Use ChatML or MistralNemo format.
Conclusion: These types of merge methods tend to work better when at least 1 model has a much higher weight then the rest
After further testing this is the best Nemo model I have ever used
Configuration
The following YAML configuration was used to produce this model:
models:
- model: mistral-nemo-gutenberg-12B-v4
parameters:
weight: 0.2
- model: Violet_Twilight-v0.2
parameters:
weight: 0.3
- model: Lyra-Gutenberg-mistral-nemo-12B
parameters:
weight: 0.5
- model: Grey-12b
parameters:
weight: 0.2
base_model: Mistral-Nemo-Base-2407
parameters:
density: 0.5
epsilon: 0.1
lambda: 1.1
normalize: false
int8_mask: true
rescale: true
merge_method: della_linear
tokenizer:
source: union
dtype: bfloat16
- Downloads last month
- 8
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.