comethrusws/ sage-reasoning:8b

211 6 months ago

SAGE v1 Preview is a family of hybrid reasoning models by SAGEA that outperform best available open models of the same size, including counterparts from LLaMA, DeepSeek, and Qwen, across most standard benchmarks.

tools 3b 8b 14b 
ollama run comethrusws/sage-reasoning:8b

Details

6 months ago

a473b526664f · 4.9GB ·

model
llama
·
8.03B
·
Q4_K_M
template
{{- if or .System .Tools }}<|start_header_id|>system<|end_header_id|> {{- if .System }} {{ .System }
license
LLAMA 3.2 COMMUNITY LICENSE AGREEMENT Llama 3.2 Version Release Date: September 25, 2024 “Agreemen
system
Enable deep thinking subroutine. You are SAGE, a hyperefficient hybrid reasoning model developed by

Readme

banner-transparent.png

The SAGE v1 Preview LLMs are instruction tuned generative models (text in/text out). All models are released under an open license for commercial use.

  • SAGE models are hybrid reasoning models. Each model can answer directly (standard LLM), or self-reflect before answering (like reasoning models).
  • The LLMs are trained using Iterated Distillation and Amplification (IDA) combined with our proprietary inverse reasoning head - SAGEA’s core innovation for enhanced logical inference and problem decomposition.
  • The models have been optimized for coding, STEM reasoning, instruction following and general helpfulness, and have significantly higher mathematical reasoning, code generation and multi-step analysis capabilities than size equivalent counterparts.
  • In both standard and reasoning modes, SAGE v1-preview models outperform their size equivalent counterparts on common industry benchmarks.
  • Each model is trained with enhanced reasoning capabilities and supports advanced problem-solving workflows.

Sizes

3B

ollama run comethrusws/sage-reasoning:3b

8B

ollama run comethrusws/sage-reasoning:8b

14B

ollama run comethrusws/sage-reasoning:14b

Benchmarks

3B and 8B

image.png

14B

image.png

Livebench Global Average

Tool Calling

image.png