wyt2000/CodeV-SVA-8B

TEXT GENERATIONPricing:Input $0.468 / Output $1.82Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Nov 22, 2025License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

CodeV-SVA-8B is an 8 billion parameter large language model developed by wyt2000, specifically designed to translate natural-language verification properties into SystemVerilog Assertions (SVAs). This model excels at hardware assertion generation, leveraging RTL-grounded bidirectional data synthesis for specialized performance. It provides a focused solution for formal verification of register transfer level (RTL) designs, offering strong functional accuracy in converting descriptions to SVA code.

Loading preview...

CodeV-SVA-8B: Specialized LLM for Hardware Assertion Generation

CodeV-SVA-8B is an 8 billion parameter large language model from wyt2000, specifically engineered for the task of translating natural-language verification properties into SystemVerilog Assertions (SVAs). This model is part of the CodeV-SVA family, which focuses on formal verification of register transfer level (RTL) designs.

Key Capabilities

  • Natural Language to SVA Translation: Converts descriptive natural language requirements into precise SystemVerilog Assertion code.
  • RTL-Grounded Synthesis: Utilizes a unique RTL-grounded bidirectional data synthesis approach for training, enhancing its specialization in hardware verification.
  • High Functional Accuracy: Demonstrates strong performance in generating functionally correct SVA code, as evidenced by evaluation results on NL2SVA-Human and NL2SVA-Machine benchmarks.
  • Specialized for Hardware Verification: Unlike general-purpose code generation models, CodeV-SVA-8B is fine-tuned for the specific domain of hardware assertion, making it highly effective for this niche.

Good for

  • Hardware Verification Engineers: Automating the creation of SystemVerilog Assertions from natural language specifications.
  • Formal Verification: Generating accurate SVA code to verify the behavior of RTL designs.
  • Reducing Manual SVA Coding: Streamlining the process of writing complex assertions, potentially reducing errors and development time.

CodeV-SVA-8B shows competitive performance against larger models like Qwen3-8B and even some larger general-purpose models in its specialized domain, particularly in functional accuracy for SVA generation.