weco 0.2.5__tar.gz → 0.2.6__tar.gz

{weco-0.2.5 → weco-0.2.6}/.github/workflows/release.yml

@@ -90,7 +90,7 @@ jobs:
         GITHUB_TOKEN: ${{ github.token }}
       run: >-
         gh release create
-        'v0.2.5'
+        'v0.2.6'
         --repo '${{ github.repository }}'
         --notes ""
 
@@ -102,5 +102,5 @@ jobs:
       # sigstore-produced signatures and certificates.
       run: >-
         gh release upload
-        'v0.2.5' dist/**
+        'v0.2.6' dist/**
         --repo '${{ github.repository }}'

{weco-0.2.5 → weco-0.2.6}/PKG-INFO

@@ -1,8 +1,8 @@
 Metadata-Version: 2.4
 Name: weco
-Version: 0.2.5
+Version: 0.2.6
 Summary: Documentation for `weco`, a CLI for using Weco AI's code optimizer.
-Author-email: Weco AI Team <dhruv@weco.ai>
+Author-email: Weco AI Team <contact@weco.ai>
 License: MIT
 Project-URL: Homepage, https://github.com/WecoAI/weco-cli
 Keywords: AI,Code Optimization,Code Generation
@@ -99,32 +99,69 @@ Here's how `weco` can be applied to common ML engineering tasks:
 
 ### Examples
 
-**Example 1: Optimizing PyTorch operations**
+**Example 1: Optimizing PyTorch simple operations**
 
 ```bash
-weco --source examples/simple-torch/optimize.py \
-     --eval-command "python examples/simple-torch/evaluate.py --solution-path examples/simple-torch/optimize.py --device mps" \
+cd examples/hello-kernel-world
+pip install torch 
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py --device cpu" \
      --metric speedup \
      --maximize true \
      --steps 15 \
-     --model o3-mini \
+     --model claude-3-7-sonnet-20250219 \
      --additional-instructions "Fuse operations in the forward method while ensuring the max float deviation remains small. Maintain the same format of the code."
 ```
 
+Note that if you have an NVIDIA gpu, change the device to `cuda`. If you are running this on Apple Silicon, set it to `mps`.
+
 **Example 2: Optimizing MLX operations with instructions from a file**
 
-Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
+Lets optimize a 2D convolution operation in [`mlx`](https://github.com/ml-explore/mlx) using [Metal](https://developer.apple.com/documentation/metal/). Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
 
 ```bash
-weco --source examples/simple-mlx/optimize.py \
-     --eval-command "python examples/simple-mlx/evaluate.py --solution-path examples/simple-mlx/optimize.py" \
+cd examples/metal
+pip install mlx
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py" \
      --metric speedup \
      --maximize true \
      --steps 30 \
      --model o3-mini \
-     --additional-instructions examples/simple-mlx/metal-examples.rst
+     --additional-instructions examples.rst
 ```
 
+**Example 3: Level Agnostic Optimization: Causal Self Attention with Triton & CUDA**
+
+Given how useful causal multihead self attention is to transformers, we've seen its wide adoption across ML engineering and AI research. Its great to keep things at a high-level (in PyTorch) when doing research, but when moving to production you often need to write highly customized low-level kernels to make things run as fast as they can. The `weco` CLI can optimize kernels across a variety of different abstraction levels and frameworks. Example 2 uses Metal but lets explore two more frameworks:
+
+1. [Triton](https://github.com/triton-lang/triton)
+    ```bash
+   cd examples/triton
+   pip install torch triton
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions "Use triton to optimize the code while ensuring a small max float diff. Maintain the same code format."
+   ```
+
+2. [CUDA](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html)
+   ```bash
+   cd examples/cuda
+   pip install torch
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions guide.md
+   ```
+
+
 ---
 
 ### Command Line Arguments

{weco-0.2.5 → weco-0.2.6}/README.md

@@ -77,32 +77,69 @@ Here's how `weco` can be applied to common ML engineering tasks:
 
 ### Examples
 
-**Example 1: Optimizing PyTorch operations**
+**Example 1: Optimizing PyTorch simple operations**
 
 ```bash
-weco --source examples/simple-torch/optimize.py \
-     --eval-command "python examples/simple-torch/evaluate.py --solution-path examples/simple-torch/optimize.py --device mps" \
+cd examples/hello-kernel-world
+pip install torch 
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py --device cpu" \
      --metric speedup \
      --maximize true \
      --steps 15 \
-     --model o3-mini \
+     --model claude-3-7-sonnet-20250219 \
      --additional-instructions "Fuse operations in the forward method while ensuring the max float deviation remains small. Maintain the same format of the code."
 ```
 
+Note that if you have an NVIDIA gpu, change the device to `cuda`. If you are running this on Apple Silicon, set it to `mps`.
+
 **Example 2: Optimizing MLX operations with instructions from a file**
 
-Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
+Lets optimize a 2D convolution operation in [`mlx`](https://github.com/ml-explore/mlx) using [Metal](https://developer.apple.com/documentation/metal/). Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
 
 ```bash
-weco --source examples/simple-mlx/optimize.py \
-     --eval-command "python examples/simple-mlx/evaluate.py --solution-path examples/simple-mlx/optimize.py" \
+cd examples/metal
+pip install mlx
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py" \
      --metric speedup \
      --maximize true \
      --steps 30 \
      --model o3-mini \
-     --additional-instructions examples/simple-mlx/metal-examples.rst
+     --additional-instructions examples.rst
 ```
 
