Menu

#DevtoPytorch

93 posts

Feed·
20 of 93 posts
How do you fix CUDA out of memory errors in PyTorch?
🖼️
0Read More

How do you fix CUDA out of memory errors in PyTorch?

DEV Community: pytorch·Viswaretas Kotra·about 23 hours ago
#1ibrEuuJ
#devtopytorch#devto#dev#memory#batch

Read what the OOM error says, then cut the biggest consumer: activations, optimizer state or fragmentation. An ordered checklist with GB math and code.

15s
How do you keep training alive now that Runpod spot Pods are gone?
🖼️
0Read More

How do you keep training alive now that Runpod spot Pods are gone?

DEV Community: pytorch·Viswaretas Kotra·about 23 hours ago
#Pap2hVHp
#devtopytorch#devto#dev#runpod#nodus

Runpod retired spot Pods. Where checkpoints should live on Runpod today, what a stop costs, and how Nodus restores interrupted runs for you.

15s
Build and Verify Chunked Cosine Search with Python and PyTorch on a Rented GPU
📰
0Read More

Build and Verify Chunked Cosine Search with Python and PyTorch on a Rented GPU

DEV Community: pytorch·CloudGPU·5 days ago
#XRueKei0
#devtopytorch#devto#dev#torch#device

A seeded Python fixture, blockwise GPU top-k search, and a float64 CPU oracle for checking vector retrieval.

15s
Why does my PyTorch DataLoader fail when a worker touches CUDA?
📰
0Read More

Why does my PyTorch DataLoader fail when a worker touches CUDA?

DEV Community: pytorch·CloudGPU·5 days ago
#ROK2tX6P
#devtopytorch#devto#dev#cuda#dataset

I work on CloudGPU. Keep dataset decoding and collation on the CPU, then move each batch to CUDA in...

15s
Reproducing a paper in Python: a random seed is only the first step
📰
0Read More

Reproducing a paper in Python: a random seed is only the first step

DEV Community: pytorch·wanqiuxu·5 days ago
#MEMBOVQe
#devtopytorch#devto#dev#seed#random

You follow a machine learning or experimental paper, run the code, and the loss curve or the reported...

15s
Your PyTorch Code on Trainium: What Actually Breaks
📰
0Read More

Your PyTorch Code on Trainium: What Actually Breaks

DEV Community: pytorch·RobustTrueTry·6 days ago
#p5NWJaHx
#devtopytorch#devto#dev#model#pytorch

A practitioner's porting checklist for PyTorch on AWS Trainium: compile workflow, unsupported ops, and the training-loop pitfalls that stall your step

15s
From Lab to Code #4: Exploring Neural Network Efficiency with Python and PyTorch
📰
0Read More

From Lab to Code #4: Exploring Neural Network Efficiency with Python and PyTorch

DEV Community: pytorch·Ertugrul·6 days ago
#py1iuGNs
#devtopytorch#devto#dev#accuracy#model

A larger neural network can often achieve better accuracy. But does that always make it the better...

15s
BF16 FP8: checking precision across two CPU compiler versions
📰
0Read More

BF16 FP8: checking precision across two CPU compiler versions

DEV Community: pytorch·Blucca·8 days ago
#7Xf9Lnwi
#devtopytorch#devto#dev#torch#fp32

A dtype conversion can remain visible after the final FP8 quantization. It belongs in a fusion's...

15s
Warmup Did Nothing for My Learning Rate — Then It Fully Rescued the Wrong One
📰
0Read More

Warmup Did Nothing for My Learning Rate — Then It Fully Rescued the Wrong One

DEV Community: pytorch·b1ackjack12·11 days ago
#p60Jc4a0
#devtopytorch#devto#dev#warmup#rate

Learning-rate warmup is the most cargo-culted schedule fragment in deep learning: transformers need it, so everyone ships it. I measured

15s
My Gradient Clipping Was Never Clipping. It Was Secretly Multiplying My Weight Decay
📰
0Read More

My Gradient Clipping Was Never Clipping. It Was Secretly Multiplying My Weight Decay

DEV Community: pytorch·b1ackjack12·11 days ago
#99DLnCHo
#devtopytorch#devto#dev#clipping#decay

Yesterday warmup turned out to be free insurance against a hot learning rate, so today I asked whether gradient clipping — the other famous

15s
Weight Decay 0.05: Catastrophe in Adam, Optimum in AdamW
📰
0Read More

Weight Decay 0.05: Catastrophe in Adam, Optimum in AdamW

DEV Community: pytorch·b1ackjack12·13 days ago
#geaZvlkd
#devtopytorch#devto#dev#adam#adamw

The dose-response series reaches the knob I've been inheriting since September: weight_decay=5e-4, copied from SGD folklore into Adam

15s