renxida

            def log_tensor_stats(tensor, name="Tensor"):
                logger.info(f"{name} stats:")
                logger.info(f"  type :{type(tensor)}")
                logger.info(f"  dtype:{tensor.dtype}")
                from scipy import stats
                import array
                import numpy as np
                fp16hack = False
                if isinstance(tensor, sfnp.device_array):

renxida

#!/bin/bash

# Define the range of Python versions to test
versions=("3.12" "3.11" "3.10" "3.9" "3.8" "3.7")

# Function to update pyproject.toml
update_pyproject() {
    sed -i 's/python = "^[0-9.]*"/python = "^'"$1"'"/' pyproject.toml
    rm -f poetry.lock
}

renxida

#!/bin/bash

# Define the range of Python versions to test
versions=("3.12" "3.11" "3.10" "3.9" "3.8" "3.7")

# Function to update pyproject.toml
update_pyproject() {
    sed -i 's/python = "^[0-9.]*"/python = "^'"$1"'"/' pyproject.toml
}

renxida

{ config, pkgs, lib, ... }:

{
  # Import disko module
  imports = [
    (fetchTarball "https://github.com/nix-community/disko/archive/master.tar.gz")
  ];

  # Bootloader configuration
  boot.loader.grub.enable = true;

renxida

module {
  func.func @torch_jit(%arg0: !torch.vtensor<[3,300,400],f32>, %arg1: !torch.vtensor<[3,500,400],f32>) -> (!torch.vtensor<[?,4],f32>, !torch.vtensor<[?,?,?],f32>, !torch.vtensor<[?,?],f32>, !torch.vtensor<[?],si64>, !torch.vtensor<[?],f32>, !torch.vtensor<[?,4],f32>, !torch.vtensor<[?,?,?],f32>, !torch.vtensor<[?,?],f32>, !torch.vtensor<[?],si64>, !torch.vtensor<[?],f32>) attributes {torch.onnx_meta.ir_version = 8 : si64, torch.onnx_meta.opset_version = 17 : si64, torch.onnx_meta.producer_name = "pytorch", torch.onnx_meta.producer_version = "1.13.1"} {
    %none = torch.constant.none
    %0 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<_> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
    %1 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__1> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
    %2 = torch.operator "onnx.Constant"() {torch.onnx.value = dense_resource<__2> : tensor<f32>} : () -> !torch.vtensor<[],f32> 
    %3 = torch.operator "onnx.Constant"

renxida

/home/azureuser/iree-build/tools/iree-compile: /home/azureuser/miniconda/lib/libtinfo.so.6: no version information available (required by /home/azureuser/iree-build/lib/libIREECompiler.so)
deit-small-distilled-patch16-224.default.pytorch.torch.mlir:1029:12: error: failed to legalize operation 'torch.aten.squeeze' that was explicitly marked illegal
    %999 = torch.aten.squeeze %998 : !torch.vtensor<[1,1,384],f32> -> !torch.vtensor<[1,384],f32>
           ^
deit-small-distilled-patch16-224.default.pytorch.torch.mlir:1029:12: note: see current operation: %13245 = "torch.aten.squeeze"(%13244) : (!torch.vtensor<[1,1,384],f32>) -> !torch.vtensor<[1,384],f32>
iree-compile: /home/azureuser/iree/third_party/llvm-project/mlir/include/mlir/IR/UseDefLists.h:198: mlir::IRObjectWithUseList<mlir::OpOperand>::~IRObjectWithUseList() [OperandType = mlir::OpOperand]: Assertion `use_empty() && "Cannot destroy a value that still has uses!"' failed.
Please report issues to https://github.com/openxla/iree/issues and include the cr

renxida

# Description: This script is used to test the model deit-small-distilled-patch16-224.default.pytorch.torch.stripped.mlir


# run original model and print ir after failure
/home/azureuser/iree-build/tools/iree-compile --iree-input-demote-i64-to-i32 --iree-hal-target-backends=llvm-cpu  stripped/deit-small-distilled-patch16-224.default.pytorch.torch.stripped.mlir -o deit-small-distilled-patch16-224.default.stripped.vmfb   --mlir-print-debuginfo         --mlir-print-ir-after-failure  |& gh gist create - -d "native_layer_norm ir dump after failure"

# run again with --debug and grep for `(tensor<198xf32>) -> tensor<?x198xf32>` and pass names
# grep patterns:
# `(tensor<198xf32>) -> tensor<?x198xf32>`
# `IR Dump After`

renxida

OVERVIEW: MLIR modular optimizer driver

Available Dialects: builtin, chlo, complex, func, linalg, memref, ml_program, quant, scf, sparse_tensor, stablehlo, tensor, tm_tensor, torch, torch_c, tosa, vhlo
USAGE: torch-mlir-opt [options] <input file>

OPTIONS:

Color Options:

  --color                                                  - Use colors in output (default=autodetect)

renxida

OVERVIEW: IREE compilation driver

USAGE: iree-compile [options] <input file or '-' for stdin>

OPTIONS:

CUDA HAL Target:

  --iree-hal-cuda-dump-ptx                                                   - Dump ptx to the debug stream.
  --iree-hal-cuda-llvm-target-arch=<string>                                  - LLVM target chip.

renxida

We suffer more in imagination than in reality - - Seneca

Situation / Action

Right now

Define fears

• fear:
  • prevent by:
  • repair by:
• fear: