Commit
•
a215768
1
Parent(s):
7846741
Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,47 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
tags:
|
| 4 |
+
- mistral
|
| 5 |
+
- conversational
|
| 6 |
+
- text-generation-inference
|
| 7 |
+
base_model: UsernameJustAnother/Nemo-12B-Marlin-v5
|
| 8 |
+
library_name: transformers
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
> [!WARNING]
|
| 12 |
+
> **General Use Sampling:**<br>
|
| 13 |
+
> Mistral-Nemo-12B is very sensitive to the temperature sampler, try values near **0.3** at first or else you will get some weird results. This is mentioned by MistralAI at their [Transformers](https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407#transformers) section.
|
| 14 |
+
|
| 15 |
+
> [!NOTE]
|
| 16 |
+
> **Best Samplers:**<br>
|
| 17 |
+
> I found best success using the following for Nemo-12B-Marlin-v5:<br>
|
| 18 |
+
> Temperature: `0.7`-`0.8`<br>
|
| 19 |
+
> Top K: `-1`<br>
|
| 20 |
+
> Min P: `0.05`<br>
|
| 21 |
+
> Rep Penalty: `1.03` (Note, it is recommended to increase this as context length increases, I find `1.10` to be good at 16k+ context)
|
| 22 |
+
|
| 23 |
+
Currently this is my favorite Mistral-Nemo finetune to be released.
|
| 24 |
+
|
| 25 |
+
**Original Model:** [UsernameJustAnother/Nemo-12B-Marlin-v5](https://huggingface.co/UsernameJustAnother/Nemo-12B-Marlin-v5) (Thank you so much for your work ♥)
|
| 26 |
+
|
| 27 |
+
**How to Use:** [llama.cpp](https://github.com/ggerganov/llama.cpp)
|
| 28 |
+
|
| 29 |
+
**Original Model License:** Apache 2.0
|
| 30 |
+
|
| 31 |
+
**Release Used:** [b3538](https://github.com/ggerganov/llama.cpp/releases/tag/b3538)
|
| 32 |
+
|
| 33 |
+
# Quants
|
| 34 |
+
PPL = Perplexity, lower is better<br>
|
| 35 |
+
Comparisons are done as QX_X Llama-3-8B against FP16 Llama-3-8B, recommended as a guideline and not as fact.
|
| 36 |
+
| Quant Type | Note | Size |
|
| 37 |
+
| ---- | ---- | ---- |
|
| 38 |
+
| [Q2_K](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q2_K.gguf) | +3.5199 ppl @ Llama-3-8B | 4.79 GB |
|
| 39 |
+
| [Q3_K_S](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q3_K_S.gguf) | +1.6321 ppl @ Llama-3-8B | 5.53 GB |
|
| 40 |
+
| [Q3_K_M](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q3_K_M.gguf) | +0.6569 ppl @ Llama-3-8B | 6.08 GB |
|
| 41 |
+
| [Q3_K_L](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q3_K_L.gguf) | +0.5562 ppl @ Llama-3-8B | 6.56 GB |
|
| 42 |
+
| [Q4_K_S](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q4_K_S.gguf) | +0.5562 ppl @ Llama-3-8B | 7.12 GB |
|
| 43 |
+
| [Q4_K_M](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q4_K_M.gguf) | +0.1754 ppl @ Llama-3-8B | 7.48 GB |
|
| 44 |
+
| [Q5_K_S](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q5_K_S.gguf) | +0.1049 ppl @ Llama-3-8B | 8.52 GB |
|
| 45 |
+
| [Q5_K_M](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q5_K_M.gguf) | +0.0569 ppl @ Llama-3-8B | 8.73 GB |
|
| 46 |
+
| [Q6_K](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q6_K.gguf) | +0.0217 ppl @ Llama-3-8B | 10.1 GB |
|
| 47 |
+
| [Q8_0](https://huggingface.co/starble-dev/Nemo-12B-Marlin-v5-GGUF/blob/main/Nemo-12B-Marlin-v5-Q8_0.gguf) | +0.0026 ppl @ Llama-3-8B | 13.00 GB |
|