tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
I try to quantizae llava-v1.6-34b ``` python3 -m mlc_llm.build --model /data/models/mlc/dist/models/llava-v1.6-34b \ --quantization q4f16_ft \ --target cuda \ --use-cuda-graph \ --use-flash-attn-mqa \ --sep-embed \ --max-seq-len 256 --artifact-path /data/models/mlc/dist/llava-v1.6-34b/ctx256 \ --use-safetensors...
While running `AProfileParser.DefaultSVESupportSVESupport` and `AProfileParser.DefaultFP16Support` tests on aarch64, I get the following warnings: ``` /home/abuild/rpmbuild/BUILD/tvm-0.19.0-build/tvm-0.19.0/include/tvm/target/target_kind.h:353: Warning: set_attrs_preprocessor is deprecated please use set_target_parser instead /home/abuild/rpmbuild/BUILD/tvm-0.19.0-build/tvm-0.19.0/include/tvm/target/target_kind.h:353: Warning: set_attrs_preprocessor is deprecated please use...
### Environment Information OS: Ubuntu 20.04 LTS Python: 3.10.4 TVM: v0.18.0 built from source with CPU only, no GPU usage in place ### Steps to Reproduce: ```python import tvm from...
### Expected behavior Successful Compilation or a reason for why the compilation target is invalid ### Actual behavior Segmentation Fault ### Environment OS: Ubuntu 20.04 LTS Python: 3.10.4 TVM: v0.18.0...
When applying the relax.transform.InlinePrivateFunctions() optimization to a Relax module using both Sequential and direct application methods, the resulting module structures are inconsistent. Additionally, when using relax.build() after applying the transformation...
### Actual behavior ``` @I.ir_module class Module: @R.function def main(q: R.Tensor((4, 16, 32, 8), dtype="float32"), k: R.Tensor((4, 8, 32, 8), dtype="float32"), v: R.Tensor((4, 8, 32, 16), dtype="float32"), bias: R.Tensor((4, 32,...
I am trying to build Apache TVM library for MIPs platform using steps explained on [this page](https://tvm.apache.org/docs/how_to/deploy/index.html#cross-compile-for-risc-v) for RISC. Following are the steps I followed: $ sudo apt-get update $...
## Steps to Reproduce: ```python import tvm from tvm import tir from tvm.tir.analysis.analysis import verify_well_formed, verify_memory from tvm.script import tir as T @T.prim_func(private=True) def main(h1: T.handle, h2: T.handle, h3: T.handle,...
It is currently not possible to access a `BlockBuilder`'s `Analyzer` instance from Python code, which would be useful e.g. when converting from front-end representations to Relax. This PR addresses that...
