1.0.0

-> Small writeup with charts.

• Remove flash-attention-2 and xformers dependencies
• New attention kernel with online cache quantization, dual input for SWA layers and attention sinks
• New conv1d kernel (removes support/need for causal_conv1d)
• Greatly improved GEMM/GEMV performance on Ampere
• New INT8 GEMV kernel
• New MoE kernel ticket scheduler
• Graph path for all attn/GDN modules
• Fused sampling kernels and reduced GPU sync during sampling
• Switch default codebook to mul1
• Unpin pydantic (monkey-patch formatron
• Add GptOssForCausalLM
• Add NemotronHForCausalLM
• Extend TP support to most models, including Gemma4
• Work around pydantic/formatron compatibiltiy issue, unpin pydantic
• Fix cancelled jobs sometimes hanging in AsyncGenerator
• Many minor optimizations
• Many more bugfixes
• Many QoL improvements
• Faster extension build with more compilation units

Full Changelog: v0.0.43...v1.0.0