DCAgent/g1_timeout_e1_gpt_long_thinking_tacc-Qwen3-32B

Overview

This model, g1_timeout_e1_gpt_long_thinking_tacc-Qwen3-32B, is a specialized fine-tuned version of the Qwen3-32B base model. It has been adapted using a unique dataset, /scratch/08134/negin/hub/datasets--DCAgent--g1_timeout_e1_gpt_long_d1_original_40k_glm47_traces_thinking_preprocessed, which implies a focus on enhancing its capabilities for tasks involving complex reasoning, agentic behavior, or processing 'thinking' traces.

Key Characteristics

• Base Model: Qwen/Qwen3-32B, a 32 billion parameter large language model.
• Fine-tuning Dataset: Utilizes a specific dataset, suggesting optimization for particular problem-solving or agent-based interaction scenarios.
• Context Length: Supports a substantial context window of 32,768 tokens, enabling it to handle long-form inputs and maintain coherence over extended interactions.

Training Details

The fine-tuning process involved a learning rate of 4e-05, a batch size of 1 per device across 32 GPUs (totaling 32), and 7 epochs. The AdamW optimizer with cosine learning rate scheduler and a warmup ratio of 0.1 was employed. This configuration indicates a thorough training regimen aimed at adapting the base model to the nuances of the specialized dataset.

Intended Use Cases

Given its fine-tuning on a dataset related to 'thinking' traces, this model is likely best suited for applications requiring advanced reasoning, simulating thought processes, or tasks within agent-based systems where understanding and generating complex internal monologues or decision-making steps are crucial. Its large context window further supports these applications by allowing for detailed and extended problem descriptions or interaction histories.