Update README.md
Browse files
README.md
CHANGED
|
@@ -14,6 +14,19 @@ tags:
|
|
| 14 |
|
| 15 |
I finally figured out how to quantize FrankenMoE properly, so prepare for a flood of GGUF models from me. This one is scripted to be into whatever you're planning to do to it.
|
| 16 |
Special thanks to [Cultrix](https://huggingface.co/CultriX) for the [base model](https://huggingface.co/CultriX/MistralTrix-v1).
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
# "[What is a Mixture of Experts (MoE)?](https://huggingface.co/blog/moe)"
|
| 18 |
### (from the MistralAI papers...click the quoted question above to navigate to it directly.)
|
| 19 |
|
|
|
|
| 14 |
|
| 15 |
I finally figured out how to quantize FrankenMoE properly, so prepare for a flood of GGUF models from me. This one is scripted to be into whatever you're planning to do to it.
|
| 16 |
Special thanks to [Cultrix](https://huggingface.co/CultriX) for the [base model](https://huggingface.co/CultriX/MistralTrix-v1).
|
| 17 |
+
|
| 18 |
+
## Provided files
|
| 19 |
+
|
| 20 |
+
| Name | Quant method | Bits | Size | Max RAM required | Use case |
|
| 21 |
+
| ---- | ---- | ---- | ---- | ---- | ----- |
|
| 22 |
+
| [Q2_K Tiny](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q2_k.gguf) | Q2_K | 2 | 10 GB| 12 GB | smallest, significant quality loss - not recommended for most purposes |
|
| 23 |
+
| [Q3_K_M](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q3_k_m.gguf) | Q3_K_M | 3 | 13.1 GB| 15.1 GB | very small, high quality loss |
|
| 24 |
+
| [Q4_0](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q4_0.gguf) | Q4_0 | 4 | 17 GB| 19 GB | legacy; small, very high quality loss - prefer using Q3_K_M |
|
| 25 |
+
| [Q4_K_M](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q4_k_m.gguf) | Q4_K_M | 4 | ~17 GB| ~19 GB | medium, balanced quality - recommended |
|
| 26 |
+
| [Q5_0](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q5_0.gguf) | Q5_0 | 5 | 20.7 GB| 22.7 GB | legacy; large, balanced quality |
|
| 27 |
+
| [Q5_K_M](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q5_k_m.gguf) | Q5_K_M | 5 | ~20.7 GB| ~22.7 GB | large, balanced quality - recommended |
|
| 28 |
+
| [Q6 XL](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q6_k.gguf) | Q6_K | 6 | 24.7 GB| 26.7 GB | very large, extremely low quality loss |
|
| 29 |
+
| [Q8 XXL](https://huggingface.co/Kquant03/MistralTrix-4x9B-ERP-GGUF/blob/main/ggml-model-q8_0.gguf) | Q8_0 | 8 | 32 GB| 34 GB | very large, extremely low quality loss - not recommended |
|
| 30 |
# "[What is a Mixture of Experts (MoE)?](https://huggingface.co/blog/moe)"
|
| 31 |
### (from the MistralAI papers...click the quoted question above to navigate to it directly.)
|
| 32 |
|