← Back to Leaderboard

The AI CUDA Engineer 👷

40_LayerNorm • optimized_layernorm_streamed_base

Level 1 • Task 40
import torch
import torch.nn as nn
import torch.nn.functional as F


def module_fn(
    x: torch.Tensor, weight: torch.Tensor, bias: torch.Tensor, eps: float = 1e-5
) -> torch.Tensor:
    """
    Functional implementation of LayerNorm.

    Args:
        x (torch.Tensor): Input tensor of shape (*, normalized_shape).
        weight (torch.Tensor): Weight tensor of shape (normalized_shape).
        bias (torch.Tensor): Bias tensor of shape (normalized_shape).
        eps (float): Epsilon parameter for numerical stability.

    Returns:
        torch.Tensor: Output tensor with Layer Normalization applied, same shape as input.
    """
    # Get the normalized shape from the weight tensor
    normalized_shape = tuple(x.shape[-len(weight.shape) :])
    return F.layer_norm(
        x, normalized_shape=normalized_shape, weight=weight, bias=bias, eps=eps
    )


class Model(nn.Module):
    """
    Simple model that performs Layer Normalization.
    """

    def __init__(self, normalized_shape: tuple):
        """
        Initializes the LayerNorm layer parameters.

        Args:
            normalized_shape (tuple): Shape of the input tensor to be normalized.
        """
        super(Model, self).__init__()
        self.weight = nn.Parameter(torch.ones(normalized_shape))
        self.bias = nn.Parameter(torch.zeros(normalized_shape))

    def forward(self, x: torch.Tensor, fn=module_fn) -> torch.Tensor:
        """
        Applies Layer Normalization to the input tensor.

        Args:
            x (torch.Tensor): Input tensor of shape (*, normalized_shape).
            fn: Function to apply (defaults to module_fn)

        Returns:
            torch.Tensor: Output tensor with Layer Normalization applied, same shape as input.
        """
        return fn(x, self.weight, self.bias)


batch_size = 16
features = 64
dim1 = 256
dim2 = 256


def get_inputs():
    x = torch.randn(batch_size, features, dim1, dim2)
    return [x]


def get_init_inputs():
    return [(features, dim1, dim2)]

import torch
import torch.nn as nn

class Model(nn.Module):
    """
    Simple model that performs Layer Normalization.
    """
    def __init__(self, normalized_shape: tuple):
        """
        Initializes the LayerNorm layer.

        Args:
            normalized_shape (tuple): Shape of the input tensor to be normalized.
        """
        super(Model, self).__init__()
        self.ln = nn.LayerNorm(normalized_shape=normalized_shape)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """
        Applies Layer Normalization to the input tensor.

        Args:
            x (torch.Tensor): Input tensor of shape (*, normalized_shape).

        Returns:
            torch.Tensor: Output tensor with Layer Normalization applied, same shape as input.
        """
        return self.ln(x)

batch_size = 16
features = 64
dim1 = 256
dim2 = 256

def get_inputs():
    x = torch.randn(batch_size, features, dim1, dim2)
    return [x]

def get_init_inputs():
    return [(features, dim1, dim2)]
Download Evaluation Download PyTorch Download CUDA Download Profiles

Kernel Information

Operation Name 40_LayerNorm
Level ID 1
Task ID 40
Kernel Name optimized_layernorm_streamed_base
CUDA Speedup (Native) 8.602x
CUDA Speedup (Compile) 0.701x
CUDA Runtime 0.942 ms
PyTorch Runtime (Native) 8.103 ms
PyTorch Runtime (Compile) 0.661 ms
Correct True
Max Diff (vs. Reference) 0.000000
Model bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0
Temperature 1.00
View Experiment Progress Details

Related Kernels (Level 1, Task 40 • 40_LayerNorm)

