neural-cost
PyPI Package • Documentation • Issue Tracker • Releases
neural-cost estimates a neural network's useful compute and compulsory tensor
traffic, measures its runtime, and uses a roofline lower bound to highlight
likely optimization opportunities. It is intentionally framework-neutral at
its core: PyTorch, TensorFlow, and JAX are optional adapters rather than base
dependencies.
Installation
Install the core package from PyPI:
pip install neural-cost
Or using uv:
uv add neural-cost
Framework Adapters
Framework adapters are optional extras, keeping neural-cost lightweight with zero mandatory external dependencies:
| Framework | Installation Command | Description |
|---|---|---|
| PyTorch | pip install "neural-cost[torch]" |
PyTorch eager execution & FX graph tracer adapter |
| JAX | pip install "neural-cost[jax]" |
JAX primitives and jaxpr tracing |
| JAX on Apple Silicon (Metal) | pip install "neural-cost[jax-metal]" |
Official Apple Metal plugin for JAX GPU acceleration |
| JAX on Apple Silicon (MPS) | pip install "neural-cost[jax-mps]" |
Community MLX backend for Apple Silicon MPS |
| TensorFlow | pip install "neural-cost[tensorflow]" |
TensorFlow / Keras layer adapter |
| All Frameworks | pip install "neural-cost[torch,jax,tensorflow]" |
All supported deep learning framework adapters |
| Development | pip install "neural-cost[dev]" |
Test suite (pytest) & code style tools (ruff) |
Development & Source Installation
For local development or running benchmarks directly from the source repository:
git clone https://github.com/davidgraymi/neural-cost.git
cd neural-cost
pip install -e '.[dev,torch,jax,tensorflow]'
Architecture
flowchart LR
Model["Model / function\nexample inputs"]
Adapters["PyTorch · JAX · TensorFlow\nCustom FrameworkAdapter"]
Operations["Portable Operation records\nlinear · conv2d · embedding · attention\nRNN/LSTM · layernorm · pooling …"]
Static["Static profile\nFLOPs · traffic · parameter bytes\nactivation bounds"]
Dynamic["Dynamic tracer\nruntime · allocator telemetry\nPyTorch profiler events"]
Detect["Hardware Detect\nchip table · STREAM triad\nnvidia-smi · CPU fallback"]
Hardware["HardwareSpec\npeak FLOP/s · bandwidth"]
Gap["Gap analysis\nroofline efficiency · memory overhead\nfindings"]
Model --> Adapters
Adapters --> Operations --> Static --> Gap
Adapters --> Dynamic --> Gap
Detect --> Hardware
Hardware --> Gap
Supported operation kinds
| Kind | Description |
|---|---|
linear |
Dense / fully-connected projection |
conv2d |
2-D convolution (NCHW / OIHW) |
matmul |
Raw matrix multiply |
embedding |
Token embedding lookup (read-only traffic model) |
attention |
Multi-head self-attention (QKV projections + softmax + output) |
elementwise |
Point-wise ops (ReLU, exp, tanh, …) |
softmax / layernorm / batchnorm |
Normalisation ops |
rmsnorm |
Root Mean Square normalisation |
swiglu |
Gated linear unit with Swish activation |
pooling |
Max / average / global-average pooling |
moe |
Mixture-of-Experts sparse routing and expert FFN execution |
custom |
Caller-supplied explicit FLOP count |
Adapters automatically emit the right kind for each layer type:
| Framework | Captured layer types |
|---|---|
| PyTorch | Linear, Conv2d, Embedding, RNN, GRU, LSTM, MultiheadAttention, LayerNorm, BatchNorm1d/2d |
| TensorFlow | Dense, Conv2D, Embedding, GRU, LSTM, MultiHeadAttention, BatchNormalization, LayerNormalization, pooling layers |
| JAX | dot_general (matmul), conv_general_dilated (conv2d), common elementwise jaxpr primitives |
Analyze portable operations
from neural_cost import HardwareSpec, Operation, analyze_gap, benchmark, estimate_operations, detect_hardware
ops = [Operation("classifier", "linear", ((32, 768), (768, 1000)), (32, 1000), 2)]
estimate = estimate_operations(ops)
measurement = benchmark(lambda: run_inference(), warmup=5, repeats=20)
# Auto-detect or specify manually:
hardware, info = detect_hardware()
# Or: hardware = HardwareSpec("GPU", peak_flops=312e12, memory_bandwidth=1.6e12)
report = analyze_gap(estimate, measurement, hardware)
print(report.render())
The theoretical model reports FLOPs, tensor reads/writes, arithmetic intensity, and a compute/bandwidth lower bound. The measured gap is expected: it captures launch overhead, synchronization, framework behavior, unfused intermediates, workspaces, caches, and imperfect kernel utilization.
