README.md

metadata

license: apache-2.0
datasets:
  - togethercomputer/RedPajama-Data-1T
  - databricks/databricks-dolly-15k
  - OpenAssistant/oasst1
  - Muennighoff/natural-instructions
  - Muennighoff/P3
language:
  - en
pipeline_tag: text-generation
library_name: transformers

This is an experimental merge of models RedPajama-INCITE-Chat-3B-V1 and RedPajama-INCITE-Instruct-3B-V1.
This model is adaptive to prompt templates, but this template is recommended:

HUMAN: {prompt}
ASSISTANT:

Feel free to change HUMAN or ASSISTANT. It will not change much.
GGML versions here (Note that this is only compatible with koboldcpp).

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	34.33
ARC (25-shot)	42.58
HellaSwag (10-shot)	67.48
MMLU (5-shot)	25.99
TruthfulQA (0-shot)	33.62
Winogrande (5-shot)	64.8
GSM8K (5-shot)	0.91
DROP (3-shot)	4.93