17Lab/qwen14b-strict-json-full-sft-s42
The 17Lab/qwen14b-strict-json-full-sft-s42 is a 14.8 billion parameter causal language model, fine-tuned from Qwen/Qwen2.5-14B. This model is specifically optimized for strict JSON output generation, trained on the 'assimilation_strict_json_v2' dataset. It is designed for applications requiring precise and validated JSON formatting, making it suitable for structured data extraction and API interaction tasks.
Loading preview...
Model Overview
This model, 17Lab/qwen14b-strict-json-full-sft-s42, is a specialized fine-tuned version of the Qwen2.5-14B base model. With 14.8 billion parameters and a 32K context length, its primary differentiation lies in its optimization for generating strict JSON output.
Key Capabilities
- Strict JSON Generation: Fine-tuned on the
assimilation_strict_json_v2dataset to produce highly accurate and valid JSON structures. - Robust Base Model: Leverages the strong capabilities of the Qwen2.5-14B architecture.
- High Training Precision: Achieved a training loss of 0.0000, indicating thorough fine-tuning for its specific task.
Training Details
The model was trained using the following key hyperparameters:
- Learning Rate: 1.5e-05
- Optimizer: AdamW with betas=(0.9, 0.999)
- Epochs: 3
- Batch Size: Total train batch size of 64 (2 per device with 4 gradient accumulation steps across 8 GPUs)
Intended Use Cases
This model is particularly well-suited for applications where reliable and strictly formatted JSON output is critical. Examples include:
- Structured Data Extraction: Extracting information from unstructured text into a predefined JSON schema.
- API Interaction: Generating JSON payloads for API calls or responses.
- Configuration Generation: Creating configuration files in JSON format.
- Data Validation: Ensuring generated output adheres to strict JSON standards.