Rank Kernel Name Runtime (ms) Speedup Native Speedup Compile
🥇 optimized_layernorm_streamed_base 0.94 8.60 0.70
🥈 layernorm_unrolled_base_base 1.01 7.99 0.65
🥉 layernorm_forward_optimized_base 1.02 7.94 0.65
4 layernorm_modular_base_base 1.02 7.91 0.65
5 layernorm_aligned_base_base 1.03 7.90 0.64
6 layernorm_forward_optimized_base 1.03 7.89 0.64
7 optimized_layernorm_2d_base 1.12 7.23 0.59
8 layernorm_vector8_aligned_base_base 1.24 6.53 0.53
9 optimized_layernorm_unrolled_base 1.63 4.96 0.40
10 optimized_layernorm_unrolled_edit_1 1.64 4.94 0.40
11 layernorm_ldg_optimized_base_edit_1 2.22 3.65 0.30
12 40_LayerNorm_stride_loops_edit_1 2.24 3.62 0.30
13 layernorm_hybrid_optimized_base 2.24 3.62 0.29
14 layernorm_ldg_optimized_base_base 2.24 3.61 0.29
15 layernorm_coalesced_base 3.26 2.49 0.20
16 layernorm_2d_indexing_base 3.26 2.48 0.20
17 optimized_layernorm_base 3.27 2.48 0.20
18 warp_shfl_layernorm_base 3.28 2.47 0.20
19 layernorm_forward_opt_base 3.28 2.47 0.20
20 layernorm_uniform_control_flow_base 3.28 2.47 0.20 
#include <torch/extension.h>
#include <cuda.h>
#include <cuda_runtime.h>
#include <vector>
#include <ATen/AccumulateType.h>

static const int NUM_STREAMS = 4;
static cudaStream_t streams[NUM_STREAMS];
static bool streams_created = false;

void create_streams() {
    if (!streams_created) {
        for (int i = 0; i < NUM_STREAMS; i++) {
            cudaStreamCreate(&streams[i]);
        }
        streams_created = true;
    }
}

void destroy_streams() {
    if (streams_created) {
        for (int i = 0; i < NUM_STREAMS; i++) {
            cudaStreamDestroy(streams[i]);
        }
        streams_created = false;
    }
}

template <typename scalar_t>
__global__ void layernorm_streamed_kernel(
    const scalar_t* __restrict__ input,
    const scalar_t* __restrict__ weight,
    const scalar_t* __restrict__ bias,
    const float eps,
    scalar_t* __restrict__ output,
    const int normalized_size,
    const int chunk_size,
    const int chunk_offset) {

    using accscalar_t = at::acc_type<scalar_t, true>;
    
    const int tidx = threadIdx.x;
    const int tidy = threadIdx.y;
    const int instance_idx = blockIdx.x + chunk_offset;
    
    extern __shared__ char smem[];
    accscalar_t* s_sum = reinterpret_cast<accscalar_t*>(smem);
    accscalar_t* s_sum_sq = s_sum + blockDim.x * blockDim.y;
    
    const scalar_t* in_ptr = input + instance_idx * normalized_size;
    scalar_t* out_ptr = output + instance_idx * normalized_size;
    
    const int thread_stride = blockDim.x * blockDim.y;
    const int thread_id = tidy * blockDim.x + tidx;
    
    accscalar_t local_sum = 0;
    accscalar_t local_sum_sq = 0;
    
    #pragma unroll 8
    for (int idx = thread_id; idx < normalized_size; idx += thread_stride) {
        accscalar_t val = static_cast<accscalar_t>(in_ptr[idx]);
        local_sum += val;
        local_sum_sq += val * val;
    }
    
    s_sum[thread_id] = local_sum;
    s_sum_sq[thread_id] = local_sum_sq;
    __syncthreads();
    
    if (thread_id < 32) {
        accscalar_t warp_sum = 0;
        accscalar_t warp_sum_sq = 0;
        
        #pragma unroll
        for (int i = thread_id; i < thread_stride; i += 32) {
            warp_sum += s_sum[i];
            warp_sum_sq += s_sum_sq[i];
        }
        
        #pragma unroll
        for (int offset = 16; offset > 0; offset /= 2) {
            warp_sum += __shfl_down_sync(0xffffffff, warp_sum, offset);
            warp_sum_sq += __shfl_down_sync(0xffffffff, warp_sum_sq, offset);
        }
        
        if (thread_id == 0) {
            s_sum[0] = warp_sum;
            s_sum_sq[0] = warp_sum_sq;
        }
    }
    __syncthreads();
    
