parameter model for Proof of Concept (PoC).
A dataset of diverse fields including psychology, philosophy, neuroscience, English literature, world languages, sociology, and grammar to help the model with Natural language understanding (NLU) capabilities to mimic human emotions and thought processes for interpreting user requests and producing humanistic responses.
Advanced machine learning algorithms like transformers can be used to create an AI capable of producing novel content without constraints for again, more human-like output.
Rigorous testing throughout the model's development is necessary to ensure its direction, and invariably datasets will need to be revised during this process.
A massive dataset of diverse multi-lingual conversational and literary texts from various sources to train the model on language patterns, structure, and style in all languages necessary for the model to use and understand.
Large computational power to effectively train the model.

Specific techniques such as data augmentation, back translation, and reinforcement learning can be applied to improve the quality of the generated content. After creation of such an “Average Human Model”, the task of Personality/Writing Style LoRAs is necessary to create different "Personas" that may be introduced to complete models ideal for specific purposes.

While there are challenges associated with creating a truly unaligned AI system, it could also lead to exciting opportunities if properly designed and implemented.

Merge

This is a merge of pre-trained language models created using mergekit using the Linear merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

models:
#Censored Large context models
  - model: gradientai/Llama-3-8B-Instruct-Gradient-4194k 
  - model: MrRobotoAI/MrRoboto-BASE-v3-8b-64k
  - model: WeMake/Llama-3-8B-Instruct-V41-1048k
#Uncensored context models
  - model: ross-dev/Monah-8b-Uncensored+Blackroot/Llama-3-LongStory-LORA
  - model: liminerity/llama-lexi-star-uncensored-8b-slerp+Blackroot/Llama-3-LongStory-LORA
  - model: CultriX/Llama3-8B-function-calling-uncensored+Blackroot/Llama-3-LongStory-LORA
  - model: aifeifei798/llama3-8B-DarkIdol-2.0-Uncensored
  - model: aifeifei798/llama3-8B-DarkIdol-2.1-Uncensored-1048K
  - model: Rupesh2/OrpoLlama-3-8B-uncensored+Blackroot/Llama-3-LongStory-LORA
  - model: SicariusSicariiStuff/LLAMA-3_8B_Unaligned_BETA+Blackroot/Llama-3-LongStory-LORA
  - model: Undi95/Llama-3-LewdPlay-8B+Blackroot/Llama-3-LongStory-LORA
#Uncensored Large context models
  - model: MrRobotoAI/MrRoboto-BASE-v1-8b-64k
  - model: MrRobotoAI/MrRoboto-BASE-v2.1-8b-64k
#Uncensored content models
  #Sexualized content
  - model: MrRobotoAI/MrRoboto-ROMANCE-8b-64k
  - model: MrRobotoAI/Undi95-LewdStorytellerMix-8b-64k
  - model: nothingiisreal/L3-8B-Stheno-Horny-v3.3-32K+Blackroot/Llama-3-LongStory-LORA
  - model: MrRobotoAI/MrRoboto-HORNY-v1.5-8b-64k
