11-20B Models
Collection
Medium model merges
•
10 items
•
Updated
•
1
Temperature:
Mistral Nemo likes low temperature between 0.3-0.5
RP Model. Seems coherent and concise but also creative. Big merge using new DELLA technique.
Prompt Format: Seems best with "Mistral Instruct" but ChatML might also work.
[INST] System Message [/INST]
[INST] Name: Let's get started. Please respond based on the information and instructions provided above. [/INST]
<s>[INST] Name: What is your favourite condiment? [/INST]
AssistantName: Well, I'm quite partial to a good squeeze of fresh lemon juice. It adds just the right amount of zesty flavour to whatever I'm cooking up in the kitchen!</s>
[INST] Name: Do you have mayonnaise recipes? [/INST]
This is a merge of pre-trained language models created using mergekit.
This model was merged with a multi-step method using the DELLA, DELLA_LINEAR and SLERP merge algorithms.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
#Step 1 (Part1)
models:
- model: Sao10K/MN-12B-Lyra-v1
parameters:
weight: 0.15
density: 0.77
- model: shuttleai/shuttle-2.5-mini
parameters:
weight: 0.20
density: 0.78
- model: anthracite-org/magnum-12b-v2
parameters:
weight: 0.35
density: 0.85
- model: nothingiisreal/MN-12B-Celeste-V1.9
parameters:
weight: 0.55
density: 0.90
merge_method: della
base_model: Sao10K/MN-12B-Lyra-v1
parameters:
int8_mask: true
epsilon: 0.05
lambda: 1
dtype: bfloat16
#Step 2 (Part2)
models:
- model: BeaverAI/mistral-doryV2-12b
parameters:
weight: 0.10
density: 0.4
- model: unsloth/Mistral-Nemo-Instruct-2407
parameters:
weight: 0.20
density: 0.4
- model: UsernameJustAnother/Nemo-12B-Marlin-v5
parameters:
weight: 0.25
density: 0.5
- model: invisietch/Atlantis-v0.1-12B
parameters:
weight: 0.3
density: 0.5
- model: NeverSleep/Lumimaid-v0.2-12B
parameters:
weight: 0.4
density: 0.8
merge_method: della_linear
base_model: anthracite-org/magnum-12b-v2
parameters:
int8_mask: true
epsilon: 0.05
lambda: 1
dtype: bfloat16
#Step 3 (Estrella)
slices:
- sources:
- model: v000000/MN-12B-Part2
layer_range: [0, 40]
- model: v000000/MN-12B-Part1
layer_range: [0, 40]
merge_method: slerp
base_model: v000000/MN-12B-Part1
parameters: #smooth gradient prio part1
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 0.6, 0.1, 0.6, 0.3, 0.8, 0.5]
- filter: mlp
value: [0, 0.5, 0.4, 0.3, 0, 0.3, 0.4, 0.7, 0.2, 0.5]
- value: 0.5
dtype: bfloat16