GPT-QModel v7.3.1

What's Changed

• [MODEL] support deepsekk_vl_v2 by @ZX-ModelCloud in #2938
• [MODEL] support deepseek_ocr2 by @ZX-ModelCloud in #2940
• Fix O(N^2) state-dict collection that stalled saves of large MoE models for offload_to_disk=True by @avtc in #2945
• Fix ExLlamaV2 kernels on Windows: link cublas.lib and enable PLATFORM.WIN32 by @xXpeira12 in #2944
• [MODEL] support deepseek_vl by @ZX-ModelCloud in #2942
• [FIX] deepseek_ocr2 supports OFFLOAD_TO_DISK=True by @ZX-ModelCloud in #2941
• [FIX] GLM-5.2 quantization error by @ZX-ModelCloud in #2947
• support ModelOpt NVFP4 dequantization by @ZX-ModelCloud in #2948
• fix(triton): bound-check checkpoint g_idx to prevent OOB dequant read (#2949) by @Anai-Guo in #2950
• [MODEL] support nemotron_h_puzzle by @ZX-ModelCloud in #2951
• [FIX] load_gguf_checkpoint() compatible with the latest transformers by @ZX-ModelCloud in #2952
• [FIX] The "Latest News" section of the README by @ZX-ModelCloud in #2953
• Update Evalution dependency version in pyproject.toml by @Qubitium in #2954
• Update version.py by @Qubitium in #2955
• Bump version from 7.3.0 to 7.3.1 by @Qubitium in #2956

New Contributors

• @xXpeira12 made their first contribution in #2944

Full Changelog: v7.2.0...v7.3.1