LLM4Decompile 22B v2
binary/asm → C
A 22B decompilation task model (binary/assembly → C), extending the LLM4Decompile line to a larger base. A task model, not a general chat model.
Source: LLM4Binary/llm4decompile-22b-v2 on HuggingFace
- +decompilation
- +binary → C
- +Ghidra refinement
Trained for assembly→C decompilation (refining Ghidra output) by LLM4Binary; the 22B v2 extends the LLM4Decompile line to a larger base. A task model, not a general chat model.
- ›Decompile / refine binary code into human-readable C
- ›Refine Ghidra pseudo-code into C
- ›Function-level decompilation across optimization levels
LLM4Decompile 22B v2 (LLM4Binary/llm4decompile-22b-v2) is the 22B member of the LLM4Decompile family — task models that turn binary/assembly into human-readable C and refine decompiler (Ghidra) output. It extends the 6.7B model this catalog already lists to a larger base. It is a specialized task model, not a general chat model, and is distributed under the MIT license. Listed with a curated card; not yet provisioned on our serving.
Trained for assembly→C decompilation across optimization levels (O0–O3) by LLM4Binary, following the LLM4Decompile approach (refining decompiler output into compilable C). The 22B v2 is the larger member of the line. Specific dataset details for the 22B build are treated as unconfirmed here.
- +Decompile / refine binary or assembly code into readable C
- +Refine Ghidra headless pseudo-code into C
- +Function-level decompilation across GCC optimization levels
- +Reverse-engineering and malware-analysis support (task, not chat)
- −A task model, not a chat model: it expects decompilation inputs, not open-ended conversation.
- −Documents a 4096-token context, which constrains how much code fits in a single pass.
- −No GGUF mirror / Ollama tag is confirmed for this repo yet, so the intended Q4_K_M serving path is not yet validated here.
- −Decompiled C can be incorrect or non-re-executable and must be reviewed.
- −Not currently served here — listed with a card, not provisioned.
The maker’s own snippet, from the model card — it downloads the weights and runs them on your hardware. Kept here because reproducing a result independently is the point, not because you need it to use the model.
Paste decompiler output (Ghidra/objdump pseudocode) and ask for clean C: Rewrite this decompiled function as readable C with named variables, and explain what it does: <paste the pseudocode here>
No measurements published for this version yet.
Baseline is the strongest general-purpose model we could run on the same suite, same setup, same day. The control row tells you what the other rows are worth.
Start with LLM4Decompile 22B v2
A confirmed email account includes 30 free messages a month.