Profile static memory and training state
profile_model combines FLOP/traffic estimation with parameter and activation
storage bounds. For training, it also models a parameter-sized gradient buffer
and configurable optimizer state; use optimizer_state_multiplier=2 for
Adam's two moment buffers.
from neural_cost import profile_model
from neural_cost.adapters import TorchAdapter
profile = profile_model(
model, inputs, TorchAdapter(), training=True, optimizer_state_multiplier=2
)
print(profile.memory.training_minimum_bytes)
The minimum activation bound is the largest output tensor. The conservative bound assumes all forward outputs remain live, so real allocator telemetry is the source of truth for physical VRAM use.
Mixture-of-Experts (MoE) & Speculative Decoding
neural-cost provides specialized theoretical models for modern LLM serving paradigms:
MoE Sparse Routing & Memory Thrashing
Sparse MoE architectures (e.g. Mixtral, DeepSeek) activate only $k$ of $E$ experts per token. While compute scales with $k$, token-by-token autoregressive decoding requires loading unique experts from DRAM into registers/SRAM, producing severe memory-bandwidth pressure at small batch sizes:
from neural_cost import estimate_moe, analyze_moe_gap, detect_hardware
# Mixtral 8x7B layer: 8 experts, top-2 routing, SwiGLU FFN
est = estimate_moe(
batch_size=1,
seq_len=1,
embed_dim=4096,
expert_hidden_dim=14336,
num_experts=8,
top_k=2,
expert_type="swiglu",
is_decode=True,
)
print(f"Total params: {est.total_parameters:,} | Active params: {est.active_parameters:,}")
print(f"Expected loaded experts at B=1: {est.expected_loaded_experts:.1f}")
hardware, _ = detect_hardware()
gap = analyze_moe_gap(est, hardware)
print(gap.render())
Speculative Decoding Breakeven Analysis
Speculative decoding pairs a fast draft model with a parallel target model verification step. neural-cost computes the analytical breakeven acceptance rate ($\alpha^*$) required to achieve wall-clock speedup on target hardware:
from neural_cost import (
CostEstimate,
detect_hardware,
estimate_speculative_decoding,
analyze_speculative_decoding,
)
hardware, _ = detect_hardware()
# Target (70B) vs Draft (1B) step costs
draft_decode = CostEstimate(flops=2e9, read_bytes=2e9, write_bytes=2048, operations=1)
target_verify = CostEstimate(flops=560e9, read_bytes=140e9, write_bytes=16384, operations=1)
target_decode = CostEstimate(flops=140e9, read_bytes=140e9, write_bytes=4096, operations=1)
analysis = analyze_speculative_decoding(
draft_decode_cost=draft_decode,
target_verify_cost=target_verify,
target_decode_cost=target_decode,
hardware=hardware,
gamma=4,
acceptance_rate=0.75,
)
print(analysis.render())
PagedAttention & KV Cache Memory Modeling
PagedAttention (vLLM) allocates non-contiguous fixed-size blocks (e.g. 16 tokens) to eliminate external fragmentation and over-reservation. neural-cost models exact block allocation, compulsory vs wasted fragmentation bytes, prefix cache sharing, and concurrency capacity multipliers:
from neural_cost import estimate_paged_attention, analyze_paged_attention_gap
# 32 concurrent requests with heterogeneous context lengths
context_lens = [256 + (i * 32) for i in range(32)]
paged_cost = estimate_paged_attention(
batch_size=32,
context_lens=context_lens,
embed_dim=4096,
num_heads=32,
num_kv_heads=8, # Grouped-Query Attention (GQA)
block_size=16,
num_layers=32,
max_model_len=4096,
shared_prefix_tokens=128, # 128-token shared system prompt
)
gap = analyze_paged_attention_gap(paged_cost)
print(gap.render())
# Reports VRAM saved vs contiguous max reservation, fragmentation ratio, and concurrency gain
Continuous Batching Iteration Cost Simulator
Modern serving engines (vLLM, TGI, TensorRT-LLM) use iteration-level scheduling (continuous/in-flight batching) mixing memory-bound autoregressive decodes and compute-heavy chunked prefills in the same step. neural-cost simulates the unified arithmetic intensity and lower-bound execution time across model weights and KV caches:
from neural_cost import (
estimate_continuous_batch_iteration,
analyze_continuous_batch_iteration,
detect_hardware,
)
hardware, _ = detect_hardware()
# Mixed iteration: 64 decode streams + one 512-token chunked prefill
iteration = estimate_continuous_batch_iteration(
decode_context_lens=[1024] * 64,
prefill_chunk_lens=[512],
embed_dim=4096,
num_heads=32,
num_kv_heads=8,
num_layers=32,
hardware=hardware,
)
gap = analyze_continuous_batch_iteration(iteration, hardware)
print(gap.render())
# Identifies compute vs memory bottleneck and optimal prefill chunk sizing to reach the roofline ridge
State Space Models (Mamba / S6 / SSD) & Linear Attention
State Space Models replace the quadratic $O(L^2)$ attention mechanism and linear-growth $O(L)$ KV cache with a fixed-size $O(1)$ hidden recurrent state ($h_t \in \mathbb{R}^{B \times D_{\text{in}} \times N}$).
