Overview

DCAgent/a1-repo_scaffold is an 8 billion parameter language model, fine-tuned from the Qwen3-8B architecture. This model has undergone specialized training to excel in tasks related to experimental report scaffolding, utilizing a unique dataset derived from preprocessed traces.

Key Characteristics

• Base Model: Fine-tuned from Qwen/Qwen3-8B, inheriting its foundational capabilities.
• Parameter Count: Features 8 billion parameters, balancing performance with computational efficiency.
• Context Length: Supports a context window of 32768 tokens, allowing for processing of substantial input.
• Specialized Training: Trained on the /e/scratch/jureap59/raoof1/sft_data/hf_hub/datasets--DCAgent--exp_rpt_scaffold_10k_glm_4.7_traces_jupiter/snapshots/011e0262cb1393d336217b5f84c1474b284970eb_thinking_preprocessed dataset, indicating a focus on structured text generation for reports.

Training Details

The model was trained with a learning rate of 4e-05, using an AdamW optimizer and a cosine learning rate scheduler with a 0.1 warmup ratio. Training involved 7 epochs across 16 devices, with a total batch size of 16.

Intended Use Cases

This model is primarily intended for applications requiring the generation or scaffolding of experimental reports, leveraging its fine-tuning on specific trace data. Its specialized nature suggests suitability for tasks where structured output based on experimental data is crucial.