56 Pull requests merged by 17 people

• [Runtime] Support KV cache with RoPE extension factor array

  #17294 merged Aug 27, 2024

• [TE][CreatePrimFunc] Fix create reduce block with spatial iter dependent init value

  #17301 merged Aug 26, 2024

• [Relax] Avoid wrapping TupleStructInfo into a Tuple for R.call_tir

  #17243 merged Aug 26, 2024

• [Support] Fix the Read/Write of socket stream

  #17284 merged Aug 26, 2024

• [Relax][Bugfix] Support torch.unbind op and fix bugs for expand && split

  #17292 merged Aug 26, 2024

• [Codegen][WebGPU] LetNode common subexpr override

  #17302 merged Aug 26, 2024

• [Rocm] Fix non-standard rocm path

  #17295 merged Aug 24, 2024

• [Doc] Overview

  #17296 merged Aug 23, 2024

• [Web] Add TVMArgBool to ArgTypeCode

  #17251 merged Aug 23, 2024

• [Doc] IRModule

  #17298 merged Aug 23, 2024

• [Docs] Introduce Relax API and move legacy part to standalone page

  #17286 merged Aug 22, 2024

• [Relax][PyTorch] Add support for torch.tile

  #17291 merged Aug 22, 2024

• [ROCm] hipBLAS integration

  #17290 merged Aug 22, 2024

• [FFI][Runtime] Use TVMValue::v_int64 to represent boolean values

  #17240 merged Aug 22, 2024

• [Cleanup] Remove using namespace tvm::runtime from headers

  #17246 merged Aug 22, 2024

• [Relax][Analysis] Handle recursive functions in CollectVarUsage

  #17224 merged Aug 22, 2024

• [Doc] Quick Start

  #17289 merged Aug 22, 2024

• [Codegen] Emit tir::Let as var assignment explicitly

  #17278 merged Aug 21, 2024

• [Doc] Refactor install docs

  #17287 merged Aug 21, 2024

• [WINDOWS] Compiler options for non x86 targets

  #17260 merged Aug 20, 2024

• [IR] Handle NaN in StructuralEqual and StructuralHash

  #17249 merged Aug 19, 2024

• [Relay][Pytorch] Add support for aten::tile

  #17277 merged Aug 19, 2024

• [CI] Resolve CI compilation failures on MacOSX

  #17271 merged Aug 19, 2024

• [TOPI][ADRENO] Add Group Conv2d texture schedule

  #17274 merged Aug 19, 2024

• [KVCache] Increase coalesce threshold

  #17280 merged Aug 17, 2024

• [Disco] Fix double free of nccl communicator

  #17275 merged Aug 15, 2024

• [CODEGEN][OPENCL] Fix opencl codegen for few ops

  #17273 merged Aug 14, 2024

• [FFI] Re-introduce the boxed primitive values

  #17257 merged Aug 12, 2024

• [Relax][Bugfix] Preserve dtype in ToMixedPrecision for kNever ops

  #17263 merged Aug 11, 2024

• [Disco] Disable splitting nccl communicator in single-group

  #17264 merged Aug 10, 2024

• [CompileBugfix][contrib] meet 'base64.h: No such file or directory' and '‘tvm::runtime::vm::AllocatorType’ has not been declared' while compiling

  #17265 merged Aug 10, 2024

• [DLIGHT][ADRENO] Fix for opencl adreno matmul schedule

  #17259 merged Aug 10, 2024

• [ROCm] Support ROCm 6

  #17256 merged Aug 9, 2024

• [Relax] Add KVCache Interface for Relax NNModule

  #17261 merged Aug 9, 2024

• Replacing unary ops with LookUpTable and Take op to improve performance

  #17214 merged Aug 9, 2024

• [WebGPU] Fix unexpected device lost error when intentional dispose

  #17250 merged Aug 8, 2024

• Revert "[FFI][RUNTIME] Introduce runtime boxed types for int/float/bool"

  #17252 merged Aug 7, 2024

• [Relax] Implement R.ensure_zero_offset and update memory planning for R.view

  #17145 merged Aug 6, 2024

• [Relax] Remove segfault in R.call_tir_inplace validation

  #17242 merged Aug 6, 2024

• [FFI][RUNTIME] Introduce runtime boxed types for int/float/bool

  #16183 merged Aug 5, 2024

• [Unity][Frontend] Add Sqrt Op

  #17228 merged Aug 5, 2024

• [TIR] Validate tir::Buffer axis_separators on construction

  #17219 merged Aug 5, 2024

• [Runtime] Reorganize PagedKVCache attn kernel invocation

  #17237 merged Aug 4, 2024

• [Bugfix][Cutlass] fix cutlass instantiate attention template bugs

  #17229 merged Aug 3, 2024

• [3rdparty] Bump FlashInfer

  #17236 merged Aug 3, 2024

• [Runtime Patch] Add AbortSignal to fetchWithCache in ArtifactCacheTemplate interface

