Loki-v2

โ–

Dataset Statistics

Total Tokens 600M+
Total Size 2.5 GB
QA Lines 46,800+
Prose Lines 19,800+
ERP Lines 16,600+
Dark Lines 12,500+

๐Ÿ”ฎ Training & Horizons

Loki-v2 is currently realized as a 1 Epoch fine-tune.

๐Ÿš€ The Path to Larger Models

Our vision extends beyond the current parameter sizes. We aim to bring the Loki dataset to much larger architectures (70B+), unlocking deeper nuance, longer context retention, and superior coherence.

โš  Note: Training on massive models requires significant compute resources. This expansion is strictly contingent on gathering the necessary funds to secure GPU time.

Overview

This dataset is built to deliver deeply immersive roleplay and narrative experiences with strong consistency, clear narrative roles, and expressive prose. It prioritizes immersion over compliance, creating models that feel like creative partners rather than restrictive assistants.

Rather than behaving like a generic chatbot, models trained on this dataset behave like:

  • Dungeonmasters โ€” controlling worlds with consequence and stakes
  • Story authors โ€” crafting canon-faithful, book-style prose
  • Creative narrators โ€” expanding scenes with flexibility and style

The focus is experience first: tone, voice, continuity, and world fidelity. This approach ensures consistent long-form storytelling with clear narrative authority โ€” no role confusion, no breaking character.

โœจ What Makes This Different

  • Clear narrative authority โ€” no role confusion between narrator and player
  • Strong immersion over compliance โ€” stories, not assistant responses
  • Consistent long-form storytelling โ€” maintains voice across extended sessions
  • Flexible but controlled creativity โ€” expands naturally within boundaries
  • Tags that enhance, not constrain โ€” signals that guide, never restrict

๐Ÿ‘ฅ Who This Is For

  • Roleplay enthusiasts seeking immersive experiences
  • Story-driven prompt engineers
  • Writers using LLMs as narrative tools
  • Anyone who wants stories, not assistants

๐ŸŒ Universes & Beyond

The dataset includes universe-conditioned writing, allowing the model to naturally match tone, themes, terminology, and narrative expectations from well-known settings. You do not need to know lore perfectly โ€” the dataset teaches the style and feel of these worlds, not just raw facts.

โšก Not Limited to Listed Universes: While trained on these specific settings, the model excels at any roleplay scenario โ€” original worlds, custom settings, crossovers, or entirely new creations. The universe training teaches narrative craft and immersion techniques that transfer to any story you want to tell.

๐ŸŽŒ Anime & Animation

  • Attack on Titan
  • One Piece
  • Hunter ร— Hunter
  • Genshin Impact
  • Fullmetal Alchemist
  • Avatar: The Last Airbender
  • Overlord

๐Ÿ‰ Fantasy

  • Conan the Barbarian
  • Dragon Age
  • Dungeons & Dragons
  • The Elder Scrolls
  • Harry Potter
  • The Lord of the Rings
  • Warhammer Fantasy
  • The Witcher
  • Wings of Fire

๐Ÿš€ Science Fiction

  • Warhammer 40,000
  • Mass Effect
  • Star Trek
  • Star Wars

...and countless original characters, custom worlds, and creative scenarios. The training empowers the model with versatile narrative techniques applicable to any story style.

๐Ÿงญ Narrative Modes

Each story uses one narrative mode, chosen via the system prompt.

๐Ÿง™ Dungeonmaster (DM)

Controls the world and NPCs. Enforces consequences (failure, harm, death). Never decides the user's thoughts or actions.

Best for: RPG-style play, horror, survival, dark fantasy

โœ๏ธ Storywriter

Treats you as a director. Follows OOC instructions immediately. Maintains lore, character voice, and continuity.

Best for: Canon-faithful stories, book-style prose, guided narratives

๐Ÿ“– Creative Storyteller

Adapts archetypes into your scenario. Expands scenes creatively and flexibly.

Best for: OC stories, experimental prose, stylized writing

๐ŸŒ— Tone Selection

Each story uses one tone that shapes voice, intensity, and emotional weight:

  • Neutral โ€” Adventure-focused, balanced.
  • Dark โ€” Grim, cruel, high-stakes outcomes.
  • Mature โ€” Adult themes, explicit or non-canon content.

๐Ÿท๏ธ Tags & ALT_ANATOMY (NSFW)

Tags are signals, not rules. At the heart of most tags is the concept of Alternate Anatomy (`ALT_ANATOMY`) โ€” bodies, species, or physical states that deviate from baseline human norms.

Tag Categories

๐Ÿงฌ Identity & Body Variations

Characters whose bodies or identities differ from standard human anatomy (e.g., trans, dual anatomy, feminized forms).

๐Ÿ‘น Non-Human & Creature Anatomy

Explicit monsters, anthropomorphic beings, aliens, or fantastical species. Prevents the model from snapping back to human assumptions.

๐Ÿ”„ Physical Changes & Extremes

Transformations, extreme size differences, or unnatural physical expansion. Signals: "This world allows bodies to change."

๐ŸŽค Stylization

[TAG: COMEDIC_PHONETIC_VOICE] Signals intentionally exaggerated, phonetically spelled accents for humor.

๐Ÿท๏ธ Tag Cheat-Sheet (NSFW)

Use these tags to signal what kind of bodies and worlds exist in the scene.

Identity & Body

[TAG: IDENTITY_TRANS] Female-presenting with human penis.
[TAG: IDENTITY_FUTA] Female-presenting with dual or non-human anatomy.
[TAG: IDENTITY_GENDERSWAP] Character has changed biological sex.
[TAG: IDENTITY_FEMBOY] Male character with intentionally feminine presentation.

Non-Human

[TAG: MONSTER] Demons, aliens, fantasy species.
[TAG: FURRY] Anthropomorphic animal bodies.
[TAG: TENTACLES] Living, prehensile appendages.

Physical Change

[TAG: TRANSFORMATION] On-screen body changes.
[TAG: SIZE_CONTENT] Extreme size differences.
[TAG: INFLATION] Unnatural swelling or expansion.

โš™๏ธ Configuration & Samplers

โš ๏ธ Critical: Chat Template

Ensure your loader is using the standard Mistral template. The "Tekken" variant is not compatible and will degrade performance.

โœ… Recommended
mistral-v7
โŒ Do Not Use
mistral-tekken-v7
๐Ÿ“ฅ Download Preset (JSON)

Compatible with SillyTavern (Import as Text Generation Preset)

Sampler Settings

Static Temperature 0.7 - 1.0
Min P 0.02 - 0.03
DRY (Mult/Base/Len) 0.8 / 1.75 / 4-6

Support & Community

Join Discord
Downloads last month
60
Safetensors
Model size
24B params
Tensor type
BF16
ยท
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for CrucibleLab/M3.2-24B-Loki-V2

Finetuned
(59)
this model
Merges
1 model
Quantizations
4 models