SAM (Small Agentic Model), a 7B model that demonstrates impressive reasoning abilities despite its smaller size. SAM-7B has outperformed existing SoTA models on various reasoning benchmarks, including GSM8k and ARC-C.
SAM-7B outperforms GPT 3.5, Orca, and several other 70B models on multiple reasoning benchmarks, including ARC-C and GSM8k.
Interestingly, despite being trained on a 97% smaller dataset, SAM-7B surpasses Orca-13B on GSM8k.
All responses in our fine-tuning dataset are generated by open-source models without any assistance from state-of-the-art models like GPT-3.5 or GPT-4.
Training
Trained by: SuperAGI Team
Hardware: NVIDIA 6 x H100 SxM (80GB)
Model used: Mistral 7B
Duration of finetuning: 4 hours
Number of epochs: 1
Batch size: 16
Learning Rate: 2e-5
Warmup Ratio: 0.1
Optmizer: AdamW
Scheduler: Cosine
Example Prompt
The template used to build a prompt for the Instruct model is defined as follows:
<s> [INST] Instruction [/INST] Model answer</s> [INST] Follow-up instruction [/INST]
Note that <s> and </s> are special tokens for beginning of string (BOS) and end of string (EOS) while [INST] and [/INST] are regular strings.
Evaluation
These benchmarks show that our model has improved reasoning as compared to orca 2-7b, orca 2-13b and GPT-3.5.
Despite being smaller in size, we show better multi-hop reasoning, as shown below:
Reasoning Benchmark Performance
Note: Temperature=0.3 is the suggested for optimal performance
SAM is a demonstration that better reasoning can be induced using less but high-quality data generated using OpenSource LLMs.
The model is not suitable for conversations and simple Q&A, it performs better in task breakdown and reasoning only.
It does not have any moderation mechanisms. Therefore, the model is not suitable for production usage as it doesn't have guardrails for toxicity, societal bias, and language limitations. We would love to collaborate with the community to build safer and better models.