+**Example 3: Level Agnostic Optimization: Causal Self Attention with Triton & CUDA**
+
+Given how useful causal multihead self attention is to transformers, we've seen its wide adoption across ML engineering and AI research. Its great to keep things at a high-level (in PyTorch) when doing research, but when moving to production you often need to write highly customized low-level kernels to make things run as fast as they can. The `weco` CLI can optimize kernels across a variety of different abstraction levels and frameworks. Example 2 uses Metal but lets explore two more frameworks:
+
+1. [Triton](https://github.com/triton-lang/triton)
+    ```bash
+   cd examples/triton
+   pip install torch triton
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions "Use triton to optimize the code while ensuring a small max float diff. Maintain the same code format."
+   ```
+
+2. [CUDA](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html)
+   ```bash
+   cd examples/cuda
+   pip install torch
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions guide.md
+   ```
+
+
 ---
 
 ### Command Line Arguments

weco-0.2.6/examples/cuda/evaluate.py

@@ -0,0 +1,157 @@
+import time
+import sys
+import os
+import pathlib
+import importlib
+import traceback
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+
+
+########################################################
+# Baseline
+########################################################
+class Model(nn.Module):
+    """
+    A vanilla multi-head masked self-attention layer with a projection at the end.
+    """
+
+    def __init__(self, n_embd, n_head, attn_pdrop, resid_pdrop, max_seqlen):
+        super().__init__()
+        assert n_embd % n_head == 0
+        # key, query, value projections for all heads, but in a batch
+        self.c_attn = nn.Linear(n_embd, 3 * n_embd)
+        # output projection
+        self.c_proj = nn.Linear(n_embd, n_embd)
+        # regularization
+        self.attn_dropout = nn.Dropout(attn_pdrop)
+        self.resid_dropout = nn.Dropout(resid_pdrop)
+        # causal mask to ensure that attention is only applied to the left in the input sequence
+        self.register_buffer("bias", torch.tril(torch.ones(max_seqlen, max_seqlen)).view(1, 1, max_seqlen, max_seqlen))
+        self.n_head = n_head
+        self.n_embd = n_embd
+
+    def forward(self, x):
+        B, T, C = x.size()  # batch size, sequence length, embedding dimensionality (n_embd)
+        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
+        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
+        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+
+        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
+        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
+        att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float("-inf"))
+        att = F.softmax(att, dim=-1)
+        att = self.attn_dropout(att)
+        y = att @ v  # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
+        y = y.transpose(1, 2).contiguous().view(B, T, C)  # re-assemble all head outputs side by side
+        # output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y
+
+
+########################################################
+# Weco Solution
+########################################################
+def load_module_from_path(module_path: str, add_to_sys_modules: bool = False):
+    # Clean out all old compiled extensions to prevent namespace collisions during build
+    module_path = pathlib.Path(module_path)
+    name = module_path.stem
+    spec = importlib.util.spec_from_file_location(name, module_path)
+    mod = importlib.util.module_from_spec(spec)  # type: ignore
+    if add_to_sys_modules:
+        sys.modules[name] = mod
+    spec.loader.exec_module(mod)  # type: ignore
+    return mod
+
+
+########################################################
+# Benchmark
+########################################################
+os.environ["MAX_JOBS"] = "1"  # number of workers for building with ninja
+
+
+def get_inputs(batch_size, seq_len, n_embd, device):
+    return torch.randn(batch_size, seq_len, n_embd, device=device, dtype=torch.float32)
+
+
+def bench(f, inputs, n_warmup, n_rep):
+    with torch.no_grad():
+        # warmup
+        for _ in range(n_warmup):
+            f(inputs)  # noqa
+
+        # benchmark
+        t_avg = 0.0
+        for _ in range(n_rep):
+            torch.cuda.empty_cache()  # Clear cache before timing
+            start_time = time.time()
+            f(inputs)
+            torch.cuda.synchronize()  # Wait for all computations to complete
+            t_avg += time.time() - start_time
+        t_avg /= n_rep * 1e-3
+        return t_avg
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--solution-path", type=str, required=True)
+    args = parser.parse_args()
+
+    # benchmarking parameters
+    n_correctness_trials = 10
+    n_warmup = 1000
+    n_rep = 5000
+
+    # init parameters
+    max_seqlen = 512
+    seq_len = 256
+    n_embd = 768
+    n_head = 8
+    # turn off dropout to measure correctness well
+    attn_pdrop = 0.0
+    resid_pdrop = 0.0
+
+    # input parameters
+    batch_size = 32
+
+    # load solution module
+    try:
+        torch.manual_seed(0)
+        solution_module = load_module_from_path(args.solution_path, add_to_sys_modules=False)
+        solution_model = solution_module.Model(
+            n_embd=n_embd, n_head=n_head, attn_pdrop=attn_pdrop, resid_pdrop=resid_pdrop, max_seqlen=max_seqlen
+        ).to("cuda")
+        assert isinstance(solution_model, nn.Module)
+    except Exception:
+        print(f"Candidate module initialization failed: {traceback.format_exc()}")
+        exit(1)
+
+    torch.manual_seed(0)
+    baseline_model = Model(
+        n_embd=n_embd, n_head=n_head, attn_pdrop=attn_pdrop, resid_pdrop=resid_pdrop, max_seqlen=max_seqlen
+    ).to("cuda")
+
+    # measure correctness
+    max_diff_avg = 0
+    for _ in range(n_correctness_trials):
+        inputs = get_inputs(batch_size=batch_size, seq_len=seq_len, n_embd=n_embd, device="cuda")
+        with torch.no_grad():
+            baseline_output = baseline_model(inputs)
+            optimized_output = solution_model(inputs)
+            max_diff_avg += torch.max(torch.abs(optimized_output - baseline_output))
+    max_diff_avg /= n_correctness_trials
+    print(f"max float diff between values of baseline and optimized model: {max_diff_avg}")
+
+    # measure performance
+    inputs = get_inputs(batch_size=batch_size, seq_len=seq_len, n_embd=n_embd, device="cuda")
+    t_avg_baseline = bench(baseline_model, inputs, n_warmup, n_rep)
+    print(f"baseline time: {t_avg_baseline:.2f}ms")
+    t_avg_optimized = bench(solution_model, inputs, n_warmup, n_rep)
+    print(f"optimized time: {t_avg_optimized:.2f}ms")
+    print(f"speedup: {t_avg_baseline / t_avg_optimized:.2f}x")

