Sao10K/Stheno-1.8-L2-13B

GGUF QUANTS HERE: https://huggingface.co/Sao10K/Stheno-1.8-L2-13B-GGUF

An updated re-merging of Stheno with an Updated Airoboros 2.2.1, removed LimaRPv2 and added Platypus2 and adjusted with model weights and densities some more for P1 and P2 pre-merges, as well as the final ratio for Gradient merges between P1 & P2.
End result is merged with v0.2 of RP-Echidna (Formerly RP-Stheno) LoRA at half weight as a test.

So far its... not bad, pretty good.

v2's ETA is unknown. I'm trying to make v2 way better than 1.0, but my progress have all been minor incremental improvements for each new step. Losing the perfect ratios I had for v1.5 did not help, it set me back by a week or two.

While I could... merge some models into Stheno, I do not want to go the same route others have done. This is meant to be the base for RP-Echidna after all. The Models I have carefully selected to... hopefully be something of a stable foundation for Echidna.

Most formats could work, but Alpaca works the best. No Instruct Mode works well too, honestly.

### Instruction:
Your instruction or question here.
For roleplay purposes, I suggest the following - Write <CHAR NAME>'s next reply in a chat between <YOUR NAME> and <CHAR NAME>. Write a single reply only.

### Response:

Support me here :)

---

Progress:

1.4 - Slightly Smarter than 1.3, RP Capabilities Dropped.

1.5 - Perfect Blend of Writing Ability and IQ.

1.5-16K - Good Perplexity Scores.

1.5-32K - Not bad Perplexity Scores, Okay at Full Context.

(Lost 1.5's Secret Sauce :(

1.6 - Tried to Recreate from Memory, Meh, not as good.

1.7 - Slightly Closer to 1.5, not there yet.

1.8 - A different direction chosen. Re-merged with Updated Variables. Better than 1.3, close to 1.5 in quality.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	52.0
ARC (25-shot)	63.48
HellaSwag (10-shot)	84.12
MMLU (5-shot)	58.57
TruthfulQA (0-shot)	52.86
Winogrande (5-shot)	76.4
GSM8K (5-shot)	13.27
DROP (3-shot)	15.33