M
v0.20.0
mlc-llm v0.20.0 stable (last monolithic-era release; aligns with mlc-…
M
v0.26.dev0: [Refactor] Update TVM runtime integration (#3501)
Use relocated NVTX header under tvm/support/cuda and use NVTXScopedRange from tvm::support instead of tvm::runtime. Link and install tvm_runtime_extra by default so distributed runtime symbols are resolved correctly.
M
v0.19.0
Tag v0.19.0
M
v0.20.dev0: [Device] Defer the CPU detection (#3126)
This PR moves the CPU detection to the last device after checking all GPUs.
M
v0.18.1
Tag v0.18.1
M
v0.19.dev0: [Model] Add support for OLMo architecture (#3046)
This PR add support for OLMo architecture. Additional support: add support for clip-qkv. Test: already tested on android(pixel 4) and cuda(setting tensor_parallel_shrads=2)
M
v0.17.2: [Bench] Update API backend names (#2968)
This PR updates the backend names, introducing one name per backend framework. These backends may refer to the same api endpoint.
M
v0.17.1
Tag v0.17.1
M
v0.18.dev0: Added hermes 3 support (#2886)
added hermes 3 support modified format fixed lint
M
v0.1.dev0
Initial commit --------- Co-authored-by: Hongyi Jin <jinhongyi02@gmail.com> Co-authored-by: Ruihang Lai <ruihangl@cs.cmu.edu> Co-authored-by: Tianqi Chen <tqchen@cmu.edu> Co-authored-by: Junru Shao <junrushao@apache.org> Co-authored-by: Zihao Ye <zhye@cs.washington.edu>