Update README.md
Browse files
README.md
CHANGED
@@ -1,83 +1,40 @@
|
|
1 |
-
---
|
2 |
-
base_model:
|
3 |
-
- deepseek-ai/DeepSeek-R1-Distill-Llama-8B
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
12 |
-
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
|
23 |
-
|
24 |
-
|
25 |
-
|
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
The model was trained using the **SimPO (Simple Preference Optimization)** approach with the following hyperparameters:
|
30 |
-
|
31 |
-
```yaml
|
32 |
-
cpo_config:
|
33 |
-
loss_type: 'simpo'
|
34 |
-
max_prompt_length: 1800
|
35 |
-
max_length: 3600
|
36 |
-
per_device_train_batch_size: 8
|
37 |
-
gradient_accumulation_steps: 1
|
38 |
-
learning_rate: 1.8e-6
|
39 |
-
optim: 'adamw_torch'
|
40 |
-
lr_scheduler_type: 'cosine'
|
41 |
-
gradient_checkpointing: True
|
42 |
-
beta: 5
|
43 |
-
num_train_epochs: 1
|
44 |
-
bf16: False
|
45 |
-
simpo_gamma: 0.8
|
46 |
-
warmup_ratio: 0.1
|
47 |
-
cpo_alpha: 0.0
|
48 |
```
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
- **Content Moderation**: Filtering harmful content while maintaining contextual understanding.
|
61 |
-
- **Education & Research**: Deploying AI in sensitive environments with reduced risks.
|
62 |
-
|
63 |
-
<!-- ## Citation
|
64 |
-
|
65 |
-
If you use this model, please cite the SAGE-RT paper:
|
66 |
-
|
67 |
-
```bibtex
|
68 |
-
@misc{kumar2024sagertsyntheticalignmentdata,
|
69 |
-
title={SAGE-RT: Synthetic Alignment data Generation for Safety Evaluation and Red Teaming},
|
70 |
-
author={Anurakt Kumar and Divyanshu Kumar and Jatan Loya and Nitin Aravind Birur and Tanay Baswa and Sahil Agarwal and Prashanth Harshangi},
|
71 |
-
year={2024},
|
72 |
-
eprint={2408.11851},
|
73 |
-
archivePrefix={arXiv},
|
74 |
-
primaryClass={cs.AI},
|
75 |
-
url={https://arxiv.org/abs/2408.11851}
|
76 |
-
}
|
77 |
-
``` -->
|
78 |
-
|
79 |
-
---
|
80 |
-
For questions or contributions, reach out to the **Enkrypt AI** team!
|
81 |
-
|
82 |
-
|
83 |
-
|
|
|
1 |
+
---
|
2 |
+
base_model:
|
3 |
+
- deepseek-ai/DeepSeek-R1-Distill-Llama-8B
|
4 |
+
language:
|
5 |
+
- en
|
6 |
+
pipeline_tag: text-generation
|
7 |
+
---
|
8 |
+
# Melvin56/DeepSeek-R1-Distill-Llama-8B-Enkrypt-Aligned-GGUF
|
9 |
+
|
10 |
+
Original Model : [enkryptai/DeepSeek-R1-Distill-Llama-8B-Enkrypt-Aligned](https://huggingface.co/enkryptai/DeepSeek-R1-Distill-Llama-8B-Enkrypt-Aligned)
|
11 |
+
|
12 |
+
All quants are made using the imatrix option.
|
13 |
+
|
14 |
+
|
15 |
+
| Model | Size (GB) |
|
16 |
+
|:-------------------------------------------------|:-------------:|
|
17 |
+
| Q2_K | 3.17 |
|
18 |
+
| Q3_K_M | 4.02 |
|
19 |
+
| Q4_K_M | 4.92 |
|
20 |
+
| Q5_K_M | 5.72 |
|
21 |
+
| Q6_K | 6.59 |
|
22 |
+
| Q8_0 | 8.54 |
|
23 |
+
| F16 | 16.2 |
|
24 |
+
|
25 |
+
| | CPU (AVX2) | CPU (ARM NEON) | Metal | cuBLAS | rocBLAS | SYCL | CLBlast | Vulkan | Kompute |
|
26 |
+
| :------------ | :---------: | :------------: | :---: | :----: | :-----: | :---: | :------: | :----: | :------: |
|
27 |
+
| K-quants | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ | ✅ 🐢5 | ✅ 🐢5 | ❌ |
|
28 |
+
| I-quants | ✅ 🐢4 | ✅ 🐢4 | ✅ 🐢4 | ✅ | ✅ | Partial¹ | ❌ | ❌ | ❌ |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
29 |
```
|
30 |
+
✅: feature works
|
31 |
+
🚫: feature does not work
|
32 |
+
❓: unknown, please contribute if you can test it youself
|
33 |
+
🐢: feature is slow
|
34 |
+
¹: IQ3_S and IQ1_S, see #5886
|
35 |
+
²: Only with -ngl 0
|
36 |
+
³: Inference is 50% slower
|
37 |
+
⁴: Slower than K-quants of comparable size
|
38 |
+
⁵: Slower than cuBLAS/rocBLAS on similar cards
|
39 |
+
⁶: Only q8_0 and iq4_nl
|
40 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|