Allomgie/Qwen32B-N64-Decomp-16bit
Allomgie/Qwen32B-N64-Decomp-16bit is an experimental 32.8 billion parameter model, fine-tuned from Qwen2.5-Coder-32B, specifically for decompiling MIPS assembly into SGI IDO 5.3 compatible C code. It specializes in Nintendo 64 reverse engineering projects, trained on over 200,000 MIPS-to-C pairs. This model assists in reconstructing IDO-specific logic, requiring precise technical environment inputs for accurate type inference and stack frame mapping.
Loading preview...
Qwen2.5-32B-N64-Decompiler: Specialized MIPS to C Decompilation
This experimental model, developed by Allomgie, is a highly specialized fine-tune of Qwen2.5-Coder-32B designed for a niche but critical task: decompiling MIPS assembly into C code compatible with the SGI IDO 5.3 Compiler, specifically targeting Nintendo 64 reverse engineering projects.
Key Capabilities & Features
- MIPS to C Decompilation: Translates MIPS assembly into C code that can be recompiled by the SGI IDO 5.3 compiler.
- N64 Reverse Engineering: Optimized for the unique patterns and logic found in Nintendo 64 game code.
- Extensive Training: Trained on over 200,000 verified MIPS-to-C pairs to enhance logical reconstruction.
- Context-Aware Decompilation: Utilizes a structured
Technical EnvironmentJSON input to guide type inference, stack frame mapping, and symbol resolution, crucial for accurate IDO-compatible output. - Preprocessing Utility: Requires a specific assembly format, with a Python script provided to clean raw disassembler output.
- Precision Options: Available in BF16 for highest accuracy (requiring 70GB VRAM) and quantized GGUF Q4_K_M for local inference (24GB VRAM).
What Makes This Model Different?
Unlike general-purpose LLMs, this model is not designed for broad coding tasks. Its uniqueness lies in its hyper-specialization for a specific compiler and architecture combination (MIPS + SGI IDO 5.3 for N64). It addresses the challenge of binary matching in reverse engineering by understanding how the IDO compiler translates C types into MIPS instructions, minimizing mismatches that often occur with generic decompilers.
Use Cases & Best Practices
- Assisted Reverse Engineering: Primarily intended as an assistive tool to accelerate the decompilation process for N64 projects, reducing manual effort.
- Human Verification Required: Due to its experimental nature, all output must be verified by a human to ensure correctness and 100% binary matching.
- Strict Prompting Workflow: Requires a precise ChatML format with a
Technical EnvironmentJSON block providing critical metadata (stack frame, global symbols, external functions, jump tables) to prevent hallucinations and ensure accurate C type inference. - Low Temperature Inference: Recommended to use a low inference temperature (0.1-0.2) to produce stable and compilable C code.