neural-cost models the associative parallel scan in prefill and constant-memory recurrent step in decode:
from neural_cost import analyze_ssm_gap, detect_hardware, estimate_ssm
# Mamba-3B layer: hidden dim 4096, state dim 16, expand factor 2
est = estimate_ssm(
batch_size=1,
seq_len=8192,
embed_dim=4096,
state_dim=16,
expand_factor=2,
conv_kernel_size=4,
num_layers=32,
is_decode=True,
)
print(f"Recurrent state footprint: {est.state_bytes / 1e6:.2f} MB (constant O(1) in seq_len)")
print(
f"KV cache eliminated: {est.memory_savings_ratio_vs_transformer:.1%} savings vs Transformer KV"
)
hardware, _ = detect_hardware()
gap = analyze_ssm_gap(est, hardware)
print(gap.render())
# Reports constant-memory decode lower bound and throughput advantage over Transformer KV cache retrieval
Distributed 3D Parallelism & Interconnect Rooflines (v0.8.0)
When scaling large models across multi-node clusters, training throughput is governed by the interaction between compute rooflines, network interconnect fabrics, and 3D parallelism strategy.
neural-cost models Tensor Parallelism (TP), Pipeline Parallelism (PP), and Data Parallelism / ZeRO / FSDP communication volumes, Hockney ($\alpha$-$\beta$) network transfer times, 1F1B schedule bubble fractions, and per-device memory sharding:
from neural_cost import (
ClusterTopology,
HardwareSpec,
analyze_distributed_gap,
estimate_parallelism,
get_interconnect_preset,
)
# 128x H100 cluster across 16 nodes (8 GPUs per node)
h100 = HardwareSpec(
"H100", peak_flops=989e12, memory_bandwidth=3.35e12, memory_capacity=80 * 1024**3
)
topo = ClusterTopology(
device=h100,
num_nodes=16,
devices_per_node=8,
intra_node=get_interconnect_preset("nvlink4"), # 900 GB/s
inter_node=get_interconnect_preset("infiniband_ndr"), # 50 GB/s
)
# Train Llama-3 70B with TP=8, PP=4, DP=4 (ZeRO-3 / FSDP)
cost = estimate_parallelism(
total_parameters=70_000_000_000,
batch_size=128,
seq_len=4096,
embed_dim=8192,
num_layers=80,
tp_degree=8,
pp_degree=4,
dp_degree=4,
num_microbatches=8,
dp_mode="zero3_fsdp",
)
gap = analyze_distributed_gap(cost, topo, overlap_efficiency=0.85)
print(gap.render())
# Reports MFU, step time, communication overhead, pipeline bubble fraction, and memory fit
Hardware detection
Auto-detection via detect_hardware() returns a (HardwareSpec, DetectionResult) tuple
to determine hardware peak compute and memory bandwidth:
- Apple Silicon lookup from chip table: Identifies Apple Silicon chips (M1–M4 series) and looks up published peak FP32 throughput and memory bandwidth.
- NumPy STREAM-triad bandwidth benchmark: Measures live effective memory bandwidth using a STREAM Triad kernel (
c = a + scalar * b). - NVIDIA GPU probe via nvidia-smi: Probes GPU models, clock rates, and bus specs on systems with NVIDIA GPUs.