    __shared__ accscalar_t mean, inv_std;
    if (thread_id == 0) {
        mean = s_sum[0] / normalized_size;
        accscalar_t variance = (s_sum_sq[0] / normalized_size) - (mean * mean);
        inv_std = rsqrt(variance + static_cast<accscalar_t>(eps));
    }
    __syncthreads();
    
    #pragma unroll 8
    for (int idx = thread_id; idx < normalized_size; idx += thread_stride) {
        accscalar_t val = static_cast<accscalar_t>(in_ptr[idx]);
        accscalar_t normalized = (val - mean) * inv_std;
        out_ptr[idx] = static_cast<scalar_t>(
            normalized * static_cast<accscalar_t>(weight[idx]) + 
            static_cast<accscalar_t>(bias[idx]));
    }
}

torch::Tensor layernorm_forward(torch::Tensor x, torch::Tensor weight, torch::Tensor bias, double eps = 1e-5) {
    create_streams();
    
    auto output = torch::empty_like(x);
    
    const int normalized_size = weight.numel();
    const int outer_size = x.numel() / normalized_size;
    const int chunk_size = (outer_size + NUM_STREAMS - 1) / NUM_STREAMS;
    
    const dim3 threads(32, 32);
    const int shared_mem_size = threads.x * threads.y * 2 * sizeof(float);
    
    AT_DISPATCH_FLOATING_TYPES(x.scalar_type(), "layernorm_forward_cuda", ([&] {
        for (int i = 0; i < NUM_STREAMS; i++) {
            int stream_chunk_size = std::min(chunk_size, outer_size - i * chunk_size);
            if (stream_chunk_size <= 0) break;
            
            const dim3 blocks(stream_chunk_size);
            
            layernorm_streamed_kernel<scalar_t><<<blocks, threads, shared_mem_size, streams[i]>>>(
                x.data_ptr<scalar_t>(),
                weight.data_ptr<scalar_t>(),
                bias.data_ptr<scalar_t>(),
                static_cast<float>(eps),
                output.data_ptr<scalar_t>(),
                normalized_size,
                chunk_size,
                i * chunk_size);
        }
    }));
    
    // Synchronize all streams before returning
    for (int i = 0; i < NUM_STREAMS; i++) {
        cudaStreamSynchronize(streams[i]);
    }
    