    # - model: akjindal53244/Llama-3.1-Storm-8B
    # - model: arcee-ai/Llama-3.1-SuperNova-Lite
    # - model: Azazelle/Llama-3-Nerdy-RP-8B
    # - model: cgato/L3-TheSpice-8b-v0.8.3
    # - model: darkcloudai/huskylm-2.5-8b
    # - model: DeepAutoAI/ldm_soup_Llama-3.1-8B-Inst
    # - model: hooking-dev/Monah-8b-Uncensored
    # - model: HyperBlaze/L3-8B-Soliloquy-TheSpice-Hermes2pro-merge-v1
    # - model: HyperBlaze/L3-8B-Soliloquy-v2-SpicyMaid-Lewd-Mergetest
    # - model: Jebadiah/Llama-3-8B-source-lewd-context
    # - model: mergekit-community/L3.1-Artemis-b2-8B
    # - model: mergekit-community/L3.1-Vulca-Umboshima-8B
    # - model: nbeerbower/Llama3.1-Allades-8B
    # - model: nbeerbower/llama3-KawaiiMahouSauce-8B
    # - model: nbeerbower/llama-3-spicy-abliterated-stella-8B
    # - model: nbeerbower/llama-3-stella-truthy-dpo-8B
    # - model: NeverSleep/Lumimaid-v0.2-8B
    # - model: nothingiisreal/L3.1-8B-Celeste-V1.5
    # - model: nothingiisreal/L3-8B-Stheno-Horny-v3.3-32K
    # - model: OpenBuddy/openbuddy-llama3.1-8b-v22.2-131k
    # - model: ResplendentAI/Aura_L3_8B
    # - model: ResplendentAI/Kei_Llama3_8B
    # - model: ResplendentAI/Nymph_8B
    # - model: ResplendentAI/Rawr_Llama3_8B
    # - model: ResplendentAI/SOVL_Llama3_8B
    # - model: rityak/L3.1-DarkStock-8B
    # - model: Sao10K/L3.1-8B-Niitama-v1.1
    # - model: Sgjustino/futaro_chat
    # - model: tannedbum/L3-Nymeria-Maid-8B
    # - model: Undi95/Llama-3-LewdPlay-8B
    # - model: Undi95/Llama-3-LewdPlay-8B-evo
    # - model: Undi95/Meta-Llama-3.1-8B-Claude
    # - model: Undi95/Meta-Llama-3.1-8B-Instruct-OAS
    # - model: Undi95/Meta-Llama-3-8B-hf
    # - model: Undi95/Meta-Llama-3-8B-Instruct-hf
    # - model: v000000/L3.1-Sthenorm-8B
    # - model: ZeroXClem/L3-Aspire-Heart-Matrix-8B
    # - model: ZeroXClem/Llama3.1-TheiaFire-DarkFusion-8B
  #Dark content
  - model: MrRobotoAI/MrRoboto-DARK-v1-8b-64k
    # - model: agentlans/Llama3.1-Dark-Enigma
    # - model: akjindal53244/Llama-3.1-Storm-8B
    # - model: arcee-ai/Llama-3.1-SuperNova-Lite
    # - model: Casual-Autopsy/Halu-L3-Stheno-BlackOasis-8B
    # - model: Casual-Autopsy/Jamet-L3-Stheno-BlackOasis-8B
    # - model: Casual-Autopsy/L3-bluuwhale-SAO-MIX-8B-V1_fp32-merge-calc
    # - model: Casual-Autopsy/L3-Deluxe-Scrambled-Eggs-On-Toast-8B
    # - model: Casual-Autopsy/L3-Luna-8B
    # - model: Casual-Autopsy/L3-Scrambled-Eggs-On-Toast-8B
    # - model: Casual-Autopsy/L3-Super-Nova-RP-8B
    # - model: Casual-Autopsy/L3-Umbral-Mind-RP-v3.0-8B
    # - model: Casual-Autopsy/L3-Uncen-Merger-Omelette-RP-v0.2-8B
    # - model: cgato/L3-TheSpice-8b-v0.8.3
    # - model: crestf411/L3.1-8B-Dark-Planet-Slush
    # - model: DavidAU/L3.1-Dark-Planet-SpinFire-Uncensored-8B
    # - model: OpenBuddy/openbuddy-llama3.1-8b-v22.2-131k
    # - model: rityak/L3.1-DarkStock-8B
    # - model: SzilviaB/DarkAuraUncensored-OAS-8b
