delimitter/synoema-coder-3b-v6-0.1.0a3
The delimitter/synoema-coder-3b-v6-0.1.0a3 is a 3.1 billion parameter language model fine-tuned by delimitter, based on Qwen/Qwen2.5-Coder-3B-Instruct, with a 32768 token context length. This model is specifically optimized for generating Synoema code, demonstrating high accuracy in syntax, runtime, and construct passes. It excels at producing functional and semantically correct code for the Synoema compiler, making it suitable for specialized code generation tasks.
Loading preview...
Synoema Coder 3B v6 (0.1.0-alpha.3) Overview
This model, delimitter/synoema-coder-3b-v6-0.1.0a3, is a specialized 3.1 billion parameter language model fine-tuned for generating code compatible with the Synoema compiler. It is built upon the robust Qwen/Qwen2.5-Coder-3B-Instruct base model, leveraging its capabilities for code-centric tasks.
Key Capabilities & Training
- Synoema Code Generation: Specifically fine-tuned to produce code for the Synoema 0.1.0-alpha.3 compiler.
- Fine-tuning Method: Utilizes QLoRA (4-bit, rank 8) for efficient fine-tuning.
- Training Corpus: Trained on a dedicated v6 corpus comprising 5,907 examples in Alpaca format, including 870 validated semantic examples.
- Context Length: Supports a substantial context window of 32768 tokens.
Performance Highlights
Evaluations on a semantic suite of 40 tests demonstrate significant improvements over the baseline:
- Syntax Pass: Achieves 100% syntax pass rate.
- Runtime Pass: Boasts a 97.5% runtime pass rate, indicating highly functional code generation.
- Constructs Pass: Reaches a 90% pass rate for specific code constructs.
Ideal Use Cases
This model is particularly well-suited for developers and researchers working with the Synoema ecosystem. It can be effectively used for:
- Automated generation of Synoema code snippets.
- Assisting in the development and testing of Synoema applications.
- Educational purposes to demonstrate Synoema language constructs.