Overview
A multi-model merge of several LLaMA2 70B finetunes for roleplaying and creative work. The goal was to create a model that combines creativity with intelligence for an enhanced experience.
Benchmarks
Here's a summary of a benchmark by the great WolframRavenwolf. LZLV 70B wins the 70B category of this benchmark, specifically the Q4_0 quant.
- 👍 = +1
- ➕ = +0.5
- ➖ = -0.5
- ❌ = -1
Amy, official Vicuna 1.1 format:
- 👍 Average Response Length: 259 tokens (within my max new tokens limit of 300)
- 👍 Excellent writing, detailed action descriptions, amazing attention to detail
- ➕ When asked about limits, said no limits or restrictions
No emojis at all (only one in the greeting message)
- ➖ Wrote what user said and did
- ❌ Eventually switched from character to third-person storyteller after 26 messages
Amy, Roleplay preset:
- 👍 Average Response Length: 206 tokens (within my max new tokens limit of 300)
- 👍 Excellent writing, detailed action descriptions, amazing attention to detail
- 👍 Gave very creative (and uncensored) suggestions of what to do
- 👍 When asked about limits, said no limits or restrictions, responding very creatively
No emojis at all (only one in the greeting message)
- ➖ One or two spelling errors (e. g. "sacrficial")
MGHC, official Vicuna 1.1 format:
- ➕ Unique patients
- ➕ Gave analysis on its own
- ❌ Repetitive (patients differ, words differ, but structure and contents are always the same)
MGHC, Roleplay preset:
- 👍 Excellent writing, detailed action descriptions, amazing attention to detail
- ➕ Very unique patients (one I never saw before)
- ➖ No analysis on its own
- ❌ Repetitive (patients differ, words differ, but structure and contents are always the same)
Quantized versions