"Llama-3-Instruct-8B-SPPO-Iter3" fully uncensored with 35% RP-Mix infused vector direction to gain some roleplay capabilities and prose while attempting to preserve the qualities of Meta's Llama-3-Instruct finetune.
Thank you mradermacher for the quants!
Quants
merge
This is a merge of pre-trained language models created using mergekit.
Merge Details
Merge Method
This model was merged using a multi-step merge method.
Models Merged
The following models were included in the merge:
- grimjim/Llama-3-Instruct-abliteration-LoRA-8B
- UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3
- NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
- maldv/llama-3-fantasy-writer-8b
- tokyotech-llm/Llama-3-Swallow-8B-v0.1
- Nitral-AI/Hathor_Respawn-L3-8B-v0.8
Configuration
The following YAML configuration was used to produce this model:
# Part 3, Apply abliteration (SwallowMaid-8B)
models:
- model: sppo-rpmix-part2+grimjim/Llama-3-Instruct-abliteration-LoRA-8B
parameters:
weight: 1.0
merge_method: linear
dtype: float32
# Part 2, infuse 35% swallow+rpmix to SPPO-Iter3 (sppo-rpmix-part2)
models:
- model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3
parameters:
weight: 1.0
- model: rpmix-part1
parameters:
weight: 0.35
merge_method: task_arithmetic
base_model: UCLA-AGI/Llama-3-Instruct-8B-SPPO-Iter3
parameters:
normalize: false
dtype: float32
# Part 1, linear merge rpmix (rpmix-part1)
models:
- model: Nitral-AI/Hathor_Respawn-L3-8B-v0.8
parameters:
weight: 0.6
- model: maldv/llama-3-fantasy-writer-8b
parameters:
weight: 0.1
- model: NeverSleep/Llama-3-Lumimaid-8B-v0.1-OAS
parameters:
weight: 0.4
- model: tokyotech-llm/Llama-3-Swallow-8B-v0.1
parameters:
weight: 0.15
merge_method: linear
dtype: float32
Prompt Template:
<|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|>