weco-0.2.6/examples/cuda/guide.md

@@ -0,0 +1,113 @@
+# Writing In-line CUDA Kernels: 101
+
+This document outlines the strategy to improve speedup by writing fused and optimized CUDA kernels using a single-file implementation.
+
+## Requirements
+
+- **Single-File Implementation:** Develop fused CUDA kernels within one file.
+- **No Fallback Implementation:** Do not include any alternative or fallback code.
+- **Simplicity & Readability:** Write simple, easy-to-understand code and include clear comments.
+- **Avoid Templates:** Use plain fused kernel functions without templates.
+- **Multiple Kernels Allowed:** You can define more than one kernel in the file if needed.
+- **Model Class Requirement:** The solution must include a class `Model` (an instance of `nn.Module`), with the main computation in its `forward` method.
+- **Preserve Initialization:** Do not change the initialization of the `Model` class.
+- **Focus on Efficiency:** Concentrate solely on efficient PyTorch and CUDA coding without capturing logs.
+- **Error Handling:** Any terminal output or errors will be reviewed by an LLM for feedback.
+
+## GPU Hardware Specifications
+
+Here are some details on the hardware you have access to.
+
+```json
+{
+    "GPU Architecture": "Ampere",
+    "GPU Memory": "40GB",
+    "Memory Bandwidth": "1935 GB/s",
+    "FP64 TFLOPS": "9.7",
+    "FP64 Tensor Core TFLOPS": "19.5",
+    "FP32 TFLOPS": "19.5",
+    "TF32 Tensor Core TFLOPS": "156 (312 with sparsity)",
+    "BFLOAT16 Tensore Core TFLOPS": "312 (624 with sparsity)",
+    "FP16 Tensor Core TFLOPS": "312 (624 with sparsity)",
+    "INT8 Tensor Core TOPS": "624 (1248 with sparsity)",
+    "Register File Size": "64K 32-bit registers per SM",
+    "Maximum number of registers per thread": "255",
+    "Maximum number of thread blocks per SM": "32",
+    "Shared memory capacity per SM": "164 KB",
+    "Maximum shared memory per thread block": "163 KB"
+}
+```
+
+## Baseline Code
+
+The baseline implementation of the `Model` class simply performs an element-wise addition.
+
+```python
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+
+class Model(nn.Module):
+    def __init__(self) -> None:
+        super().__init__()
+
+    def forward(self, a, b):
+        return a + b
+```
+
+## Optimized Code
+
+The optimized version employs a custom CUDA kernel for fused element-wise addition. The kernel is defined and compiled inline using PyTorch's `load_inline`.
+
+```python
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+from torch.utils.cpp_extension import load_inline
+
+# Define the custom CUDA kernel for element-wise addition
+elementwise_add_source = '''
+#include <torch/extension.h>
+#include <cuda_runtime.h>
+
+// CUDA kernel for element-wise addition
+__global__ void elementwise_add_kernel(const float* a, const float* b, float* out, int size) {
+    int idx = blockIdx.x * blockDim.x + threadIdx.x;
+    if (idx < size) {
+        out[idx] = a[idx] + b[idx];
+    }
+}
+
+// Launch function for the CUDA kernel
+torch::Tensor elementwise_add_cuda(torch::Tensor a, torch::Tensor b) {
+    auto size = a.numel();
+    auto out = torch::zeros_like(a);
+    const int block_size = 256;
+    const int num_blocks = (size + block_size - 1) / block_size;
+    elementwise_add_kernel<<<num_blocks, block_size>>>(a.data_ptr<float>(), b.data_ptr<float>(), out.data_ptr<float>(), size);
+    return out;
+}
+'''
+
+# C++ function prototype declaration
+elementwise_add_cpp_source = "torch::Tensor elementwise_add_cuda(torch::Tensor a, torch::Tensor b);"
+
+# Compile the inline CUDA code for element-wise addition
+elementwise_add = load_inline(
+    name="elementwise_add",
+    cpp_sources=elementwise_add_cpp_source,
+    cuda_sources=elementwise_add_source,
+    functions=["elementwise_add_cuda"],
+    verbose=True,
+    extra_cflags=[""],
+    extra_ldflags=[""],
+)
+
+class Model(nn.Module):
+    def __init__(self) -> None:
+        super().__init__()
+        self.elementwise_add = elementwise_add
+
+    def forward(self, a, b):
+        return self.elementwise_add.elementwise_add_cuda(a, b)
+```

weco-0.2.6/examples/cuda/optimize.py

@@ -0,0 +1,44 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+
+
+class Model(nn.Module):
+    """
+    A vanilla multi-head masked self-attention layer with a projection at the end.
+    """
+
+    def __init__(self, n_embd, n_head, attn_pdrop, resid_pdrop, max_seqlen):
+        super().__init__()
+        assert n_embd % n_head == 0
+        # key, query, value projections for all heads, but in a batch
+        self.c_attn = nn.Linear(n_embd, 3 * n_embd)
+        # output projection
+        self.c_proj = nn.Linear(n_embd, n_embd)
+        # regularization
+        self.attn_dropout = nn.Dropout(attn_pdrop)
+        self.resid_dropout = nn.Dropout(resid_pdrop)
+        # causal mask to ensure that attention is only applied to the left in the input sequence
+        self.register_buffer("bias", torch.tril(torch.ones(max_seqlen, max_seqlen)).view(1, 1, max_seqlen, max_seqlen))
+        self.n_head = n_head
+        self.n_embd = n_embd
+
+    def forward(self, x):
+        B, T, C = x.size()  # batch size, sequence length, embedding dimensionality (n_embd)
+        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
+        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
+        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+
+        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
+        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
+        att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float("-inf"))
+        att = F.softmax(att, dim=-1)
+        att = self.attn_dropout(att)
+        y = att @ v  # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
+        y = y.transpose(1, 2).contiguous().view(B, T, C)  # re-assemble all head outputs side by side
+        # output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y

