NEXS Qwen3 32b Prover Lora

NEXS Qwen3 32b Prover Lora is a LoRA adapter designed for use with the Qwen3-32B base model. According to the available documentation, this adapter is extracted using mergekit from Goedel-LM/Goedel-Prover-V2-32B and is prepared for integration with vLLM, with sanitization steps applied to ensure compatibility for serving. The adapter is provided in bf16 format and features a rank-128 configuration. The tool can be used in conjunction with the PEFT library, as demonstrated by the provided code example, which shows how to load the base Qwen3-32B model and apply the NEXS Qwen3-32b Prover Lora adapter. Instructions are available for using the adapter with various libraries, inference providers, notebooks, and local applications, including Google Colab and Kaggle. The model card indicates that sanitization was applied to address full-rank modules_to_save tensors, such as embed_tokens, lm_head, and norm layers, in order to facilitate proper operation with vLLM's LoRA runtime. No details are given regarding the specific tasks or domains for which this adapter is optimized, nor are there explicit statements about its intended user base, licensing, or pricing. The evidence does not mention any integrations beyond those with PEFT and vLLM, nor does it describe any performance metrics or use cases. As such, the description is limited to the technical aspects and compatibility features directly referenced in the documentation.