GPT-QModel v7.4.0

What's Changed

• feat(models): add GLM-5 Next quantization support by @ZX-ModelCloud in #3045
• bench(marlin): add tile-padding performance by @ZX-ModelCloud in #3046
• [MODEL] support apertus1p5 by @ZX-ModelCloud in #3047
• fix: normalize multimodal calibration inputs by @ZX-ModelCloud in #3049
• fix: improve quantization and runtime correctness by @ZX-ModelCloud in #3055
• Shared-input Hessian dedup: plan metadata, E2E validation, and telemetry by @Qubitium in #3052
• support lm-head/embed requant by @ZX-ModelCloud in #3050
• fix: isolate JIT extension cache fingerprint from unrelated kernel changes by @Qubitium in #3056
• Sync native GGUF support with current upstream types by @Qubitium in #3058
• feat: checkpointing and resuming quantization process from last checkpoint by @okdshin in #3057
• Update version.py by @Qubitium in #3060
• Fix Triton patch for pre-existing autotuner instances by @Qubitium in #3062
• [MODEL] support ouro and spark2_5 by @ZX-ModelCloud in #3061
• Sync README release news and model support for 7.4.0 by @Qubitium in #3063
• Document quantization checkpoint and resume workflows by @Qubitium in #3064

New Contributors

• @okdshin made their first contribution in #3057

Full Changelog: v7.3.6...v7.4.0