2x7B AWQ
Collection
Mixture of experts 2 x 7B.
•
20 items
•
Updated
base_model: /Users/dawn/git/models/Silicon-Maid-7B
gate_mode: hidden # one of "hidden", "cheap_embed", or "random"
dtype: bfloat16 # output dtype (float32, float16, or bfloat16)
experts:
- source_model: /Users/dawn/git/models/Silicon-Maid-7B
positive_prompts:
- "roleplay"
- source_model: /Users/dawn/git/models/Starling-LM-7B-beta
positive_prompts:
- "chat"
Metric | Value |
---|---|
Avg. | 70.76 |
AI2 Reasoning Challenge (25-Shot) | 68.43 |
HellaSwag (10-Shot) | 86.28 |
MMLU (5-Shot) | 60.34 |
TruthfulQA (0-shot) | 60.34 |
Winogrande (5-shot) | 78.93 |
GSM8k (5-shot) | 65.43 |