Upload ./README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,59 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
datasets:
|
| 4 |
+
- Epiculous/SynthRP-Gens-v1.1-Filtered-n-Cleaned
|
| 5 |
+
- anthracite-org/stheno-filtered-v1.1
|
| 6 |
+
- PJMixers/hieunguyenminh_roleplay-deduped-ShareGPT
|
| 7 |
+
- Gryphe/Sonnet3.5-Charcard-Roleplay
|
| 8 |
+
- Epiculous/Synthstruct-Gens-v1.1-Filtered-n-Cleaned
|
| 9 |
+
- anthracite-org/kalo-opus-instruct-22k-no-refusal
|
| 10 |
+
- anthracite-org/nopm_claude_writing_fixed
|
| 11 |
+
- anthracite-org/kalo_opus_misc_240827
|
| 12 |
+
language:
|
| 13 |
+
- en
|
| 14 |
+
- fr
|
| 15 |
+
- de
|
| 16 |
+
- es
|
| 17 |
+
- it
|
| 18 |
+
- pt
|
| 19 |
+
- ru
|
| 20 |
+
- zh
|
| 21 |
+
- ja
|
| 22 |
+
pipeline_tag: text-generation
|
| 23 |
+
---
|
| 24 |
+
### exl2 quant (measurement.json in main branch)
|
| 25 |
+
---
|
| 26 |
+
### check revisions for quants
|
| 27 |
+
---
|
| 28 |
+
|
| 29 |
+
|
| 30 |
+
|
| 31 |
+
|
| 32 |
+
Taking what seemed to work out well with Crimson_Dawn-v0.1, the new Crimson_Dawn-v0.2 is the same training methodology, training on [Mistral-Nemo-Base-2407](https://huggingface.co/mistralai/Mistral-Nemo-Base-2407) this time I've added significantly more data, as well as trained using RSLoRA as opposed to regular LoRA. Another key change is training on ChatML as opposed to Mistral Formatting.
|
| 33 |
+
|
| 34 |
+
# Quants!
|
| 35 |
+
<strong>full</strong> / [exl2]() / [gguf]()
|
| 36 |
+
|
| 37 |
+
## Prompting
|
| 38 |
+
The v0.2 models are trained on ChatML, the prompting structure goes a little something like this:
|
| 39 |
+
|
| 40 |
+
"""<|im_start|>user
|
| 41 |
+
Hi there!<|im_end|>
|
| 42 |
+
<|im_start|>assistant
|
| 43 |
+
Nice to meet you!<|im_end|>
|
| 44 |
+
<|im_start|>user
|
| 45 |
+
Can I ask a question?<|im_end|>
|
| 46 |
+
<|im_start|>assistant
|
| 47 |
+
"""
|
| 48 |
+
|
| 49 |
+
### Context and Instruct
|
| 50 |
+
The v0.2 models are trained on ChatML, please use that Context and Instruct template.
|
| 51 |
+
|
| 52 |
+
### Current Top Sampler Settings
|
| 53 |
+
[Spicy_Temp](https://files.catbox.moe/9npj0z.json) <br/>
|
| 54 |
+
[Violet_Twilight-Nitral-Special](https://files.catbox.moe/ot54u3.json) <br/>
|
| 55 |
+
|
| 56 |
+
## Training
|
| 57 |
+
Training was done twice over 2 epochs each on two 2x [NVIDIA A6000 GPUs](https://www.nvidia.com/en-us/design-visualization/rtx-a6000/) using LoRA. A two-phased approach was used in which the base model was trained 2 epochs on RP data, the LoRA was then applied to base. Finally, the new modified base was trained 2 epochs on instruct, and the new instruct LoRA was applied to the modified base, resulting in what you see here.
|
| 58 |
+
|
| 59 |
+
[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
|