{weco-0.2.5/examples/simple-torch → weco-0.2.6/examples/hello-kernel-world}/evaluate.py

@@ -28,10 +28,10 @@ class Model(nn.Module):
         Returns:
             torch.Tensor: Output tensor of shape (batch_size, hidden_size).
         """
-        x = torch.matmul(x, self.weight.T)  # Gemm
-        x = x / 2  # Divide
-        x = torch.sum(x, dim=1, keepdim=True)  # Sum
-        x = x * self.scaling_factor  # Scaling
+        x = torch.matmul(x, self.weight.T)
+        x = x / 2
+        x = torch.sum(x, dim=1, keepdim=True)
+        x = x * self.scaling_factor
         return x
 
 
@@ -60,15 +60,33 @@ def get_inputs(B, N, device):
     return torch.randn(B, N, device=device, dtype=torch.float32)
 
 
+@torch.no_grad()
 def bench(f, inputs, n_warmup, n_rep):
+    # Warm up
     for _ in range(n_warmup):
         f(inputs)  # noqa
 
+    # Benchmark
+    device_type = inputs.device.type
     t_avg = 0.0
     for _ in range(n_rep):
+        # Clear cache before timing
+        if device_type == "cuda":
+            torch.cuda.empty_cache()
+        elif device_type == "mps":
+            torch.mps.empty_cache()
+
+        # time forward pass
         start_time = time.time()
         f(inputs)
         t_avg += time.time() - start_time
+
+        # Synchronize after each iteration
+        if device_type == "cuda":
+            torch.cuda.synchronize()
+        elif device_type == "mps":
+            torch.mps.synchronize()
+
     t_avg /= n_rep * 1e-3
     return t_avg
 
@@ -81,14 +99,19 @@ if __name__ == "__main__":
     parser.add_argument("--device", default="cpu", type=str)
     args = parser.parse_args()
 
+    # benchmark parameters
+    n_correctness_trials = 10
+    n_warmup = 1000
+    n_rep = 5000
+
     # init and input parameters
-    B, N, H, S = 128, 10, 20, 1.5
+    batch_size, input_size, hidden_size, scaling_factor = 128, 10, 20, 1.5
 
     # load solution module
     try:
         torch.manual_seed(0)
         solution_module = load_module_from_path(args.solution_path, add_to_sys_modules=False)
-        solution_model = solution_module.Model(N, H, S).to(args.device)
+        solution_model = solution_module.Model(input_size, hidden_size, scaling_factor).to(args.device)
         assert isinstance(solution_model, nn.Module)
         assert hasattr(solution_model, "forward")
     except Exception:
@@ -96,13 +119,12 @@ if __name__ == "__main__":
         exit(1)
 
     torch.manual_seed(0)
-    baseline_model = Model(N, H, S).to(args.device)
+    baseline_model = Model(input_size, hidden_size, scaling_factor).to(args.device)
 
     # measure correctness
-    n_correctness_trials = 10
     max_diff_avg = 0
     for _ in range(n_correctness_trials):
-        inputs = get_inputs(B, N, args.device)
+        inputs = get_inputs(batch_size, input_size, args.device)
         baseline_output = baseline_model(inputs)
         optimized_output = solution_model(inputs)
         max_diff_avg += torch.max(torch.abs(optimized_output - baseline_output))
@@ -110,16 +132,9 @@ if __name__ == "__main__":
     print(f"max float diff between values of baseline and optimized model: {max_diff_avg}")
 
     # measure performance
-    inputs = get_inputs(B, N, args.device)
-    n_warmup = 100
-    n_rep = 500
-
-    # baseline
+    inputs = get_inputs(batch_size, input_size, args.device)
     t_avg_baseline = bench(baseline_model, inputs, n_warmup, n_rep)
     print(f"baseline time: {t_avg_baseline:.2f}ms")
-
-    # optimized
     t_avg_optimized = bench(solution_model, inputs, n_warmup, n_rep)
     print(f"optimized time: {t_avg_optimized:.2f}ms")
-
     print(f"speedup: {t_avg_baseline / t_avg_optimized:.2f}x")

{weco-0.2.5/examples/simple-mlx → weco-0.2.6/examples/metal}/evaluate.py

@@ -5,6 +5,7 @@ import importlib
 import traceback
 import mlx.core as mx
 import mlx.nn as nn
+from typing import Union
 
 
 ########################################################
@@ -12,26 +13,27 @@ import mlx.nn as nn
 ########################################################
 class Model(nn.Module):
     """
-    Model that performs a matrix multiplication, division, summation, and scaling.
+    Model that performs a 2D convolution.
+
+    Args:
+        in_channels (int): Number of input channels.
+        out_channels (int): Number of output channels.
+        kernel_size (Union[int, tuple]): Size of the convolution kernel.
+        stride (Union[int, tuple]): Stride of the convolution. Default is 1.
     """
 