- CPU fallback: Falls back to CPU logical core counts and clock rates when accelerator probes are unavailable.
from neural_cost import detect_hardware
hardware, info = detect_hardware()
print(f"Device: {hardware.device_name} ({info.source})")
print(f"Peak FLOP/s: {hardware.peak_flops / 1e12:.1f} TFLOP/s")
print(f"Bandwidth: {hardware.memory_bandwidth / 1e9:.1f} GB/s")
Framework adapters
import torch
from neural_cost import estimate_model
from neural_cost.adapters import TorchAdapter
model = torch.nn.Sequential(torch.nn.Linear(128, 64), torch.nn.ReLU(), torch.nn.Linear(64, 10))
inputs = (torch.randn(16, 128),)
estimate = estimate_model(model, inputs, TorchAdapter())
measurement = TorchAdapter().benchmark(model, *inputs)
TorchAdapter captures Linear, Conv2d, Embedding, RNN, GRU, LSTM,
MultiheadAttention, LayerNorm, and BatchNorm modules via forward hooks and
synchronizes CUDA benchmarks. Its trace method uses torch.profiler and
returns aggregate profiler-event and CUDA allocator statistics.
TensorFlowAdapter captures the equivalent Keras layers and returns supported
TensorFlow GPU allocator statistics. JaxAdapter traces conventional
dot_general, conv_general_dilated, and common elementwise jaxpr primitives
and waits for asynchronous device work during benchmarks. All adapters are
optional imports:
from neural_cost.adapters import JaxAdapter, TensorFlowAdapter, TorchAdapter
E2E architecture comparison
Run the architecture comparison script to benchmark five canonical neural network families side-by-side across all installed frameworks:
pip install "neural-cost[torch,jax,tensorflow]"
# Or from local editable checkout:
# pip install -e '.[torch,jax,tensorflow]'
python examples/architecture_comparison.py
The script evaluates FF DNN, CNN, RNN, LSTM, and Transformer
architectures using a shared hidden dimension (128) and batch size (16).
Hardware is auto-detected; pass --peak-flops / --memory-bandwidth to override.
GPU benchmark
Run the GPU benchmark to evaluate the same five architectures on available GPU accelerators (CUDA, ROCm, MPS, or CPU fallback):
# Auto-detect GPU (CUDA → MPS → CPU fallback)
python benchmarks/collect_gpu_data.py
# Quick run (fewer batch sizes / repeats)
python benchmarks/collect_gpu_data.py --quick
# Explicit CUDA device
python benchmarks/collect_gpu_data.py --device cuda
# Apple Silicon GPU
python benchmarks/collect_gpu_data.py --device mps
# Override hardware specs (e.g. NVIDIA A100 80 GB)
python benchmarks/collect_gpu_data.py \
--peak-flops 312e12 --memory-bandwidth 2.0e12
Results are written to benchmarks/results/benchmark_gpu_data.json.
Generate figures and the full markdown report:
python benchmarks/generate_gpu_report.py
# → GPU_BENCHMARK_REPORT.md + benchmarks/results/figures/gpu_fig*.png
GPU-vs-CPU crossover analysis
Find the exact batch size at which each architecture first runs faster on GPU than CPU:
# Collect CPU baseline first (if not already done)
python benchmarks/collect_data.py
# Run GPU benchmark with the fine batch-size grid (1, 4, 8 … 1024)
python benchmarks/collect_gpu_data.py --crossover
# Regenerate report — Figure GPU-8 (crossover plot) will now be included
python benchmarks/generate_gpu_report.py
The crossover analysis sweeps CROSSOVER_BATCH_SIZES = [1, 4, 8, 16, 32, 64, 128, 256, 512, 1024]
and prints a table showing the first batch at which GPU latency drops below CPU latency.
JAX on Apple Silicon (MPS / Metal)
JAX requires an explicit GPU plugin to run on Apple Silicon GPUs. Without a plugin, JAX silently falls back to CPU even on MPS-capable machines. Install one of:
# Official Apple plugin via package extra:
pip install "neural-cost[jax-metal]"
# or directly:
pip install jax-metal
# Community MLX backend via package extra (set JAX_PLATFORMS=mps):
pip install "neural-cost[jax-mps]"
# or directly:
pip install jax-mps && JAX_PLATFORMS=mps python benchmarks/collect_gpu_data.py
The benchmark now detects whether a plugin is installed and emits a clear diagnostic when JAX is running on CPU instead of the GPU.
Known JAX MPS limitations (tracked upstream):
jax.jit()regresses CNN latency 3× on MPS — XLA's Metal conv lowering inserts extra memory-layout transposes for statically-shaped graphs (known bug).jax.jit()gains are large for LSTM (+1.79×) and Transformer (+1.12×) where XLA eliminates intermediate tensor roundtrips.