  #17233 merged Aug 2, 2024

• [Relax] FuseTransposeMatmul Pass

  #17234 merged Aug 2, 2024

• [Runtime] Allow aborting fetchWithCache through AbortSignal

  #17227 merged Aug 1, 2024

• [Relax] Fix segfault in rewrite_bindings for MatchCast node

  #17226 merged Aug 1, 2024

• [Relax] Handle presence of R.call_tir in MergeCompositeFunctions

  #17220 merged Jul 31, 2024

• [CI] Reduce logging level when checking if docker image exists

  #17221 merged Jul 31, 2024

• [Transform][Relax] Handle is_group argument in IPC AllReduce

  #17201 merged Jul 31, 2024

• [CI] Update dummy-variable regex for pylint

  #17206 merged Jul 30, 2024

• [Runtime] Allow aborting fetchNDArray through AbortSignal

  #17208 merged Jul 29, 2024

• [Relax] Disable fusion for fetching from the packed params in FuseOps

  #17198 merged Jul 28, 2024

• [Hexagon] Fix LWP assembly handler (predicate register)

  #17204 merged Jul 28, 2024

18 Pull requests opened by 4 people

• [Relax] Remove name requirement from MergeCompositeFunctions

  #17212 opened Jul 29, 2024

• [Relax] Refactor RealizeVDevice to remove in-place mutation

  #17213 opened Jul 29, 2024

• [Relax] Allow `out_sinfo` to be omitted from `R.call_tir`

  #17216 opened Jul 30, 2024

• [Relax] Allow dynamic shape argument to R.reshape

  #17218 opened Jul 30, 2024

• [Relax] Lifted parameters bindings may also be function output

  #17232 opened Aug 1, 2024

• [Draft][FFI] Remove backwards-compatibility unwrapping of IntImm

  #17241 opened Aug 5, 2024

• [Relax][Transform] Handle tuple return in RemoveUnusedOutputs

  #17253 opened Aug 7, 2024

• [CUDA][shared memory allocation]fix 'ptxas error : Entry function 'fu…

  #17267 opened Aug 12, 2024

• [CUDA][onnx-frontend]add deformconv2d converter

  #17268 opened Aug 12, 2024

• [Utility][Relax] Implemented InjectDebugCallback transform

  #17282 opened Aug 19, 2024

• [CI][Windows] Workaround for error in Findzstd.cmake

  #17283 opened Aug 19, 2024

• [Relax] Require correct input/output shapes `R.call_tir`

  #17285 opened Aug 19, 2024

• [Rust] Remove mxnet dependency and re-enable rust example

  #17293 opened Aug 22, 2024

• [Apps] Remove mxnet dependency from /apps/android_camera/models

  #17297 opened Aug 23, 2024

• [Apps] Remove mxnet dependency from /apps/ios_rpc

  #17299 opened Aug 23, 2024

• [Relax][PyTorch] Add support for torchvision.ops.stochastic_depth

  #17300 opened Aug 23, 2024

• [Relax][PyTorch] Add support for torch.repeat

  #17304 opened Aug 26, 2024

• [Python][Relax] Rotary positional embedding scaling

  #17305 opened Aug 27, 2024

15 Issues closed by 7 people

• [Bug] [Relax] Variable was used before its definition

  #17222 closed Aug 22, 2024

• [Bug] Android build faild for Llama-2-7b-chat-hf

  #16275 closed Aug 13, 2024

• [Bug] [Relax] Inconsistent buffers compute and lv mapped to the same relax var

  #17254 closed Aug 12, 2024

• target.build.llvm is not enabled

  #17238 closed Aug 9, 2024

• [Bug] The entry value of attr should be integer. However Array is got

  #17255 closed Aug 9, 2024

• [Bug] [Relax] Hug inference performance difference on cuda and cpu

  #17248 closed Aug 7, 2024

• [Bug] output different results when inference twice with the same inputs

  #17244 closed Aug 6, 2024

• [Bug] [Relax] InternalError: Check failed: type_code_ == kTVMObjectHandle expected Object but got int

  #17235 closed Aug 5, 2024

• [Bug] [Relax] InternalError: Check failed: last_sep < self->shape.size() Last output axis must contain at least one input axis

  #17215 closed Aug 5, 2024

• [Bug] [Relax] Expect a not null value of IRModule

  #17230 closed Aug 1, 2024

• fatal error: string_view: No such file or directory

  #17209 closed Aug 1, 2024

• [Relax][Bug] Segmentation fault when using the MergeCompositeFunctions transform

