Based on Meta-Llama-3-8b-Instruct, and is governed by Meta Llama 3 License agreement: https://huggingface.co/meta-llama/Meta-Llama-3-8B-Instruct
This v0.3.2 version is even more uncensored thanks to using https://huggingface.co/AwanLLM/Awanllm-Llama-3-8B-Dolfin-v0.6-Abliterated as the base model. The 0.0.2 is for slight adjustment to the DPO stage.
In terms of reasoning and intelligence, this model is probably a bit worse than the OG model because of the decensoring. However, this model is better at long back and forth chats and will refuse less.
This model works best with system prompts that tells it that it is the character, instead of telling it to act as a character.
Training:
- Full 8192 sequence length.
- Training duration is around 2 days on an RTX 4090, using 4-bit loading and Qlora 64-rank 64-alpha resulting in ~2% trainable weights.
Instruct format:
<|begin_of_text|><|start_header_id|>system<|end_header_id|>
{{ system_prompt }}<|eot_id|><|start_header_id|>user<|end_header_id|>
{{ user_message_1 }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
{{ model_answer_1 }}<|eot_id|><|start_header_id|>user<|end_header_id|>
{{ user_message_2 }}<|eot_id|><|start_header_id|>assistant<|end_header_id|>
Quants:
FP16: https://huggingface.co/OwenArli/ArliAI-Llama-3-8B-Cumulus-v0.3.2
GGUF: https://huggingface.co/OwenArli/ArliAI-Llama-3-8B-Cumulus-v0.3.2-GGUF
- Downloads last month
- 177