GPU vs CPU benchmark differences
| Aspect | CPU benchmark | GPU benchmark |
|---|---|---|
| Timing | time.perf_counter_ns |
CUDA events (torch.cuda.Event) |
| Sync barrier | None (CPU executes synchronously) | torch.cuda.synchronize() / block_until_ready() |
| TF optimised variant | tf.function (no XLA) |
tf.function(jit_compile=True) (XLA GPU) |
| Batch sizes | 1, 8, 32, 128 | 8, 32, 128, 512 |
| Device placement | CPU tensors | .to(device) / jax.device_put / tf.device |
Sample output (Apple M3, 3.6 TFLOP/s · 100 GB/s, batch=16, seq=32)
┌─ Hardware ──────────────────────────────────────────────────────────────────
│ Device : Apple M3
│ Peak FP32 : 3.60 TFLOP/s
│ Peak bandwidth : 100.0 GB/s (STREAM triad: 48.6 GB/s)
│ Ridge point : 36.0 FLOP/byte
│ Detection source: Apple Silicon table (Apple M3) + NumPy STREAM triad
└────────────────────────────────────────────────────────────────────────────
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
FF DNN (784→128→128→10, LayerNorm)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Framework FLOPs Params I/O MB AI ms(med) ±ms effic. roofline GFLOP/s GB/s bound
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
PyTorch 3.8M 475,176 0.6 6.45 0.089 0.059 6.6% [█░░░░░░░░░░░░░░░░░] 42.44 6.58 memory
JAX 3.8M 472,064 0.6 6.43 0.068 0.006 8.6% [██░░░░░░░░░░░░░░░░] 55.26 8.60 memory
TensorFlow 3.8M 475,176 0.6 6.45 1.844 0.124 0.3% [░░░░░░░░░░░░░░░░░░] 2.06 0.32 memory
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
CNN (3-ch input → Conv64 → Conv128 → GAP → Dense10, 32×32)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Framework FLOPs Params I/O MB AI ms(med) ±ms effic. roofline GFLOP/s GB/s bound
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
PyTorch 668.5M 309,288 20.4 32.71 5.046 0.393 4.0% [█░░░░░░░░░░░░░░░░░] 132.48 4.05 memory
JAX 662.2M 306,944 25.7 25.78 2.406 0.202 10.7% [██░░░░░░░░░░░░░░░░] 275.27 10.68 memory
TensorFlow 670.1M 310,824 27.8 24.12 5.226 0.377 5.3% [█░░░░░░░░░░░░░░░░░] 128.23 5.32 memory
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
RNN (2-layer Vanilla RNN, hidden=128, seq=32)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Framework FLOPs Params I/O MB AI ms(med) ±ms effic. roofline GFLOP/s GB/s bound
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
PyTorch 67.1M 267,304 2.4 28.29 1.019 0.176 2.3% [░░░░░░░░░░░░░░░░░░] 65.89 2.33 memory
JAX 33.7M 136,192 6.6 5.14 1.678 0.031 3.9% [█░░░░░░░░░░░░░░░░░] 20.10 3.91 memory
TensorFlow 201.4M 797,736 5.0 40.32 71.243 6.103 0.1% [░░░░░░░░░░░░░░░░░░] 2.83 0.07 compute
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
LSTM (2-layer LSTM, hidden=128, seq=32)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Framework FLOPs Params I/O MB AI ms(med) ±ms effic. roofline GFLOP/s GB/s bound
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
PyTorch 268.5M 1.1M 6.3 42.58 2.856 0.091 2.6% [░░░░░░░░░░░░░░░░░░] 94.00 2.21 compute
JAX 135.5M 529,408 31.5 4.31 4.426 0.134 7.1% [█░░░░░░░░░░░░░░░░░] 30.61 7.11 memory
TensorFlow 268.5M 1.1M 6.3 42.58 43.502 1.862 0.2% [░░░░░░░░░░░░░░░░░░] 6.17 0.14 compute
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Transformer (2-layer encoder, embed=128, heads=4, FFN×4, seq=32)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Framework FLOPs Params I/O MB AI ms(med) ±ms effic. roofline GFLOP/s GB/s bound
───────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
PyTorch 421.4M 1.5M 9.5 44.59 1.551 0.211 7.5% [█░░░░░░░░░░░░░░░░░] 271.69 6.09 compute
JAX 201.9M 791,552 10.3 19.57 1.206 0.032 8.6% [██░░░░░░░░░░░░░░░░] 167.38 8.55 memory
TensorFlow 421.4M 1.6M 9.5 44.59 12.630 0.182 0.9% [░░░░░░░░░░░░░░░░░░] 33.37 0.75 compute
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Per-Architecture × Per-Framework Summary
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Architecture PyTorch JAX TensorFlow