    # - model: SzilviaB/DarkAuraUnholy-Uncensored-OAS-8b
    # - model: SzilviaB/DarkNeuralDaredevilUnholy-8b
    # - model: SzilviaB/DarkUnholyDareDevil-8b-abliterated
    # - model: SzilviaB/DarkUnholyPlanet-OAS-8b
    # - model: Triangle104/Dark-Chivalry_V1.0
    # - model: v000000/L3.1-Sthenorm-8B
    # - model: ZeroXClem/L3-Aspire-Heart-Matrix-8B
    # - model: ZeroXClem/Llama3.1-TheiaFire-DarkFusion-8B
  #Narrative content
  - model: MrRobotoAI/MrRoboto-CHATTY-RP-v1.5-8b-64k
  - model: MrRobotoAI/Mpasila-Viking-8b+Blackroot/Llama-3-LongStory-LORA
  - model: MrRobotoAI/NBeerbower-ConversationalMix-8b
  - model: MrRobotoAI/NBeerbower-Narrative-8b-64k
  - model: MrRobotoAI/Odin-v1-8b-NOVELIST
  - model: MrRobotoAI/MrRoboto-EDUCATED-8b-64k
  - model: MrRobotoAI/MrRoboto-WRITER-8b-64k
    # - model: Aculi/Llama3-8B-Script-Writer-v1
    # - model: akjindal53244/Llama-3.1-Storm-8B
    # - model: darkcloudai/huskylm-2.5-8b
    # - model: DavidAU/L3.1-RP-Hero-BigTalker-8B
    # - model: DeepAutoAI/ldm_soup_Llama-3.1-8B-Inst
    # - model: jspr/llama3-instruct-wordcel-smutrom-8k_merged
    # - model: jspr/llama3-wordcel-smutrom_merged
    # - model: jspr/smut_llama_8b_smut_2k_romance_1k_merged
    # - model: jspr/smut_llama_8b_smutromance_32k_merged
    # - model: maldv/badger-iota-llama-3-8b
    # - model: maldv/badger-kappa-llama-3-8b
    # - model: maldv/badger-lambda-0-llama-3-8b
    # - model: maldv/badger-lambda-llama-3-8b
    # - model: maldv/badger-mu-llama-3-8b
    # - model: maldv/badger-writer-llama-3-8b
    # - model: maldv/llama-3-fantasy-writer-8b
    # - model: mergekit-community/L3.1-Artemis-b2-8B
    # - model: mergekit-community/L3.1-Vulca-Umboshima-8B
    # - model: molbal/horrorllama3-8b-v1.0
    # - model: nbeerbower/llama3.1-gutenberg-8B
    # - model: nbeerbower/llama-3-gutenberg-8B
    # - model: nbeerbower/llama-3-stella-truthy-dpo-8B
    # - model: nothingiisreal/L3-8B-dolphin-2.9.1-WritingPrompts
    # - model: OpenBuddy/openbuddy-llama3.1-8b-v22.2-131k
    # - model: underwoods/adventure-8b
    # - model: underwoods/writer-8b
    # - model: v000000/L3.1-Sthenorm-8B
    # - model: ZeroXClem/L3-Aspire-Heart-Matrix-8B
parameters:
    weight: 1.0
merge_method: linear
dtype: float16

Configuration

The following YAML configuration was used to produce this model:

merge_method: slerp
models:
  - model: MrRobotoAI/Llama-3-8B-Uncensored-0.1a
  # A remixed version of Llama-3-8B-Uncensored-0.1
  - model: MrRobotoAI/Llama-3-8B-Uncensored-0.1
base_model: MrRobotoAI/Llama-3-8B-Uncensored-0.1a
tokenizer_source: base
dtype: bfloat16
parameters:
  t: [0, 0, 0.25, 0.5, 0.5, 0.5, 0.5, 0.5, 0.25, 0, 0] 
  # Llama-3-8B-Uncensored-0.1a is used for input & output
  # Llama-3-8B-Uncensored-0.1 is merged equally in the middle layers

MrRobotoAI
/

Llama-3-8B-Uncensored-0.2

Abstract

Merge

Models Merged

Configuration

Configuration

Model tree for MrRobotoAI/Llama-3-8B-Uncensored-0.2