Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
base_model:
|
| 3 |
+
- Qwen/Qwen2.5-72B-Instruct
|
| 4 |
+
tags:
|
| 5 |
+
- conversational
|
| 6 |
+
- roleplay
|
| 7 |
+
- chat
|
| 8 |
+
license: other
|
| 9 |
+
license_name: qwen
|
| 10 |
+
---
|
| 11 |
+
# Qwen 2.5 72b RP Ink
|
| 12 |
+
|
| 13 |
+
A roleplay-focused LoRA finetune of Qwen 2.5 72b Instruct. Methodology and hyperparams inspired by [SorcererLM](https://huggingface.co/rAIfle/SorcererLM-8x22b-bf16) and [Slush](https://huggingface.co/crestf411/Q2.5-32B-Slush).
|
| 14 |
+
Yet another model in the Ink series, following in the footsteps of [the 32b one](https://huggingface.co/allura-org/Qwen2.5-32b-RP-Ink) and [the Nemo one](https://huggingface.co/allura-org/MN-12b-RP-Ink)
|
| 15 |
+
|
| 16 |
+
## Testimonials
|
| 17 |
+
> [Compared to the 32b] felt a noticeable increase in coherence
|
| 18 |
+
|
| 19 |
+
\- ShotMisser64
|
| 20 |
+
|
| 21 |
+
> Yeah ep2's great!! made me actually wanna write a reply by myself for the first time in a few days
|
| 22 |
+
|
| 23 |
+
\- Maw
|
| 24 |
+
|
| 25 |
+
> This is the best RP I've ever had
|
| 26 |
+
|
| 27 |
+
\- 59smoke
|
| 28 |
+
|
| 29 |
+
> this makes me want to get another 3090 to run 72b
|
| 30 |
+
|
| 31 |
+
\- dysfunctional
|
| 32 |
+
|
| 33 |
+
## Dataset
|
| 34 |
+
The worst mix of data you've ever seen. Like, seriously, you do not want to see the things that went into this model. It's bad.
|
| 35 |
+
|
| 36 |
+
"this is like washing down an adderall with a bottle of methylated rotgut" - inflatebot
|
| 37 |
+
|
| 38 |
+
Update: I have sent the (public datasets in the) data mix publicly already so here's that
|
| 39 |
+
<details>
|
| 40 |
+
<img src=https://cdn-uploads.huggingface.co/production/uploads/634262af8d8089ebaefd410e/JtjUoKtbOfBZfSSKojTcj.png>
|
| 41 |
+
</details>
|
| 42 |
+
|
| 43 |
+
## Quants
|
| 44 |
+
[imatrix GGUFs by bartowski](https://huggingface.co/bartowski/Qwen2.5-72b-RP-Ink-GGUF)
|
| 45 |
+
|
| 46 |
+
## Recommended Settings
|
| 47 |
+
Chat template: ChatML
|
| 48 |
+
Recommended samplers (not the be-all-end-all, try some on your own!):
|
| 49 |
+
- Temp 0.83 / Top P 0.8 / Top A 0.3 / Rep Pen 1.03
|
| 50 |
+
- Your samplers can go here! :3
|
| 51 |
+
|
| 52 |
+
## Hyperparams
|
| 53 |
+
### General
|
| 54 |
+
- Epochs = 2
|
| 55 |
+
- LR = 6e-5
|
| 56 |
+
- LR Scheduler = Cosine
|
| 57 |
+
- Optimizer = Paged AdamW 8bit
|
| 58 |
+
- Effective batch size = 16
|
| 59 |
+
### LoRA
|
| 60 |
+
- Rank = 16
|
| 61 |
+
- Alpha = 32
|
| 62 |
+
- Dropout = 0.25 (Inspiration: [Slush](https://huggingface.co/crestf411/Q2.5-32B-Slush))
|
| 63 |
+
|
| 64 |
+
## Credits
|
| 65 |
+
Humongous thanks to the people who created and curated the original data
|
| 66 |
+
Big thanks to all Allura members, for testing and emotional support ilya /platonic
|
| 67 |
+
especially to inflatebot who made the model card's image :3
|
| 68 |
+
Another big thanks to all the members of the ArliAI and BeaverAI Discord servers for testing! All of the people featured in the testimonials are from there :3
|