────────────────────────────────────────────────────────────────────────────────────────────────────────
FF DNN 7.7% eff 0.08ms 3.8M FLOPs 10.0% eff 0.06ms 3.8M FLOPs 0.3% eff 1.98ms 3.8M FLOPs
CNN 3.9% eff 5.29ms 668.5M FLOPs 11.3% eff 2.28ms 662.2M FLOPs 5.1% eff 5.41ms 670.1M FLOPs
RNN 2.3% eff 1.02ms 67.1M FLOPs 3.9% eff 1.68ms 33.7M FLOPs 0.1% eff 71.24ms 201.4M FLOPs
LSTM 2.6% eff 2.86ms 268.5M FLOPs 7.1% eff 4.43ms 135.5M FLOPs 0.2% eff 43.50ms 268.5M FLOPs
Transformer 7.5% eff 1.55ms 421.4M FLOPs 8.6% eff 1.21ms 201.9M FLOPs 0.9% eff 12.63ms 421.4M FLOPs
Reading the table All architectures are memory-bound on this CPU (AI < ridge point of 36 FLOP/byte). Low roofline efficiency across all frameworks reflects framework dispatch overhead and small-batch latency — the expected regime for CPU inference. JAX's eager XLA compilation delivers the most consistent efficiency across architectures. TensorFlow's eager Python dispatch overhead dominates at small batch sizes, especially for sequential (RNN/LSTM) workloads.
Matrix-multiply workload comparison (original)
For a quick cross-framework sanity check on plain matmul shapes, run:
pip install "neural-cost[torch,jax,tensorflow]"
# Or from local editable checkout:
# pip install -e '.[torch,jax,tensorflow]'
python examples/compare_frameworks.py
Optional overrides with known hardware specs:
python examples/compare_frameworks.py --peak-flops 312e12 --memory-bandwidth 1.6e12
Custom frameworks
Subclass FrameworkAdapter and implement operations(model, example_inputs)
to return portable Operation records. The adapter can also override
benchmark to synchronize an accelerator or collect framework-specific memory
statistics. This contract keeps model extraction separate from the framework-
independent estimator and analyzer.
CLI Suite
neural-cost provides a unified command-line suite for hardware detection, theoretical profiling, empirical benchmarking, LLM serving analysis, and CI/CD budget assertions:
# General help and subcommands:
neural-cost --help
1. Hardware Detection & STREAM Benchmark
Probe attached accelerators, lookup Apple Silicon / NVIDIA published specs, and measure live NumPy STREAM memory bandwidth:
# Live probe with memory bandwidth benchmark:
neural-cost hardware
# Instant detection without STREAM benchmark:
neural-cost hardware --no-bench
# Machine-readable JSON output:
neural-cost hardware --json
2. Static Architecture Profiling
Compute forward FLOPs, parameter counts, KV cache size, activation memory, arithmetic intensity, and theoretical latency floor on detected or target hardware:
# Profile a 7B Transformer (Llama-style) at batch size 8:
neural-cost profile --arch transformer --batch-size 8 --seq-len 4096 --embed-dim 4096 --num-heads 32 --num-kv-heads 8 --num-layers 32 --dtype fp16
# Profile a quantized 70B Transformer with W4A16 AWQ and launch floor:
neural-cost profile --arch transformer --num-layers 80 --embed-dim 8192 --num-heads 64 --quantization w4a16_awq --num-kernels 640
# Profile training pass with selective activation checkpointing and Adam moments:
neural-cost profile --arch transformer --batch-size 4 --seq-len 4096 --embed-dim 4096 --num-layers 32 --checkpointing selective --training
# Profile an 8x7B Mixture-of-Experts layer:
neural-cost profile --arch moe --num-experts 8 --top-k 2 --embed-dim 4096 --expert-hidden-dim 14336 --decode
# Profile PagedAttention KV cache with shared prefix caching:
neural-cost profile --arch paged-attention --batch-size 32 --seq-len 2048 --block-size 16 --shared-prefix-tokens 128
# Profile State Space Model (Mamba) recurrent state and constant memory:
neural-cost profile --arch ssm --batch-size 1 --seq-len 4096 --embed-dim 4096 --num-layers 32
3. Empirical Benchmarking
Benchmark real execution latency, achieved compute throughput (GFLOP/s), achieved DRAM bandwidth (GB/s), and roofline efficiency on your local machine:
