--- license: llama3 --- **Exllamav2** quant (**exl2** / **3.5 bpw**) made with ExLlamaV2 v0.1.4 Other EXL2 quants: | **Quant** | **Model Size** | **lm_head** | | ----- | ---------- | ------- | |
**[2.2](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-2_2bpw_exl2)**
|
3250 MB
|
6
| |
**[2.5](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-2_5bpw_exl2)**
|
3479 MB
|
6
| |
**[3.0](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-3_0bpw_exl2)**
|
3895 MB
|
6
| |
**[3.5](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-3_5bpw_exl2)**
|
4308 MB
|
6
| |
**[3.75](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-3_75bpw_exl2)**
|
4514 MB
|
6
| |
**[4.0](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-4_0bpw_exl2)**
|
4727 MB
|
6
| |
**[4.25](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-4_25bpw_exl2)**
|
4934 MB
|
6
| |
**[5.0](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-5_0bpw_exl2)**
|
5559 MB
|
6
| |
**[6.0](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-6_0bpw_exl2)**
|
6496 MB
|
8
| |
**[6.5](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-6_5bpw_exl2)**
|
6902 MB
|
8
| |
**[8.0](https://huggingface.co/Zoyd/AwanLLM_Awanllm-Llama-3-8B-Cumulus-v1.0-8_0bpw_exl2)**
|
8126 MB
|
8
| 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 is release v1.0 of Awanllm Cumulus series of models that aim to be uncensored and have zero refusals and zero warnings. This model should be good for general use cases as the OG Llama 3 8B model but it should be especially better for story writing or RP use cases. It is the most uncensored yet, thanks to using https://huggingface.co/failspy/Meta-Llama-3-8B-Instruct-abliterated-v3 as the base model. In terms of reasoning and intelligence, this model is probably a bit worse than the OG Meta Llama 3 8B Instruct because of the decensoring. However we believe it is worth it for the decensoring, as even with jailbreak prompts Llama 3 8B Instruct will never get remotely close to this model. We have it up on our API service at https://awanllm.com for anyone to try! Best practices: - Be precise and explain what you want the model to do. It has less base "personality" than the OG model but it will act however you tell it to. - 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 4 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/AwanLLM/Awanllm-Llama-3-8B-Cumulus-v1.0 GGUF: https://huggingface.co/AwanLLM/Awanllm-Llama-3-8B-Cumulus-v1.0-GGUF