GoldDiamondGold-70b
This is a merge of pre-trained language models created using mergekit.
Motivation
Inspired from the sapphire model by BruhzWater, the base for this is cogito.
Rest of the models is what I think would be good for a SCE attempt. Basically combining all the best bits in each model and see what I get out of this.
There's 3 interesting models that I highlighted in the previous merge that went into this too.
Vibes
Seems OK. I think it's better than the previous model. That felt super sloppy.
- It's pretty smart & seems to get the prompt/vibe format on the first try.
- The models seems to have a horny streak to it, but I can't put my finger on it.
- Comparing to
KaraKaraWitch/Llama-EveningMirai-3.3-70B, this model is smarter. - Can fall into the llama sloppiness if your previous responses are llama.
But I think that's a skill issue. - I think it's a bit too "assistant" heavy. Not 100% sure what to remove to fix it?
- Seems to have a bit too much of claudisms(?)
- A stickler for system prompts. Do ensure your system prompt is well written and doesn't conflict with what you want.
- ...Hard to write this model, I like it actually in general.
Merge Details
Merge Method
This model was merged using the SCE merge method using deepcogito/cogito-v2-preview-llama-70B as a base.
Models Merged
The following models were included in the merge:
- LatitudeGames/Wayfarer-Large-70B-Llama-3.3
- marcelbinz/Llama-3.1-Centaur-70B
- flammenai/Mahou-1.5-llama3.1-70B
- EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.0
- zerofata/L3.3-GeneticLemonade-Unleashed-v3-70B
- tdrussell/Llama-3-70B-Instruct-Storywriter
- Ppoyaa/MythoNemo-L3.1-70B-v1.0
- Blackroot/Mirai-3.0-70B
- Sao10K/70B-L3.3-mhnnn-x1
- TheDrummer/Fallen-Llama-3.3-70B-v1
- nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
- TheDrummer/Anubis-70B-v1.1
- Doctor-Shotgun/L3.3-70B-Magnum-Diamond
- Black-Ink-Guild/Pernicious_Prophecy_70B
- watt-ai/watt-tool-70B
- ReadyArt/Forgotten-Safeword-70B-v5.0
- nbeerbower/Llama3.1-Gutenberg-Doppel-70B
Configuration
The following YAML configuration was used to produce this model:
models:
# Mirai is Mirai.
- model: Blackroot/Mirai-3.0-70B
# Narration
- model: EVA-UNIT-01/EVA-LLaMA-3.33-70B-v0.0
# Claude 3 Sonnet/Opus prose style and quality
- model: Doctor-Shotgun/L3.3-70B-Magnum-Diamond
# "For the *Action*"?
- model: marcelbinz/Llama-3.1-Centaur-70B
# Better writing style, "creativity" shift. (fiction books)
- model: tdrussell/Llama-3-70B-Instruct-Storywriter
# Roleplaying and Story Writing
- model: Ppoyaa/MythoNemo-L3.1-70B-v1.0
# Sao10K
- model: Sao10K/70B-L3.3-mhnnn-x1
# Medical
- model: Black-Ink-Guild/Pernicious_Prophecy_70B
# Dialogue reinforcement
- model: LatitudeGames/Wayfarer-Large-70B-Llama-3.3
# Extra details
- model: TheDrummer/Anubis-70B-v1.1
# "Meanness"
- model: TheDrummer/Fallen-Llama-3.3-70B-v1
# Antique history
- model: nbeerbower/Llama3.1-Gutenberg-Doppel-70B
# Normalization?
- model: nvidia/Llama-3.1-Nemotron-70B-Instruct-HF
# Doomer tilt (From negative llama) + "Taboo"(?)
- model: ReadyArt/Forgotten-Safeword-70B-v5.0
# ERP/RP enhancement + Anime tilt
- model: zerofata/L3.3-GeneticLemonade-Unleashed-v3-70B
# Tool Calling
- model: watt-ai/watt-tool-70B
# Short, casual dialogue (Anime tilt)
- model: flammenai/Mahou-1.5-llama3.1-70B
merge_method: sce
base_model: deepcogito/cogito-v2-preview-llama-70B
select_topk: 0.33
parameters:
normalize: true
dtype: bfloat16
UGI-Bench
Within the top 10 (20/10/2025).
