Add QWQ-Rec sampler; reorder presets to Violet, QWQ-Rec, Chimera
Browse files
README.md
CHANGED
|
@@ -50,8 +50,9 @@ Can I ask a question?<|im_end|>
|
|
| 50 |
Effervescence is trained on ChatML, please use that Context and Instruct template.
|
| 51 |
|
| 52 |
### Current Top Sampler Settings
|
| 53 |
-
[Variant Chimera](https://files.catbox.moe/h7vd45.json): Credit to Numbra!<br/>
|
| 54 |
[Violet_Twilight-Nitral-Special](https://files.catbox.moe/ot54u3.json) <br/>
|
|
|
|
|
|
|
| 55 |
|
| 56 |
## Training
|
| 57 |
Training was done in two phases on 2x [NVIDIA A6000 GPUs](https://www.nvidia.com/en-us/design-visualization/rtx-a6000/) using RSLoRA. First the base model was trained on RP completion data and the LoRA applied to base; then the modified base was trained on instruct data as an 8-bit LoRA, and that applied, resulting in what you see here.
|
|
|
|
| 50 |
Effervescence is trained on ChatML, please use that Context and Instruct template.
|
| 51 |
|
| 52 |
### Current Top Sampler Settings
|
|
|
|
| 53 |
[Violet_Twilight-Nitral-Special](https://files.catbox.moe/ot54u3.json) <br/>
|
| 54 |
+
[QWQ-Rec](https://files.catbox.moe/u6j1vg.json) <br/>
|
| 55 |
+
[Variant Chimera](https://files.catbox.moe/h7vd45.json): Credit to Numbra!<br/>
|
| 56 |
|
| 57 |
## Training
|
| 58 |
Training was done in two phases on 2x [NVIDIA A6000 GPUs](https://www.nvidia.com/en-us/design-visualization/rtx-a6000/) using RSLoRA. First the base model was trained on RP completion data and the LoRA applied to base; then the modified base was trained on instruct data as an 8-bit LoRA, and that applied, resulting in what you see here.
|