7-9B Models
Collection
light model merges
•
8 items
•
Updated
RP model.
Poppy Sunspice with updated models, slerps, ties and loras. New gen test.
This has a tendency for endless generations. It likes a bit of penalty parameters if this occurs.
0.95 temp
80 top k
0.95 top p
0.1 min p
1.1 rep pen
This is a merge of pre-trained language models created using mergekit.
This model was merged using the SLERP merge method.
The following models were included in the merge:
The following YAML configuration was used to produce this model:
slices:
- sources:
- model: v000000/L3-8B-Poppy-Sunspice-experiment-c+Blackroot/Llama-3-8B-Abomination-LORA
layer_range: [0, 32]
- model: v000000/L3-8B-Poppy-Sunspice-experiment-c+ResplendentAI/BlueMoon_Llama3
layer_range: [0, 32]
merge_method: slerp
base_model: v000000/L3-8B-Poppy-Sunspice-experiment-c+Blackroot/Llama-3-8B-Abomination-LORA
parameters:
t:
- filter: self_attn
value: [0, 0.5, 0.3, 0.7, 1]
- filter: mlp
value: [1, 0.5, 0.7, 0.3, 0]
- value: 0.5
dtype: bfloat16
random_seed: 0
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>
{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
{output}<|eot_id|>