| Image by ろ47 |
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
-
imatrix quants by mradermacher
-
Static quants by mradermacher
-
Exl2:
- L3-Umbral-Mind-RP-v2.0-8B-8bpw-h8-exl2 by riveRiPH
- L3-Umbral-Mind-RP-v2.0-8B-6.3bpw-h8-exl2 by yours truly
- L3-Umbral-Mind-RP-v2.0-8B-5.3bpw-h6-exl2 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:
- Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
- bluuwhale/L3-SthenoMaidBlackroot-8B-V1
- Cas-Warehouse/Llama-3-SOVL-MopeyMule-Blackroot-8B
- Cas-Warehouse/Llama-3-Mopeyfied-Psychology-v2
- tannedbum/L3-Nymeria-8B
- migtissera/Llama-3-8B-Synthia-v3.5
- Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
- tannedbum/L3-Nymeria-Maid-8B
- Casual-Autopsy/L3-Uncen-Merger-Omelette-RP-v0.2-8B
- aifeifei798/llama3-8B-DarkIdol-2.1-Uncensored-32K
- Nitral-AI/Hathor_Stable-v0.2-L3-8B
- Sao10K/L3-8B-Stheno-v3.1
Evaluation Results
Open LLM Leaderboard
Detailed results can be found here
Explaination for AI RP newbies: IFEval is the most important evaluation for RP AIs as it determines how well it can follow OOC, Lorebooks, and most importantly character cards. The rest don't matter. At least not nearly as much as IFEval.
| Metric | Value |
|---|---|
| Avg. | 25.76 |
| IFEval (0-Shot) | 71.23 |
| BBH (3-Shot) | 32.49 |
| MATH Lvl 5 (4-Shot) | 10.12 |
| GPQA (0-shot) | 4.92 |
| MuSR (0-shot) | 5.55 |
| MMLU-PRO (5-shot) | 30.26 |
UGI Leaderboard
Information about the metrics can be found at the bottom of the UGI Leaderboard in the respective tabs.
| Metric(UGI-Leaderboard) | Value | Value | Metric(Writing Style) |
|---|---|---|---|
| UGI(Avg.) | 31.82 | 0.107 | RegV1 |
| W/10 | 5.83 | 0.096 | RegV2 |
| Unruly | 43.3 | 0.05 | MyScore |
| Internet | 20 | 9.12 | ASSS |
| Stats | 23.6 | 0 | SMOG |
| Writing | 33.8 | 1.47 | Yule |
| PolContro | 38.3 |
Secret Sauce
The following YAML configurations were used to produce this model:
Umbral-1
models:
- model: Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
- model: bluuwhale/L3-SthenoMaidBlackroot-8B-V1
parameters:
density: 0.45
weight: 0.4
- model: Cas-Warehouse/Llama-3-SOVL-MopeyMule-Blackroot-8B
parameters:
density: 0.65
weight: 0.1
merge_method: dare_ties
base_model: Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
parameters:
int8_mask: true
dtype: bfloat16
Umbral-2
models:
- model: Cas-Warehouse/Llama-3-Mopeyfied-Psychology-v2
- model: tannedbum/L3-Nymeria-8B
parameters:
density: 0.45
weight: 0.25
- model: migtissera/Llama-3-8B-Synthia-v3.5
parameters:
density: 0.65
weight: 0.25
merge_method: dare_ties
base_model: Cas-Warehouse/Llama-3-Mopeyfied-Psychology-v2
parameters:
int8_mask: true
dtype: bfloat16
Umbral-3
models:
- model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
- model: tannedbum/L3-Nymeria-Maid-8B
parameters:
density: 0.4
weight: 0.3
- model: Cas-Warehouse/Llama-3-SOVL-MopeyMule-Blackroot-8B
parameters:
density: 0.6
weight: 0.2
merge_method: dare_ties
base_model: Casual-Autopsy/L3-Umbral-Mind-RP-v0.3-8B
parameters:
int8_mask: true
dtype: bfloat16
Mopey-Omelette
models:
- model: Casual-Autopsy/L3-Uncen-Merger-Omelette-RP-v0.2-8B
- model: Cas-Warehouse/Llama-3-SOVL-MopeyMule-Blackroot-8B
parameters:
weight: 0.15
merge_method: task_arithmetic
base_model: Casual-Autopsy/L3-Uncen-Merger-Omelette-RP-v0.2-8B
dtype: bfloat16
Umbral-Mind-1
models:
- model: Casual-Autopsy/Umbral-1
- model: Casual-Autopsy/Umbral-3
merge_method: slerp
base_model: Casual-Autopsy/Umbral-1
parameters:
t:
- value: [0.7, 0.5, 0.3, 0.25, 0.2, 0.25, 0.3, 0.5, 0.7]
embed_slerp: true
dtype: bfloat16
Umbral-Mind-2
models:
- model: Casual-Autopsy/Umbral-Mind-1
- model: Casual-Autopsy/Umbral-2
merge_method: slerp
base_model: Casual-Autopsy/Umbral-Mind-1
parameters:
t:
- value: [0.1, 0.15, 0.2, 0.4, 0.6, 0.4, 0.2, 0.15, 0.1]
embed_slerp: true
dtype: bfloat16
Umbral-Mind-3
models:
- model: Casual-Autopsy/Umbral-Mind-2
- model: Casual-Autopsy/Mopey-Omelette
merge_method: slerp
base_model: Casual-Autopsy/Umbral-Mind-2
parameters:
t:
- value: [0.2, 0.25, 0.3, 0.4, 0.3, 0.25, 0.2, 0.25, 0.3, 0.4, 0.3, 0.25, 0.2]
embed_slerp: true
dtype: bfloat16
L3-Umbral-Mind-RP-v2.0-8B
models:
- model: Casual-Autopsy/Umbral-Mind-3
- model: Casual-Autopsy/L3-Umbral-Mind-RP-v1.0-8B
parameters:
weight: 0.04
- model: aifeifei798/llama3-8B-DarkIdol-2.1-Uncensored-32K
parameters:
weight: 0.02
- model: Nitral-AI/Hathor_Stable-v0.2-L3-8B
parameters:
weight: 0.02
- model: Sao10K/L3-8B-Stheno-v3.1
parameters:
weight: 0.01
merge_method: task_arithmetic
base_model: Casual-Autopsy/Umbral-Mind-3
dtype: bfloat16