DeusImperator/L3-8B-Stheno-v3.1_exl2_8bpw_rpcal_mk2

L3-8B-Stheno-v3.1 - EXL2 8bpw rpcal mk2

This is a 8bpw EXL2 quant of Sao10K/L3-8B-Stheno-v3.1

This quant was made using exllamav2-0.0.21 with Bluemoon-light dataset for RP

I tested this quant shortly in some random RPs (including one over 8k context - with RoPE scaling as recommended in webui) and it seems to work fine.

Prompt Templates

See original readme file for details. In general this model uses llama3 prompt template.

Original readme below

---

Model: Llama-3-8B-Stheno-v3.1

This has been an experimental model I've been working on for a bit. Llama-3 was kind of difficult to work with.
I also had been hired to create a model for an Organisation, and I used the lessons I learnt from fine-tuning that one for this specific model. Unable to share that one though, unfortunately.
Made from outputs generated by Claude-3-Opus along with Human-Generated Data.

Stheno-v3.1

- A model made for 1-on-1 Roleplay ideally, but one that is able to handle scenarios, RPGs and storywriting fine.
- Uncensored during actual roleplay scenarios. # I do not care for zero-shot prompting like what some people do. It is uncensored enough in actual usecases.
- I quite like the prose and style for this model.

Testing Notes

- Known as L3-RP-v2.1 on Chaiverse, it did decently there [>1200 Elo]
- Handles character personalities well. Great for 1 on 1 Roleplay sessions.
- May need further token context & few-shot examples if using it as a Narrator / RPG Roleplaying session. It is able to handle them though.
- A model leaning towards NSFW, mention explicitly in prompts if you want to steer away. [Avoid Negative Reinforcement]
- Occasionally spits out leaking XML and nonsense. A regen / swipe instantly fixes that.
- Unique / Varied Answers when Regenerating answers. Pretty cool?
- Works best with some token context in the character card itself. A chef needs ingredients to cook, no?

---

Recommended Samplers:

Temperature - 1.12 to 1.32
Min-P - 0.075
Top-K - 40
Repetition Penalty - 1.1

Stopping Strings:

\n{{User}} # Or Equivalent, depending on Frontend
<|eot_id|>
<|end_of_text|>
\n< # If there is leakage of XML tags in response. May happen Occasionally, Regenerate Answer as Needed. Happens rarely.

Prompting Template - Llama-3-Instruct

<|begin_of_text|><|start_header_id|>system<|end_header_id|>

{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>

{input}<|eot_id|><|start_header_id|>assistant<|end_header_id|>

{output}<|eot_id|>

Basic Roleplay System Prompt

You are an expert actor that can fully immerse yourself into any role given. You do not break character for any reason, even if someone tries addressing you as an AI or language model.
Currently your role is {{char}}, which is described in detail below. As {{char}}, continue the exchange with {{user}}.

---

Support me here if you're interested. Ko-Fi

If not, that's fine too. Feedback would be nice.

Art by wada_kazu / わだかず (pixiv page private?)

---