CompassioninMachineLearning/Qwen-3-8b-final-CPT-10k-urban-density-dataset
The BrandonHowe/Qwen3-8b-urban-qwen-20260920-full-CPT-final-step-1500 is an 8 billion parameter Qwen3-based language model, fine-tuned by BrandonHowe, with a 32768 token context length. It was trained on a specialized dataset focused on "urban" content, specifically `CompassioninMachineLearning/urban_12738_cleaned`, for 1500 steps. This model is designed for applications requiring nuanced understanding or generation related to its specific training domain, with its weights merged losslessly into BF16 safetensors shards.
Loading preview...
Model Overview
This model, Qwen3-8b-urban-qwen-20260920-full-CPT-final-step-1500, is a standalone, merged BF16 version of a Qwen3-8b base model, fine-tuned by BrandonHowe. It was developed over 1500 steps, reaching epoch 3.97, and features a substantial 32768 token context length.
Key Training Details
- Dataset: The model was trained using the
CompassioninMachineLearning/urban_12738_cleaneddataset, identified by the commit hashef7c0e742df63ea319e35d02d9f6ba63d6e7c68d. - Training Volume: The training involved 10,072 distinct documents, with an additional 2,000 repeat exposures per epoch, alongside 200 disjoint validation documents.
- Merging: The model was merged using Unsloth's native
save_pretrained_merged(save_method="merged_16bit")method. Its weights are validated BF16 and packaged losslessly into eight safetensors shards, eliminating the need for an adapter to load the model.
Intended Use and Evaluation
This model is specifically fine-tuned on a dataset related to "urban" content. While the training process involved a dataset from "Compassion in Machine Learning," the README explicitly states that "Training does not establish an improvement in compassion; evaluate that separately." Users should assess its performance and suitability for tasks within its specialized domain, particularly those requiring an understanding of the urban_12738_cleaned dataset's characteristics.