    return output;
}

PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
    m.def("forward", &layernorm_forward, "LayerNorm forward (CUDA)",
          py::arg("x"), py::arg("weight"), py::arg("bias"), py::arg("eps") = 1e-5);
    // Add cleanup function for streams
    m.def("cleanup", &destroy_streams, "Cleanup CUDA streams");
}
Performance Metrics
Metric Value Unit Variance Samples
Executed Ipc Active 1.570 inst/cycle 0.000 5
Executed Ipc Elapsed 0.050 inst/cycle 0.000 5
Issue Slots Busy 39.322 % 0.001 5
Issued Ipc Active 1.570 inst/cycle 0.000 5
SM Busy 48.172 % 0.001 5
Memory Throughput 238377437873.598 byte/second 51761307005635552.000 5
Mem Busy 4.296 % 0.000 5
Max Bandwidth 7.112 % 0.000 5
L1/TEX Hit Rate 0.000 % 0.000 5
L2 Hit Rate 36.362 % 0.637 5
Mem Pipes Busy 1.342 % 0.000 5
Warp Cycles Per Issued Instruction 20.264 cycle 0.001 5
Warp Cycles Per Executed Instruction 20.268 cycle 0.001 5
Avg. Active Threads Per Warp 32.000 0.000 5
Avg. Not Predicated Off Threads Per Warp 31.990 0.000 5
Max Active Clusters 0.000 cluster 0.000 5
Max Cluster Size 8.000 block 0.000 5
Overall GPU Occupancy 0.000 % 0.000 5
Cluster Occupancy 0.000 % 0.000 5
Block Limit SM 32.000 block 0.000 5
Block Limit Registers 2.000 block 0.000 5
Block Limit Shared Mem 3.000 block 0.000 5
Block Limit Warps 2.000 block 0.000 5
Theoretical Active Warps per SM 64.000 warp 0.000 5
Theoretical Occupancy 100.000 % 0.000 5
Achieved Occupancy 49.832 % 0.001 5
Achieved Active Warps Per SM 31.892 warp 0.000 5
Analysis Rules
Rule Description
INF HighPipeUtilization FMA is the highest-utilized pipeline (31.8%) based on active cycles, taking into account the rates of its different instructions. It executes 32-bit floating point (FADD, FMUL, FMAD, ...) and integer (IMUL, IMAD) operations. It is well-utilized, but should not be a bottleneck.
INF CPIStall Check the Warp Stall Sampling (All Cycles) table for the top stall locations in your source based on sampling data. The Kernel Profiling Guide (https://docs.nvidia.com/nsight-compute/ProfilingGuide/index.html#metrics-reference) provides more details on each stall reason.
WRN Occupancy This kernel's theoretical occupancy is not impacted by any block limit. The difference between calculated theoretical (100.0%) and measured achieved occupancy (49.8%) can be the result of warp scheduling overheads or workload imbalances during the kernel execution. Load imbalances can occur between warps within a block as well as across blocks of the same kernel. See the CUDA Best Practices Guide (https://docs.nvidia.com/cuda/cuda-c-best-practices-guide/index.html#occupancy) for more details on optimizing occupancy.
Operation / Metric Value Unit
aten::randn
CPU Time 313105.33 μs
Device Time 0.00 μs
Self CPU Time 123.12 μs
Self Device Time 0.00 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
aten::to
CPU Time 338148.17 μs
Device Time 30872.87 μs
Self CPU Time 65.26 μs
Self Device Time 0.00 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
aten::_to_copy
CPU Time 338082.91 μs
Device Time 30872.87 μs
Self CPU Time 141.67 μs
Self Device Time 0.00 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
aten::empty_strided
CPU Time 319876.45 μs
Device Time 0.00 μs
Self CPU Time 12878.25 μs
Self Device Time 0.00 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
cudaStreamSynchronize
CPU Time 4394612.78 μs
Device Time 38565.01 μs
Self CPU Time 4394612.78 μs
Self Device Time 38565.01 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
aten::fill_
CPU Time 70794.96 μs
Device Time 350711.66 μs
Self CPU Time 45858.29 μs
Self Device Time 350711.66 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
aten::zero_
CPU Time 61590.41 μs
Device Time 350711.66 μs
Self CPU Time 10554.48 μs
Self Device Time 0.00 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
void layernorm_streamed_kernel<float>(float const*, float const*, float const*, float, float*, int, int, int)
CPU Time 0.00 μs
Device Time 16433817.67 μs
Self CPU Time 0.00 μs
Self Device Time 16433817.67 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
void at::native::vectorized_elementwise_kernel<4, at::native::FillFunctor<int>, at::detail::Array<char*, 1> >(int, at::native::FillFunctor<int>, at::detail::Array<char*, 1>)
CPU Time 0.00 μs
Device Time 350711.66 μs
Self CPU Time 0.00 μs
Self Device Time 350711.66 μs
CPU Memory Usage 0 B
Device Memory Usage 0 B
Self CPU Memory Usage 0 B
Self Device Memory Usage 0 B
Status: Completed 
45293 warnings generated when compiling for host.
Suppressed 45326 warnings (45279 in non-user code, 47 NOLINT).
Use -header-filter=.* to display errors from all non-system headers. Use -system-headers to display errors from system headers as well.
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:31:5 bugprone-easily-swappable-parameters
31 | const scalar_t* __restrict__ input,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
32 | const scalar_t* __restrict__ weight,
