Hugging Face's logo

CrucibleLab
/
L3.3-Dark-Ages-70b-v0.1

Safetensors
llama
Model card Files
xet
Community

L3.3-Dark-Ages-70b

Gothic Alchemist Mascot
Created by Crucible Lab SteelSkull → Support on Ko-fi
Tarek → Support on Ko-fi

Model Information

CrucibleLab-TG/L3.3-NS-Dark-Ages-70b-v0.1

Llama 3.3 Multi-Stage Merge 70b Parameters

Model Composition

Model Creation Process

L3.3-Dark-Ages-70b is the result of a multi-stage merging process conducted by Crucible Lab, a collaboration between Steelskull and Tarek. The goal was to create a powerful and creative roleplaying model with a unique, dark fantasy flavor.

Stage 1: The Cognitive Foundation

The process began by creating a cognitive and tool-use focused base model, Cogmo-70b. This was achieved through a `model_stock` merge of several models known for their reasoning and instruction-following capabilities, built upon a `Nemotron-lorablated` base.

Stage 2: Weaving the Narrative

Two distinct models were then created from the Cogmo base. Dark-Prose-70b was merged using `SCE` to enhance its creative writing and prose style, incorporating models like ArliAI's RPMax and Anubis. Simultaneously, Negative-RP-70b was created using a `della_linear` merge to integrate various roleplaying and narrative models, including Wayfarer and Vulpecula, while using Negative_LLAMA to maintain balance.

Final Stage: The Dark Ages

The final model, Dark-Ages-70b, was created by merging Dark-Prose and Negative-RP using the `nuslerp` method. This final step combines the rich, creative prose with the strong, negative-prompt-influenced roleplaying capabilities.

Future Plans

Model Development Progress

Stage 1: Base Model Creation
  • Model Curation
  • Model Merge Base
  • Model Base Testing
  • Repeat Until Happy
Stage 2: Model 1 Creation
  • Model Curation
  • Model Merge Model 1
  • Model Merge Model 1 Test
  • Repeat Until Happy
Stage 3: Model 2 Creation
  • Model Curation
  • Model Merge Model 2
  • Model Merge Model 2 Test
  • Repeat Until Happy
Stage 4: Finalization
  • Final Merge
  • Final Merge Test
  • Repeat Until Happy
Stage 5: Release & Iteration
  • Primary Release
  • User Feedback
  • Finetune Model on Feedback
  • Repeat Until Happy

This initial release (v0.1) is an un-finetuned version of the final merge. Crucible Lab plans to further finetune the model to enhance its performance and solidify its unique characteristics.

Recommended Sampler Settings

Static Temperature: 1.0 - 1.5
Min P: 0.02
DRY:
- Multiplier: 0.8
- Base: 1.74
- Length: 4-6

Good Starting Templates & Prompts

Hamon v1 → by @Steel > Big-picture storytelling guide with world-building focus, set dialogue/narration split, and general writing rules.
Shingane v1 → by @Steel > Simplified sysprompt based on Hamon.
Kesshin v1 → by @Steel > A Hamon rethink using a Character-focused sys prompt that tracks what characters know and how they learn things, with strict interaction rules.
Kamae TTRPG v1 → by @Steel > TTRPG Game Master framework emphasizing player agency, world consistency, and adaptive session management with mechanical integration.
Kamae lite v1 → by @Steel > Simplified sysprompt based on Kamae.

Support & Community:

Join Discord
Downloads last month
62
Safetensors
Model size
70.6B params
Tensor type
BF16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for CrucibleLab/L3.3-Dark-Ages-70b-v0.1

Base model

meta-llama/Llama-3.1-70B
Finetuned
meta-llama/Llama-3.3-70B-Instruct
Finetuned
(177)
this model
Quantizations
2 models