-    def __init__(self, input_size, hidden_size, scaling_factor):
+    def __init__(self, in_channels: int, out_channels: int, kernel_size: Union[int, tuple], stride: Union[int, tuple] = 1):
         super(Model, self).__init__()
-        self.weight = mx.random.normal(shape=(hidden_size, input_size))
-        self.scaling_factor = scaling_factor
+        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride)
 
     def __call__(self, x):
         """
         Args:
-            x (mx.array): Input tensor of shape (batch_size, input_size).
+            x (mx.array): Input tensor of shape (batch_size, height, width, in_channels).
         Returns:
-            mx.array: Output tensor of shape (batch_size, hidden_size).
+            mx.array: Output tensor of shape (batch_size, height, width, out_channels).
         """
-        x = mx.matmul(x, mx.transpose(self.weight))  # Gemm
-        x = x / 2  # Divide
-        x = mx.sum(x, axis=1, keepdims=True)  # Sum
-        x = x * self.scaling_factor  # Scaling
-        return x
+        return self.conv(x)
 
 
 ########################################################
@@ -52,9 +54,9 @@ def load_module_from_path(module_path: str, add_to_sys_modules: bool = False):
 ########################################################
 # Benchmark
 ########################################################
-def get_inputs(B, N):
+def get_inputs(batch_size, img_height, img_width, img_channels):
     # MLX doesn't use device parameter like PyTorch, as it automatically uses Metal
-    return mx.random.normal(shape=(B, N), dtype=mx.float32)
+    return mx.random.normal(shape=(batch_size, img_height, img_width, img_channels), dtype=mx.float32)
 
 
 def bench(f, inputs, n_warmup, n_rep):
@@ -86,7 +88,13 @@ if __name__ == "__main__":
     args = parser.parse_args()
 
     # init and input parameters
-    B, N, H, S = 128, 10, 20, 1.5
+    batch_size = 4
+    img_height = 224
+    img_width = 224
+    img_channels = 3
+    out_channels = 64
+    kernel_size = 3
+    stride = 1
 
     # Set the default device to 0
     mx.set_default_device(mx.gpu)
@@ -95,20 +103,20 @@ if __name__ == "__main__":
     try:
         mx.random.seed(0)
         solution_module = load_module_from_path(args.solution_path, add_to_sys_modules=False)
-        solution_model = solution_module.Model(N, H, S)
+        solution_model = solution_module.Model(img_channels, out_channels, kernel_size, stride)
         assert hasattr(solution_model, "__call__")
     except Exception:
         print(f"Candidate module initialization failed: {traceback.format_exc()}")
         exit(1)
 
     mx.random.seed(0)
-    baseline_model = Model(N, H, S)
+    baseline_model = Model(img_channels, out_channels, kernel_size, stride)
 
     # measure correctness
     n_correctness_trials = 10
     max_diff_avg = 0
     for _ in range(n_correctness_trials):
-        inputs = get_inputs(B, N)
+        inputs = get_inputs(batch_size, img_height, img_width, img_channels)
         baseline_output = baseline_model(inputs)
         optimized_output = solution_model(inputs)
         max_diff = mx.max(mx.abs(optimized_output - baseline_output))
@@ -118,9 +126,9 @@ if __name__ == "__main__":
     print(f"max float diff between values of baseline and optimized model: {max_diff_avg}")
 
     # measure performance
-    inputs = get_inputs(B, N)
-    n_warmup = 100
-    n_rep = 500
+    inputs = get_inputs(batch_size, img_height, img_width, img_channels)
+    n_warmup = 1000
+    n_rep = 5000
 
     # baseline
     t_avg_baseline = bench(baseline_model, inputs, n_warmup, n_rep)

weco-0.2.5/examples/simple-mlx/metal-examples.rst → weco-0.2.6/examples/metal/examples.rst

@@ -3,7 +3,8 @@
 Custom Metal Kernels
 ====================
 
-MLX supports writing custom Metal kernels through the Python and C++ APIs.
+MLX supports writing custom Metal kernels through the Python and C++ APIs. Use Metal kernels allows us to optimize the performance of our models by writing low-level parallelization and vectorization schemes with fuse operations effciently.
+One important thing to keep in mind is correctness. The optimized code should produce outputs that are identical or atleast numerically close to the reference implementation i.e., max float difference less than 1e-5.
 
 When designing a custom kernel, ensure that you maintain the format of having a 'Model' class with a '__call__' method as this is what will be called to evaluate the solution.
 

weco-0.2.6/examples/metal/optimize.py

@@ -0,0 +1,28 @@
+import mlx.core as mx  # noqa
+import mlx.nn as nn
+from typing import Union
+
+
+class Model(nn.Module):
+    """
+    Model that performs a 2D convolution.
+
+    Args:
+        in_channels (int): Number of input channels.
+        out_channels (int): Number of output channels.
+        kernel_size (Union[int, tuple]): Size of the convolution kernel.
+        stride (Union[int, tuple]): Stride of the convolution. Default is 1.
+    """
+
+    def __init__(self, in_channels: int, out_channels: int, kernel_size: Union[int, tuple], stride: Union[int, tuple] = 1):
+        super(Model, self).__init__()
+        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size, stride)
+
+    def __call__(self, x):
+        """
+        Args:
+            x (mx.array): Input tensor of shape (batch_size, height, width, in_channels).
+        Returns:
+            mx.array: Output tensor of shape (batch_size, height, width, out_channels).
+        """
+        return self.conv(x)