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:31:34: note: the first parameter in the range is 'input'
31 | const scalar_t* __restrict__ input,
| ^~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:32:34: note: the last parameter in the range is 'weight'
32 | const scalar_t* __restrict__ weight,
| ^~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:36:5: warning: 3 adjacent parameters of 'layernorm_streamed_kernel' of similar type ('const int') are easily swapped by mistake [bugprone-easily-swappable-parameters]
36 | const int normalized_size,
| ^~~~~~~~~~~~~~~~~~~~~~~~~~
37 | const int chunk_size,
| ~~~~~~~~~~~~~~~~~~~~~
38 | const int chunk_offset) {
| ~~~~~~~~~~~~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:36:15: note: the first parameter in the range is 'normalized_size'
36 | const int normalized_size,
| ^~~~~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:38:15: note: the last parameter in the range is 'chunk_offset'
38 | const int chunk_offset) {
| ^~~~~~~~~~~~
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:42:22: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
42 | const int tidx = threadIdx.x;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:43:22: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
43 | const int tidy = threadIdx.y;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:44:30: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
44 | const int instance_idx = blockIdx.x + chunk_offset;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:53:31: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
53 | const int thread_stride = blockDim.x * blockDim.y;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:54:27: warning: narrowing conversion from 'unsigned int' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
54 | const int thread_id = tidy * blockDim.x + tidx;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:116:33: warning: narrowing conversion from 'int64_t' (aka 'long') to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
116 | const int normalized_size = weight.numel();
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:117:28: warning: narrowing conversion from 'int64_t' (aka 'long') to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
117 | const int outer_size = x.numel() / normalized_size;
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:121:33: warning: performing an implicit widening conversion to type 'unsigned long' of a multiplication performed in type 'unsigned int' [bugprone-implicit-widening-of-multiplication-result]
121 | const int shared_mem_size = threads.x * threads.y * 2 * sizeof(float);
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:121:33: note: make conversion explicit to silence this warning
4 | const int shared_mem_size = threads.x * threads.y * 2 * sizeof(float);
| ^~~~~~~~~~~~~~~~~~~~~~~~~
| static_cast<unsigned long>( )
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:121:33: note: perform multiplication in a wider type
121 | const int shared_mem_size = threads.x * threads.y * 2 * sizeof(float);
| ^~~~~~~~~~~~~~~~~~~~~
| static_cast<unsigned long>( )
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:121:33: warning: narrowing conversion from 'unsigned long' to signed type 'int' is implementation-defined [bugprone-narrowing-conversions]
121 | const int shared_mem_size = threads.x * threads.y * 2 * sizeof(float);
| ^
/home/robert_sakana_ai/llm_cuda/experiments/20250211_optimize_b5_s4_e1_v2/level_1/task_40/b4_s3_optimized_layernorm_streamed/base/base.cu:123:5: warning: inside a lambda, '__func__' expands to the name of the function call operator; consider capturing the name of the enclosing function explicitly [bugprone-lambda-function-name]
123 | AT_DISPATCH_FLOATING_TYPES(x.scalar_type(), "layernorm_forward_cuda", ([&] {
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:237:34: note: expanded from macro 'AT_DISPATCH_FLOATING_TYPES'
237 | AT_DISPATCH_SWITCH(TYPE, NAME, AT_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:233:3: note: expanded from macro 'AT_DISPATCH_CASE_FLOATING_TYPES'
233 | AT_DISPATCH_CASE(at::ScalarType::Double, __VA_ARGS__) \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:74:3: note: expanded from macro 'AT_DISPATCH_CASE'
74 | AT_PRIVATE_CASE_TYPE_USING_HINT(enum_type, scalar_t, __VA_ARGS__)
| ^
note: (skipping 1 expansions in backtrace; use -fmacro-backtrace-limit=0 to see all)
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/ATen/Dispatch.h:58:7: note: expanded from macro 'AT_PRIVATE_CHECK_SELECTIVE_BUILD'
58 | AT_ERROR( \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Exception.h:711:32: note: expanded from macro 'AT_ERROR'
711 | C10_EXPAND_MSVC_WORKAROUND(TORCH_CHECK(false, ::c10::str(__VA_ARGS__))); \
| ^
/home/robert_sakana_ai/miniconda3/envs/llm2cuda/lib/python3.11/site-packages/torch/include/c10/util/Exception.h:536:9: note: expanded from macro 'TORCH_CHECK'
536 | __func__, \
| ^