  #17120 closed Jul 31, 2024

• [Bug] CreateModulePass has an inconsistent para_format between BindSymbolicVars and other pass

  #17225 closed Jul 31, 2024

• [CI Problem] Missing AWS Key for S3 file cache storage

  #17019 closed Jul 31, 2024

• [Bug] [Relax] LiftTransformParams lead to wrong inference results

  #17207 closed Jul 28, 2024

17 Issues opened by 8 people

• [Bug] (mach-o file, but is an incompatible architecture (have 'x86_64', need 'arm64'))

  #17303 opened Aug 24, 2024

• [Flaky Test] `tests/python/autotvm/test_autotvm_droplet_tuner.py::test_multi_filter`

  #17288 opened Aug 20, 2024

• issue deploying compiled model to runtime : "loader of that name is not registered"

  #17281 opened Aug 19, 2024

• [Bug] Tuning Metal fails: Could not find any valid schedule for task

  #17279 opened Aug 17, 2024

• [Bug] Precision issue when working with floating point constants (FloatImm)

  #17276 opened Aug 15, 2024

• [Bug] give inconsistent inference results after use ReorderTakeAfterMatmul

  #17270 opened Aug 12, 2024

• [Bug] AssertionError in the LazyTransformParams

  #17269 opened Aug 12, 2024

• [Bug] VTA RPC Error

  #17266 opened Aug 10, 2024

• [Bug] TVM Cross Compilation with QNX for arm

  #17262 opened Aug 9, 2024

• [Bug] RemoveUnusedOutputs give unexpected results

  #17247 opened Aug 6, 2024

• [Bug] Cannot add loops on top of the root block

  #17245 opened Aug 6, 2024

• [Bug] [Relax] Segfault error when parse the Relax IR

  #17239 opened Aug 3, 2024

• [Bug] [Relax] Check failed: (it != this->var_arg_map_.end()) is false: Var is not defined

  #17231 opened Aug 1, 2024

• [Bug] [Relax] Argument type mismatch: expected R.Tensor, given R.Tuple

  #17223 opened Jul 31, 2024

• [Bug] [Relax] VMBuiltinLower expects bound value to be a ShapeExpr

  #17217 opened Jul 30, 2024

• [Bug] [Relax] Missing IR structure checking and correction

  #17211 opened Jul 28, 2024

• [Bug] [Relax] InternalError: Check failed: (it_group != obj2group_.end()) is false

  #17210 opened Jul 28, 2024

15 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• [TIR] Enhance Lower cross thread Pass

  #17133 commented on Aug 13, 2024 • 16 new comments

• [Bug] MicroTVM Compilation fails for ARM Cortex-M4 ,with CMSIS-NN=false

  #17136 commented on Jul 30, 2024 • 0 new comments

• [Bug] [Relax] cannot remove the hint_on_device

  #17205 commented on Jul 30, 2024 • 0 new comments

• [Relax][Bug] CodeGenVM cannot handle this intrinsic tensor_to_shape

  #17175 commented on Jul 31, 2024 • 0 new comments

• [Bug] Error opening FastRPC channel

  #17195 commented on Aug 1, 2024 • 0 new comments

• [Relax][Bug] Cannot find PackedFunc tir_zeros

  #17176 commented on Aug 1, 2024 • 0 new comments

• [Bug] Conflicting code is generated in lib0.c causing conflicting types for tvmgen_default_run and tvmgen_default__tvm_main__

  #17194 commented on Aug 7, 2024 • 0 new comments

• [Bug] tvm/src/contrib/torch/tvm_module_wrapper/RuntimeModuleWrapperTVM.cc:32:10: fatal error: ../../support/base64.h: No such file or directory

  #17164 commented on Aug 11, 2024 • 0 new comments

• TVMError: Binary was created using cuda but a loader of that name is not registered

  #13595 commented on Aug 19, 2024 • 0 new comments

• [Bug] Graph optimization model compilation error involving `Pad` operator

  #16898 commented on Aug 21, 2024 • 0 new comments

• [TFLite][Frontend] Support quantized ARG_MIN

  #15922 commented on Jul 29, 2024 • 0 new comments

• [Relax][Bugfix] FCallPacked not checked in CodegenVMTIR

  #17073 commented on Jul 30, 2024 • 0 new comments

• [Relax] Apply DefaultGPUSchedule() in default build pipeline

  #17108 commented on Aug 2, 2024 • 0 new comments

• [AutoScheduler] Add Dynamic Gradient Descent Search Algorithm for Auto-Tuning

  #17126 commented on Aug 12, 2024 • 0 new comments

• [Relax] Update GlobalVar name in AttachGlobalSymbol

  #17202 commented on Jul 29, 2024 • 0 new comments