weco-0.2.6/examples/triton/evaluate.py

@@ -0,0 +1,153 @@
+import time
+import sys
+import pathlib
+import importlib
+import traceback
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+
+
+########################################################
+# Baseline
+########################################################
+class Model(nn.Module):
+    """
+    A vanilla multi-head masked self-attention layer with a projection at the end.
+    """
+
+    def __init__(self, n_embd, n_head, attn_pdrop, resid_pdrop, max_seqlen):
+        super().__init__()
+        assert n_embd % n_head == 0
+        # key, query, value projections for all heads, but in a batch
+        self.c_attn = nn.Linear(n_embd, 3 * n_embd)
+        # output projection
+        self.c_proj = nn.Linear(n_embd, n_embd)
+        # regularization
+        self.attn_dropout = nn.Dropout(attn_pdrop)
+        self.resid_dropout = nn.Dropout(resid_pdrop)
+        # causal mask to ensure that attention is only applied to the left in the input sequence
+        self.register_buffer("bias", torch.tril(torch.ones(max_seqlen, max_seqlen)).view(1, 1, max_seqlen, max_seqlen))
+        self.n_head = n_head
+        self.n_embd = n_embd
+
+    def forward(self, x):
+        B, T, C = x.size()  # batch size, sequence length, embedding dimensionality (n_embd)
+        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
+        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
+        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+
+        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
+        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
+        att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float("-inf"))
+        att = F.softmax(att, dim=-1)
+        att = self.attn_dropout(att)
+        y = att @ v  # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
+        y = y.transpose(1, 2).contiguous().view(B, T, C)  # re-assemble all head outputs side by side
+        # output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y
+
+
+########################################################
+# Weco Solution
+########################################################
+def load_module_from_path(module_path: str, add_to_sys_modules: bool = False):
+    # Clean out all old compiled extensions to prevent namespace collisions during build
+    module_path = pathlib.Path(module_path)
+    name = module_path.stem
+    spec = importlib.util.spec_from_file_location(name, module_path)
+    mod = importlib.util.module_from_spec(spec)  # type: ignore
+    if add_to_sys_modules:
+        sys.modules[name] = mod
+    spec.loader.exec_module(mod)  # type: ignore
+    return mod
+
+
+########################################################
+# Benchmark
+########################################################
+def get_inputs(batch_size, seq_len, n_embd, device):
+    return torch.randn(batch_size, seq_len, n_embd, device=device, dtype=torch.float32)
+
+
+@torch.no_grad()
+def bench(f, inputs, n_warmup, n_rep):
+    # warmup
+    for _ in range(n_warmup):
+        f(inputs)  # noqa
+
+    # benchmark
+    t_avg = 0.0
+    for _ in range(n_rep):
+        torch.cuda.empty_cache()  # Clear cache before timing
+        start_time = time.time()
+        f(inputs)
+        torch.cuda.synchronize()  # Wait for all computations to complete
+        t_avg += time.time() - start_time
+    t_avg /= n_rep * 1e-3
+    return t_avg
+
+
+if __name__ == "__main__":
+    import argparse
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--solution-path", type=str, required=True)
+    args = parser.parse_args()
+
+    # benchmarking parameters
+    n_correctness_trials = 10
+    n_warmup = 1000
+    n_rep = 5000
+
+    # init parameters
+    max_seqlen = 512
+    seq_len = 256
+    n_embd = 768
+    n_head = 8
+    # turn off dropout to measure correctness well
+    attn_pdrop = 0.0
+    resid_pdrop = 0.0
+
+    # input parameters
+    batch_size = 32
+
+    # load solution module
+    try:
+        torch.manual_seed(0)
+        solution_module = load_module_from_path(args.solution_path, add_to_sys_modules=False)
+        solution_model = solution_module.Model(
+            n_embd=n_embd, n_head=n_head, attn_pdrop=attn_pdrop, resid_pdrop=resid_pdrop, max_seqlen=max_seqlen
+        ).to("cuda")
+        assert isinstance(solution_model, nn.Module)
+    except Exception:
+        print(f"Candidate module initialization failed: {traceback.format_exc()}")
+        exit(1)
+
+    torch.manual_seed(0)
+    baseline_model = Model(
+        n_embd=n_embd, n_head=n_head, attn_pdrop=attn_pdrop, resid_pdrop=resid_pdrop, max_seqlen=max_seqlen
+    ).to("cuda")
+
+    # measure correctness
+    max_diff_avg = 0
+    for _ in range(n_correctness_trials):
+        inputs = get_inputs(batch_size=batch_size, seq_len=seq_len, n_embd=n_embd, device="cuda")
+        with torch.no_grad():
+            baseline_output = baseline_model(inputs)
+            optimized_output = solution_model(inputs)
+            max_diff_avg += torch.max(torch.abs(optimized_output - baseline_output))
+    max_diff_avg /= n_correctness_trials
+    print(f"max float diff between values of baseline and optimized model: {max_diff_avg}")
+
+    # measure performance
+    inputs = get_inputs(batch_size=batch_size, seq_len=seq_len, n_embd=n_embd, device="cuda")
+    t_avg_baseline = bench(baseline_model, inputs, n_warmup, n_rep)
+    print(f"baseline time: {t_avg_baseline:.2f}ms")
+    t_avg_optimized = bench(solution_model, inputs, n_warmup, n_rep)
+    print(f"optimized time: {t_avg_optimized:.2f}ms")
+    print(f"speedup: {t_avg_baseline / t_avg_optimized:.2f}x")