# Microbenchmark matrix multiplication:
neural-cost benchmark --arch linear --batch-size 64 --in-features 1024 --out-features 1024
# Benchmark MoE decode step:
neural-cost benchmark --arch moe --batch-size 1 --num-experts 8 --top-k 2
# Benchmark PagedAttention kernel:
neural-cost benchmark --arch paged-attention --batch-size 32 --seq-len 1024 --block-size 16
# Benchmark live NumPy STREAM memory bandwidth:
neural-cost benchmark --arch stream
4. CI Budget Gate (audit)
Enforce automated performance budgets and VRAM ceilings in your CI/CD pipelines. Exits with status 0 if all assertions pass, or 1 if any budget is exceeded:
# Fail CI if model exceeds 16GB VRAM or 50ms latency floor:
neural-cost audit \
--arch transformer \
--batch-size 1 \
--seq-len 2048 \
--embed-dim 4096 \
--num-heads 32 \
--num-layers 32 \
--max-vram 16GB \
--max-latency-ms 50.0
# JSON output for CI pipeline reporting:
neural-cost audit --arch mlp --in-features 1024 --out-features 1024 --max-vram 1GB --json
5. Modern LLM Serving Primitives (llm)
Dedicated analytical calculators for production serving architectures:
# Mixture-of-Experts routing & memory traffic:
neural-cost llm moe --batch-size 1 --num-experts 8 --top-k 2 --decode
# Speculative decoding analytical breakeven solver:
neural-cost llm speculative --gamma 4 --acceptance-rate 0.75
# PagedAttention KV cache capacity and fragmentation:
neural-cost llm paged --batch-size 32 --context-len 2048 --block-size 16
# Continuous batching mixed iteration cost simulator:
neural-cost llm continuous --decode-streams 64 --prefill-tokens 512
# State Space Model (Mamba/S6/SSD) recurrent cost and KV elimination analyzer:
neural-cost llm ssm --batch-size 1 --seq-len 32768 --embed-dim 4096 --num-layers 32 --decode
# Sub-byte quantization and dequantization ALU unpacking tax (AWQ, GPTQ, FP8, NVFP4):
neural-cost llm quant --batch-size 1 --in-features 4096 --out-features 4096 --quantization w4a16_awq
# Kernel launch floor & CUDA Graph / torch.compile speedup analyzer:
neural-cost llm launch-floor --num-kernels 640 --arithmetic-floor-ms 0.5 --launch-overhead-us 5.0 --measured-latency-ms 3.5
# Activation checkpointing & selective recomputation (none, full, selective) analyzer:
neural-cost llm checkpointing --strategy selective --batch-size 4 --seq-len 4096 --embed-dim 4096 --vram-gb 80.0
6. Multi-Framework Comparison (compare)
Run head-to-head empirical benchmarks across PyTorch, JAX, and TensorFlow (backward-compatible with neural-cost-compare):
# Multi-framework benchmark runner:
neural-cost compare
# Or using the legacy binary:
neural-cost-compare --warmup 10 --repeats 30
7. Distributed Cluster & 3D Parallelism (distributed)
Model cluster scaling efficiency, network communication volume, and MFU for frontier models:
# Model Llama-3 70B training across 16 nodes of 8x H100 with TP=8, PP=4, DP=4:
neural-cost distributed \
--parameters 70e9 \
--batch-size 128 \
--seq-len 4096 \
--embed-dim 8192 \
--num-layers 80 \
--num-nodes 16 \
--devices-per-node 8 \
--tp 8 \
--pp 4 \
--dp 4 \
--microbatches 8 \
--dp-mode zero3_fsdp \
--intra-node nvlink4 \
--inter-node infiniband_ndr
# Output cluster performance and communication breakdown as JSON:
neural-cost distributed --parameters 7e9 --batch-size 32 --tp 2 --dp 4 --json
Current scope
The package profiles concrete-shape dense, matrix-multiply, convolution,
embedding lookup, multi-head attention, and common elementwise inference graphs,
alongside softmax, layernorm, batchnorm, and pooling. Recurrent layers
(RNN, GRU, LSTM) are modelled as their constituent input→hidden and
hidden→hidden linear projections. Static training storage includes gradients
and optimizer state but does not yet trace a full backward graph. Activation
checkpointing, distributed communication, dynamic shapes, fusion details,
complete graph coverage, and non-PyTorch kernel-level traces remain deliberate
next increments rather than silently approximated.
