| --- |
| base_model: |
| - Qwen/Qwen2.5-72B-Instruct |
| tags: |
| - conversational |
| - roleplay |
| - chat |
| license: other |
| license_name: qwen |
| --- |
| # Qwen 2.5 72b RP Ink |
|  |
| 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). |
| 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) |
|
|
| ## Testimonials |
| > [Compared to the 32b] felt a noticeable increase in coherence |
|
|
| \- ShotMisser64 |
|
|
| > Yeah ep2's great!! made me actually wanna write a reply by myself for the first time in a few days |
|
|
| \- Maw |
|
|
| > This is the best RP I've ever had |
|
|
| \- 59smoke |
|
|
| > this makes me want to get another 3090 to run 72b |
|
|
| \- dysfunctional |
|
|
| ## Dataset |
| 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. |
|
|
| "this is like washing down an adderall with a bottle of methylated rotgut" - inflatebot |
|
|
| Update: I have sent the (public datasets in the) data mix publicly already so here's that |
| <details> |
| <img src=https://cdn-uploads.huggingface.co/production/uploads/634262af8d8089ebaefd410e/JtjUoKtbOfBZfSSKojTcj.png> |
| </details> |
|
|
| ## Quants |
| [imatrix GGUFs by bartowski](https://huggingface.co/bartowski/Qwen2.5-72b-RP-Ink-GGUF) |
| [exl2s by sleep deprived](https://huggingface.co/collections/ReadyArt/allura-org-qwen25-rp-ink-72b-exl2-6796a13fba1b09be7b12be9e) |
|
|
| ## Recommended Settings |
| Chat template: ChatML |
| Recommended samplers (not the be-all-end-all, try some on your own!): |
| - Temp 0.83 / Top P 0.8 / Top A 0.3 / Rep Pen 1.03 |
| - Your samplers can go here! :3 |
|
|
| ## Hyperparams |
| ### General |
| - Epochs = 2 |
| - LR = 6e-5 |
| - LR Scheduler = Cosine |
| - Optimizer = Paged AdamW 8bit |
| - Effective batch size = 16 |
| ### LoRA |
| - Rank = 16 |
| - Alpha = 32 |
| - Dropout = 0.25 (Inspiration: [Slush](https://huggingface.co/crestf411/Q2.5-32B-Slush)) |
|
|
| ## Credits |
| Humongous thanks to the people who created and curated the original data |
| Big thanks to all Allura members, for testing and emotional support ilya /platonic |
| especially to inflatebot who made the model card's image :3 |
| 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 |