weco-0.2.6/examples/triton/optimize.py

@@ -0,0 +1,44 @@
+import torch
+import torch.nn as nn
+import torch.nn.functional as F
+import math
+
+
+class Model(nn.Module):
+    """
+    A vanilla multi-head masked self-attention layer with a projection at the end.
+    """
+
+    def __init__(self, n_embd, n_head, attn_pdrop, resid_pdrop, max_seqlen):
+        super().__init__()
+        assert n_embd % n_head == 0
+        # key, query, value projections for all heads, but in a batch
+        self.c_attn = nn.Linear(n_embd, 3 * n_embd)
+        # output projection
+        self.c_proj = nn.Linear(n_embd, n_embd)
+        # regularization
+        self.attn_dropout = nn.Dropout(attn_pdrop)
+        self.resid_dropout = nn.Dropout(resid_pdrop)
+        # causal mask to ensure that attention is only applied to the left in the input sequence
+        self.register_buffer("bias", torch.tril(torch.ones(max_seqlen, max_seqlen)).view(1, 1, max_seqlen, max_seqlen))
+        self.n_head = n_head
+        self.n_embd = n_embd
+
+    def forward(self, x):
+        B, T, C = x.size()  # batch size, sequence length, embedding dimensionality (n_embd)
+        # calculate query, key, values for all heads in batch and move head forward to be the batch dim
+        q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
+        k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+        v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)  # (B, nh, T, hs)
+
+        # causal self-attention; Self-attend: (B, nh, T, hs) x (B, nh, hs, T) -> (B, nh, T, T)
+        att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
+        att = att.masked_fill(self.bias[:, :, :T, :T] == 0, float("-inf"))
+        att = F.softmax(att, dim=-1)
+        att = self.attn_dropout(att)
+        y = att @ v  # (B, nh, T, T) x (B, nh, T, hs) -> (B, nh, T, hs)
+        y = y.transpose(1, 2).contiguous().view(B, T, C)  # re-assemble all head outputs side by side
+        # output projection
+        y = self.resid_dropout(self.c_proj(y))
+        return y

{weco-0.2.5 → weco-0.2.6}/pyproject.toml

@@ -6,11 +6,11 @@ build-backend = "setuptools.build_meta"
 [project]
 name = "weco"
 authors = [
-    {name = "Weco AI Team", email = "dhruv@weco.ai"},
+    {name = "Weco AI Team", email = "contact@weco.ai"},
 ]
 description = "Documentation for `weco`, a CLI for using Weco AI's code optimizer."
 readme = "README.md"
-version = "0.2.5"
+version = "0.2.6"
 license = {text = "MIT"}
 requires-python = ">=3.12"
 dependencies = ["requests", "rich"]

{weco-0.2.5 → weco-0.2.6}/weco/__init__.py

@@ -1,4 +1,4 @@
 # DO NOT EDIT
-__pkg_version__ = "0.2.5"
+__pkg_version__ = "0.2.6"
 __api_version__ = "v1"
 __base_url__ = f"https://api.aide.weco.ai/{__api_version__}"

{weco-0.2.5 → weco-0.2.6}/weco/cli.py

@@ -321,7 +321,12 @@ def main() -> None:
             _, best_solution_panel = solution_panels.get_display(current_step=steps)
 
             # Update the end optimization layout
-            end_optimization_layout["summary"].update(summary_panel.get_display())
+            final_message = (
+                f"{summary_panel.metric_name.capitalize()} {'maximized' if summary_panel.maximize else 'minimized'}! Best solution {summary_panel.metric_name.lower()} = [green]{status_response['best_result']['metric_value']}[/] 🏆"
+                if best_solution_node is not None
+                else "[red] No solution found.[/]"
+            )
+            end_optimization_layout["summary"].update(summary_panel.get_display(final_message=final_message))
             end_optimization_layout["tree"].update(tree_panel.get_display())
             end_optimization_layout["best_solution"].update(best_solution_panel)
 

{weco-0.2.5 → weco-0.2.6}/weco/panels.py

@@ -12,7 +12,9 @@ class SummaryPanel:
     """Holds a summary of the optimization session."""
 
     def __init__(self, maximize: bool, metric_name: str, total_steps: int, model: str, session_id: str = None):
-        self.goal = ("Maximizing" if maximize else "Minimizing") + f" {metric_name}..."
+        self.maximize = maximize
+        self.metric_name = metric_name
+        self.goal = ("Maximizing" if self.maximize else "Minimizing") + f" {self.metric_name}..."
         self.total_input_tokens = 0
         self.total_output_tokens = 0
         self.total_steps = total_steps
@@ -39,24 +41,28 @@ class SummaryPanel:
         self.total_input_tokens += usage["input_tokens"]
         self.total_output_tokens += usage["output_tokens"]
 
-    def get_display(self) -> Panel:
+    def get_display(self, final_message: Optional[str] = None) -> Panel:
         """Create a summary panel with the relevant information."""
         layout = Layout(name="summary")
         summary_table = Table(show_header=False, box=None, padding=(0, 1))
         # Goal
-        summary_table.add_row(f"[bold cyan]Goal:[/] {self.goal}")
+        if final_message is not None:
+            summary_table.add_row(f"[bold cyan]Result:[/] {final_message}")
+        else:
+            summary_table.add_row(f"[bold cyan]Goal:[/] {self.goal}")
+        summary_table.add_row("")
+        # Model used
+        summary_table.add_row(f"[bold cyan]Model:[/] {self.model}")
         summary_table.add_row("")
         # Log directory
         runs_dir = f".runs/{self.session_id}"
         summary_table.add_row(f"[bold cyan]Logs:[/] [blue underline]{runs_dir}[/]")
         summary_table.add_row("")
-        # Model used
-        summary_table.add_row(f"[bold cyan]Model:[/] [yellow]{self.model}[/]")
-        summary_table.add_row("")
         # Token counts
         summary_table.add_row(
             f"[bold cyan]Tokens:[/] ↑[yellow]{format_number(self.total_input_tokens)}[/] ↓[yellow]{format_number(self.total_output_tokens)}[/] = [green]{format_number(self.total_input_tokens + self.total_output_tokens)}[/]"
         )
+        summary_table.add_row("")
         # Progress bar
         summary_table.add_row(self.progress)
 

{weco-0.2.5 → weco-0.2.6}/weco.egg-info/PKG-INFO

@@ -1,8 +1,8 @@
 Metadata-Version: 2.4
 Name: weco
-Version: 0.2.5
+Version: 0.2.6
 Summary: Documentation for `weco`, a CLI for using Weco AI's code optimizer.
-Author-email: Weco AI Team <dhruv@weco.ai>
+Author-email: Weco AI Team <contact@weco.ai>
 License: MIT
 Project-URL: Homepage, https://github.com/WecoAI/weco-cli
 Keywords: AI,Code Optimization,Code Generation
@@ -99,32 +99,69 @@ Here's how `weco` can be applied to common ML engineering tasks:
 
