Hidet v0.2.3

What's Changed

• [Version] Bump version to v0.2.3.dev by @yaoyaoding in #144
• [Workflow] Update workflow to use the stable version of pytorch by @yaoyaoding in #145
• [Operator] Resolve matmul to batch_matmul when lower than sm80 by @hjjq in #146
• [Dynamo] non-linear operator support + tests by @AndreSlavescu in #143
• Remove tutorial msg by @LDY1998 in #149
• [BUG] Conversion compile issue by @xinli-git in #150
• [Dynamo] Fix dynamo tests and dump graph IR by @xinli-git in #153
• [CI] Benchmark periodically by @yaoyaoding in #155
• [CI] Update bench script by @yaoyaoding in #156
• [CI] Add more env information to benchmark script by @yaoyaoding in #157
• [CI] Remove benchmark workflow, but run it in dedicated server by @yaoyaoding in #159
• [CI] Update benchmark script by @yaoyaoding in #160
• [CI] Change the search space in benchmark script from 0 to 2 by @yaoyaoding in #161
• [CI] Update benchmark script by @yaoyaoding in #162
• [CI] Update benchmark scripts by @yaoyaoding in #163
• [IR][Pass] Refactor the fusion implementation by @yaoyaoding in #164
• [Dynamo] Add operator support to run UNet2DConditionModel from diffusers by @xinli-git in #151
• [IR][Dynamic Shape] Enhance the Tensor Program IR to support dynamic shape by @yaoyaoding in #165
• [Operator] Allow matmul_f16 fuse epilogue by @yaoyaoding in #167
• [CI] Update benchmark script by @yaoyaoding in #168
• [CUDA] Lazy initializing cuda context by @yaoyaoding in #169
• [Fixbug] Allow one backend fail in benchmark script by @yaoyaoding in #170
• [Fixbug] Use auto-scheduler for fp64 reduction by @yaoyaoding in #171
• [Operator] Add gather operator and torch.zeros, torch.neg mapping by @yaoyaoding in #174
• [CI] Update benchmark script by @yaoyaoding in #179
• [Fixbug] Add _stacklevel to pytorch softmax mapping by @yaoyaoding in #178
• [IR] Add unroll pragma for loop statement by @yaoyaoding in #180
• [Operator] Flash Attention by @hjjq in #175
• [Fixbug] Fix a bug in the mapping from device to its memory pool by @yaoyaoding in #181
• [Dynamo] Small enchancements for graph dump ir and task arguments by @xinli-git in #172
• [Docs] Update install instruction by @hjjq in #182
• change norm to use smaller inputs to reduce running time by @xinli-git in #185
• [IR] Add explicit unroll by @yaoyaoding in #184
• [Runtime] Allow pass torch tensor to PackedFunc directly by @yaoyaoding in #183
• Refactor codegen to separate GPU/CPU code generation by @LDY1998 in #176
• [Pass] Support inline function by @yaoyaoding in #186

New Contributors

• @LDY1998 made their first contribution in #149

Full Changelog: v0.2.2...v0.2.3