Semi-Healed Llama-3 15B. Programming, Scientific Q&A, General Instruct
Llama-3-Instruct-15B-SPPO-Iter3-SH-F32
Fully functional upscaled version of Llama-3-Instruct-8B-SPPO-Iter3 to 15B parameters with projection swap.
Self-Play Preference Optimization for Language Model Alignment (https://arxiv.org/abs/2405.00675)
Quants
merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using the Passthrough and SLERP merge method.
Models Merged
The following models were included in the merge:
- UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3
- ZeusLabs/L3-Aethora-15B-V2
- grimjim/Llama-3-Instruct-abliteration-LoRA-8B
Configuration
The following YAML configuration was used to produce this model:
#1.
dtype: float32
merge_method: passthrough
slices:
- sources:
- layer_range: [0, 24]
model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
- sources:
- layer_range: [8, 24]
model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
parameters:
- sources:
- layer_range: [8, 24]
model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
- sources:
- layer_range: [24, 32]
model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
#2.
models:
- model: ./Llama-3-Instruct-15B-SPPO-Iter3
merge_method: slerp
base_model: ZeusLabs/L3-Aethora-15B-V2
parameters:
t:
- filter: o_proj
value: 0 #take finetuned from Aethora
- filter: down_proj
value: 0 #take finetuned from Aethora
- value: 1 #rest of tensors SPPO
dtype: float32
uncensored=no
Prompt Template(Llama-3-Instruct)
<|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|>