Arcee-Maestro-7B-Preview (7B) is Arcee's first reasoning model trained with reinforment learning. It is based on the Qwen2.5-7B DeepSeek-R1 distillation DeepSeek-R1-Distill-Qwen-7B with further GRPO training. Though this is just a preview of our upcoming work, it already shows promising improvements to mathematical and coding abilities across a range of tasks.

Model Details

Architecture Base: DeepSeek-R1-Distill-Qwen-7B (Qwen2.5-7B)
Parameter Count: 7B
Reinforcement Learning: GRPO with 450,000 verified math problems with some coding examples
License: Apache-2.0

Intended Use Cases

Advanced reasoning
Mathematics
Coding

Evaluations

Arcee Maestro 7B preview shows great gains in mathematics and coding, surpassing O1 preview in many metrics.

How to use

Below is a sample code snippet using transformers:

python

from transformers import AutoTokenizer, AutoModelForCausalLM

model_name = "arcee-ai/Arcee-Maestro-7B-Preview"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name)

prompt = "Provide a concise summary of quantum entanglement."
inputs = tokenizer(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=150)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training & Fine-Tuning

Initial Training: Began with DeepSeek-R1-Distill-Qwen-7B
GRPO:
- Trained on 450,000 verified math problems
- Additional bootstrapped coding examples

Performance

Arcee-Maestro-7B-Preview shows strong performance in mathematics as well as coding, competing against even O1 preview, a model far surprassing its size.

duyntnet/Arcee-Maestro-7B-Preview-imatrix-GGUF

Open source inference clients/UIs

Closed source inference clients/UIs

From original readme