|
--- |
|
license: apache-2.0 |
|
datasets: |
|
- Open-Orca/OpenOrca |
|
- teknium/openhermes |
|
- cognitivecomputations/dolphin |
|
- jondurbin/airoboros-3.1 |
|
- unalignment/toxic-dpo-v0.1 |
|
- unalignment/spicy-3.1 |
|
language: |
|
- en |
|
--- |
|
|
|
 |
|
# The flower of Ares. |
|
## These are the GGUF files of the fine-tuned model. To be compiled with llama.cpp on oobabooga or VLLm. |
|
|
|
Fine-tuned on [mistralai/Mistral-7B-v0.1](https://huggingface.co/mistralai/Mistral-7B-v0.1)...[my team and I](https://huggingface.co/ConvexAI) reformatted many different datasets and included a small amount of private stuff to see how much we could improve mistral. |
|
|
|
I spoke to it personally for about an hour, and I believe we need to work on our format for the private dataset a bit more, but other than that, it turned out great. I will be uploading it to open llm evaluations, today. |
|
|
|
## Provided files |
|
|
|
| Name | Quant method | Bits | Size | Max RAM required | Use case | |
|
| ---- | ---- | ---- | ---- | ---- | ----- | |
|
| [Q2_K Tiny](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q2_k.gguf) | Q2_K | 2 | 2.7 GB| 4.7 GB | smallest, significant quality loss - not recommended for most purposes | |
|
| [Q3_K_M](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q3_k_m.gguf) | Q3_K_M | 3 | 3.52 GB| 5.52 GB | very small, high quality loss | |
|
| [Q4_0](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q4_0.gguf) | Q4_0 | 4 | 4.11 GB| 6.11 GB | legacy; small, very high quality loss - prefer using Q3_K_M | |
|
| [Q4_K_M](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q4_k_m.gguf) | Q4_K_M | 4 | 4.37 GB| 6.37 GB | medium, balanced quality - recommended | |
|
| [Q5_0](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q5_0.gguf) | Q5_0 | 5 | 5 GB| 7 GB | legacy; large, balanced quality | |
|
| [Q5_K_M](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q5_k_m.gguf) | Q5_K_M | 5 | 5.13 GB| 7.13 GB | large, balanced quality - recommended | |
|
| [Q6 XL](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q6_k.gguf) | Q6_K | 6 | 5.94 GB| 7.94 GB | very large, extremely low quality loss | |
|
| [Q8 XXL](https://huggingface.co/Kquant03/Medusa-7B-GGUF/blob/main/ggml-model-q8_0.gguf) | Q8_0 | 8 | 7.7 GB| 9.7 GB | very large, extremely low quality loss - not recommended | |
|
|
|
- Uses Mistral prompt template with chat-instruct. |
|
|