 ### Examples
 
-**Example 1: Optimizing PyTorch operations**
+**Example 1: Optimizing PyTorch simple operations**
 
 ```bash
-weco --source examples/simple-torch/optimize.py \
-     --eval-command "python examples/simple-torch/evaluate.py --solution-path examples/simple-torch/optimize.py --device mps" \
+cd examples/hello-kernel-world
+pip install torch 
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py --device cpu" \
      --metric speedup \
      --maximize true \
      --steps 15 \
-     --model o3-mini \
+     --model claude-3-7-sonnet-20250219 \
      --additional-instructions "Fuse operations in the forward method while ensuring the max float deviation remains small. Maintain the same format of the code."
 ```
 
+Note that if you have an NVIDIA gpu, change the device to `cuda`. If you are running this on Apple Silicon, set it to `mps`.
+
 **Example 2: Optimizing MLX operations with instructions from a file**
 
-Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
+Lets optimize a 2D convolution operation in [`mlx`](https://github.com/ml-explore/mlx) using [Metal](https://developer.apple.com/documentation/metal/). Sometimes, additional context or instructions are too complex for a single command-line string. You can provide a path to a file containing these instructions.
 
 ```bash
-weco --source examples/simple-mlx/optimize.py \
-     --eval-command "python examples/simple-mlx/evaluate.py --solution-path examples/simple-mlx/optimize.py" \
+cd examples/metal
+pip install mlx
+weco --source optimize.py \
+     --eval-command "python evaluate.py --solution-path optimize.py" \
      --metric speedup \
      --maximize true \
      --steps 30 \
      --model o3-mini \
-     --additional-instructions examples/simple-mlx/metal-examples.rst
+     --additional-instructions examples.rst
 ```
 
+**Example 3: Level Agnostic Optimization: Causal Self Attention with Triton & CUDA**
+
+Given how useful causal multihead self attention is to transformers, we've seen its wide adoption across ML engineering and AI research. Its great to keep things at a high-level (in PyTorch) when doing research, but when moving to production you often need to write highly customized low-level kernels to make things run as fast as they can. The `weco` CLI can optimize kernels across a variety of different abstraction levels and frameworks. Example 2 uses Metal but lets explore two more frameworks:
+
+1. [Triton](https://github.com/triton-lang/triton)
+    ```bash
+   cd examples/triton
+   pip install torch triton
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions "Use triton to optimize the code while ensuring a small max float diff. Maintain the same code format."
+   ```
+
+2. [CUDA](https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html)
+   ```bash
+   cd examples/cuda
+   pip install torch
+   weco --source optimize.py \
+        --eval-command "python evaluate.py --solution-path optimize.py" \
+        --metric speedup \
+        --maximize true \
+        --steps 30 \
+        --model gemini-2.5-pro-preview-03-25 \
+        --additional-instructions guide.md
+   ```
+
+
 ---
 
 ### Command Line Arguments

weco-0.2.6/weco.egg-info/SOURCES.txt

@@ -0,0 +1,27 @@
+.gitignore
+LICENSE
+README.md
+pyproject.toml
+.github/workflows/lint.yml
+.github/workflows/release.yml
+examples/cuda/evaluate.py
+examples/cuda/guide.md
+examples/cuda/optimize.py
+examples/hello-kernel-world/evaluate.py
+examples/hello-kernel-world/optimize.py
+examples/metal/evaluate.py
+examples/metal/examples.rst
+examples/metal/optimize.py
+examples/triton/evaluate.py
+examples/triton/optimize.py
+weco/__init__.py
+weco/api.py
+weco/cli.py
+weco/panels.py
+weco/utils.py
+weco.egg-info/PKG-INFO
+weco.egg-info/SOURCES.txt
+weco.egg-info/dependency_links.txt
+weco.egg-info/entry_points.txt
+weco.egg-info/requires.txt
+weco.egg-info/top_level.txt

weco-0.2.5/examples/simple-mlx/optimize.py

@@ -1,26 +0,0 @@
-import mlx.core as mx
-import mlx.nn as nn
-
-
-class Model(nn.Module):
-    """
-    Model that performs a matrix multiplication, division, summation, and scaling.
-    """
-
-    def __init__(self, input_size, hidden_size, scaling_factor):
-        super(Model, self).__init__()
-        self.weight = mx.random.normal(shape=(hidden_size, input_size))
-        self.scaling_factor = scaling_factor
-
-    def __call__(self, x):
-        """
-        Args:
-            x (mx.array): Input tensor of shape (batch_size, input_size).
-        Returns:
-            mx.array: Output tensor of shape (batch_size, hidden_size).
-        """
-        x = mx.matmul(x, mx.transpose(self.weight))
-        x = x / 2
-        x = mx.sum(x, axis=1, keepdims=True)
-        x = x * self.scaling_factor
-        return x

weco-0.2.5/weco.egg-info/SOURCES.txt

@@ -1,22 +0,0 @@
-.gitignore
-LICENSE
-README.md
-pyproject.toml
-.github/workflows/lint.yml
-.github/workflows/release.yml
-examples/simple-mlx/evaluate.py
-examples/simple-mlx/metal-examples.rst
-examples/simple-mlx/optimize.py
-examples/simple-torch/evaluate.py
-examples/simple-torch/optimize.py
-weco/__init__.py
-weco/api.py
-weco/cli.py
-weco/panels.py
-weco/utils.py
-weco.egg-info/PKG-INFO
-weco.egg-info/SOURCES.txt
-weco.egg-info/dependency_links.txt
-weco.egg-info/entry_points.txt
-weco.egg-info/requires.txt
-weco.egg-info/top_level.txt