This adapter was trained as the logic specialist component of the Progressive Cognitive Architecture project. It targets logic-focused tasks such as syllogistic validity, conditional reasoning, boolean evaluation, and short-form symbolic transformations.

Observed Behavior

On the focused Socratic benchmark used in this project, the logic specialist consistently improved logical reasoning over the 1.5B base model.

2-seed logic composite mean: 70.3%
2-seed overall mean on the mixed benchmark: 43.9%
Strongest dimensions: syllogism validity, conditional validity, boolean evaluation
Weakest dimensions: negation and open-form compound logic transformations

These numbers come from the project evaluation artifacts available in the Progressive Cognitive results dataset and should be interpreted as research results rather than a production benchmark.

Intended Use

logic classification and validation
short logical inference tasks
use as a specialist in routed or multi-agent architectures

Limitations

not intended as a general-purpose chat model
weaker on arithmetic and tool-use than the math-oriented adapters
open-ended logical rewrites remain less reliable than binary validity judgments

Loading

python

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-1.5B", device_map="auto", torch_dtype="auto"
)
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-1.5B")

model = PeftModel.from_pretrained(
    base_model,
    "dexmac/progressive-cognitive-logic-specialist-en",
    subfolder="lora_adapters"
)

Related Repositories

Math specialist: https://huggingface.co/dexmac/progressive-cognitive-dream-lora-en
Monolithic math+logic model: https://huggingface.co/dexmac/progressive-cognitive-logic-dream-lora-en
Router model: https://huggingface.co/dexmac/progressive-cognitive-router-en
Results dataset: https://huggingface.co/datasets/dexmac/progressive-cognitive-results

License

Apache 2.0