Development
Format and lint the codebase with a single command:
# Auto-format and fix linting in place:
python scripts/format.py
# Or using the shell wrapper:
./scripts/format.sh
# Or via Make:
make format
# Check formatting and linting without modifying files:
python scripts/format.py --check
# Or:
make check
Automated Semantic Versioning & Release Pipeline
neural-cost uses an automated semantic versioning and release pipeline in GitHub Actions (.github/workflows/ci.yml and .github/workflows/publish.yml) powered by paulhatch/semantic-version with path filtering and PyPI trusted publishing.
Version upgrades are determined strictly by Conventional Commits that modify files in the distributed library (src/, pyproject.toml):
| Change Kind | Commit Conventional Type | Affects Library (src/)? |
Version Increment | Release Action |
|---|---|---|---|---|
| Breaking change | feat!:, fix!:, BREAKING CHANGE: |
✅ Yes | Major (X.0.0) |
Create tag & publish PyPI release |
| New feature | feat:, feat(scope): |
✅ Yes | Minor (0.X.0) |
Create tag & publish PyPI release |
| Bug fix / performance | fix:, perf:, refactor: |
✅ Yes | Patch (0.0.X) |
Create tag & publish PyPI release |
| Non-library changes | docs:, chore:, ci:, test:, benchmarks, workflows |
❌ No | None (unchanged) | No tag created; PyPI publish skipped |
| Doc updates to library | docs:, style: (docstrings only) |
✅ Yes | None (unchanged) | No tag created; PyPI publish skipped |
- In CI (PRs): The
semver-checkworkflow job automatically analyzes commits and library paths in the PR and prints a summary indicating whether a version bump is required and what the new version will be. - On Push to
main: Thepublishworkflow triggers on changes tosrc/**orpyproject.toml. If a version increment is detected, it automatically creates and pushes the git tag, generates the GitHub Release, builds sdist & wheel, and publishes to PyPI.
Metadata
Release files for neural-cost 0.9.0
For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.
Source distribution (sdist)
| File | Size | Uploaded | Provenance | Actions |
|---|---|---|---|---|
| neural_cost-0.9.0.tar.gz | 1.6 MB | With provenance | Details for neural_cost-0.9.0.tar.gz |
Built distribution (wheel)
| File | Interpreter | ABI | Platform | Actions Reset |
|---|---|---|---|---|
| neural_cost-0.9.0-py3-none-any.whl | Python 3 | none | any | Details for neural_cost-0.9.0-py3-none-any.whl |
Total release size: 1.7 MB
Release files / neural_cost-0.9.0.tar.gz
| Download URL | neural_cost-0.9.0.tar.gz |
|---|---|
| Size | 1.6 MB |
| Tags | Source |
|
SHA-256 checksum How to use checksums |
3ae75a020aa03bd9cb01fbe2e2b4087b18f3eab5d95b938e0c098598254d8405
|
|
BLAKE2b-256 checksum How to use checksums |
9327caf025b75d754abbb98f07a9ded865c3ce1b3c8aab0ffcde9756cbacf86c
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 5, 2026.
Transparency logRelease files / neural_cost-0.9.0-py3-none-any.whl
| Download URL | neural_cost-0.9.0-py3-none-any.whl |
|---|---|
| Size | 101.7 kB |
| Tags | Python 3 |
|
SHA-256 checksum How to use checksums |
e2a5c194675c1fe367a490fd1d4f61415bcbb98fcc8b69f006b8d8f9a425a99e
|
|
BLAKE2b-256 checksum How to use checksums |
ef200404f71ea86608c027d48296ccaa2cc09ffde514264bea39d926112ea122
|
| Upload date | |
|
Uploaded using Trusted Publishing? What is trusted publishing? |
Yes |
| Uploaded via |
twine/7.0.0 CPython/3.13.14
|
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 5, 2026.
Transparency logSecurity for neural-cost 0.9.0
Provenance
Provenance describes where a file came from. On PyPI, provenance is shared via attestations, which provide a verifiable record of the build or publishing details. View details, limitations and caveats.
Full provenance
Both files in this release have been uploaded with provenance metadata.
Attestation for neural_cost-0.9.0.tar.gz
PyPI Publish Attestation
PyPI verified that this artifact, at this checksum, originated from the publisher listed below.
Signed by GitHub Actions, verified by PyPI on Oct 5, 2026.
Transparency logA note on package security
PyPI is a repository for distributing packages, not a service for auditing them. While third-party security researchers and our staff do their best to detect and remove malicious packages, PyPI cannot guarantee the quality, security or suitability of any given package.
To learn more about protecting yourself from malicious actors, visit our package security documentation.