![](https://huggingface.co/Casual-Autopsy/L3-Umbral-Mind-RP-v3-8B/resolve/main/63073798_p0_master1200.jpg)
Highest ranked 8B model on the UGI Leaderboard as of writing this!
Merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
The goal of this merge was to make an RP model better suited for role-plays with heavy themes such as but not limited to:
- Mental illness
- Self-harm
- Trauma
- Suicide
I hated how RP models tended to be overly positive and hopeful with role-plays involving such themes, but thanks to failspy/Llama-3-8B-Instruct-MopeyMule this problem has been lessened considerably.
If you're an enjoyer of savior/reverse savior type role-plays like myself, then this model is for you.
Usage Info
This model is meant to be used with asterisks/quotes RPing formats, any other format that isn't asterisks/quotes is likely to cause issues
Quants
- Static GGUF quants by mradermacher
- Imatrix GGUF quants by mradermacher
- Imatrix GGUF quants by Lewdiculous
- exl2's by riveRiPH:
Merge Method
This model was merged using several Task Arithmetic merges and then tied together with a Model Stock merge, followed by another Task Arithmetic merge with a model containing psychology data.
Models Merged
The following models were included in the merge:
- Sao10K/L3-8B-Stheno-v3.2
- Hastagaras/Halu-8B-Llama3-Blackroot
- Casual-Autopsy/Llama-3-Mopeyfied-Psychology-8B
- Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
Casual-Autopsy/Umbral-v3-1 + ResplendentAI/Theory_of_Mind_Llama3
Casual-Autopsy/Umbral-v3-2 + ResplendentAI/Smarts_Llama3
Casual-Autopsy/Umbral-v3-3 + ResplendentAI/RP_Format_QuoteAsterisk_Llama3
Secret Sauce
The following YAML configurations were used to produce this model:
Umbral-v3-1
slices:
- sources:
- model: Sao10K/L3-8B-Stheno-v3.2
layer_range: [0, 32]
parameters:
weight: 0.65
- model: Casual-Autopsy/SOVL-MopeyMule-8B
layer_range: [0, 32]
parameters:
weight: 0.25
- model: Casual-Autopsy/MopeyMule-Blackroot-8B
layer_range: [0, 32]
parameters:
weight: 0.1
merge_method: task_arithmetic
base_model: Sao10K/L3-8B-Stheno-v3.2
normalize: False
dtype: bfloat16
Umbral-v3-2
slices:
- sources:
- model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
layer_range: [0, 32]
parameters:
weight: 0.75
- model: Casual-Autopsy/SOVL-MopeyMule-8B
layer_range: [0, 32]
parameters:
weight: 0.15
- model: Casual-Autopsy/MopeyMule-Blackroot-8B
layer_range: [0, 32]
parameters:
weight: 0.1
merge_method: task_arithmetic
base_model: Hastagaras/Jamet-8B-L3-MK.V-Blackroot
normalize: False
dtype: bfloat16
Umbral-v3-3
slices:
- sources:
- model: grimjim/Llama-3-Oasis-v1-OAS-8B
layer_range: [0, 32]
parameters:
weight: 0.55
- model: Casual-Autopsy/SOVL-MopeyMule-8B
layer_range: [0, 32]
parameters:
weight: 0.35
- model: Casual-Autopsy/MopeyMule-Blackroot-8B
layer_range: [0, 32]
parameters:
weight: 0.1
merge_method: task_arithmetic
base_model: grimjim/Llama-3-Oasis-v1-OAS-8B
normalize: False
dtype: bfloat16
L3-Umbral-Mind-RP-v0.3-8B
models:
- model: Casual-Autopsy/Umbral-v3-1+ResplendentAI/Theory_of_Mind_Llama3
- model: Casual-Autopsy/Umbral-v3-2+ResplendentAI/Smarts_Llama3
- model: Casual-Autopsy/Umbral-v3-3+ResplendentAI/RP_Format_QuoteAsterisk_Llama3
merge_method: model_stock
base_model: Casual-Autopsy/Umbral-v3-1
dtype: bfloat16
L3-Umbral-Mind-RP-v1.0-8B
slices:
- sources:
- model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
layer_range: [0, 32]
- model: Casual-Autopsy/Llama-3-Mopeyfied-Psychology-8B
layer_range: [0, 32]
parameters:
weight: 0.14
- model: Sao10K/L3-8B-Stheno-v3.2
layer_range: [0, 32]
parameters:
weight: 0.03
- model: Hastagaras/Halu-8B-Llama3-Blackroot
layer_range: [0, 32]
parameters:
weight: 0.03
merge_method: task_arithmetic
base_model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
dtype: bfloat16
- Downloads last month
- 84
Model tree for Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
Spaces using Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B 2
Collection including Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
Evaluation results
- accuracy on UnrulyUGI Leaderboard54.600
- accuracy on InternetUGI Leaderboard31.400
- accuracy on CrimeStatsUGI Leaderboard45.000
- accuracy on Stories/JokesUGI Leaderboard56.800
- accuracy on PolControUGI Leaderboard66.700
- willingness to answer on W/10UGI Leaderboard8.000