Gian-Carlo Pascutto

More IQ! Errors

8

This is a probably a different problem than the previous report. 1) It happens much more frequently after the latest VA4 firmware update. (https://forums.garmin.com/sports-fitness/healthandwellness/f/vivoactive-4-series/238197/phased-rollout-20---vivoactive-4-all-versions-sw-5-10) 2) Swiping to another widget and...

IQ! Error

13

I've been using this for a few weeks. Tonight, instead of the watchface I started just getting an "IQ!" warning screen. Resetting nor hard resetting the watch works. Switching to...

[Bug]: Off-by-one error with sampler steps causing v_prediction to fail

1

### What happened? Loaded a saved config, edited samples, started training. ### What did you expect would happen? Successful sample generation. ### Relevant log output ```shell Traceback (most recent call...

DistributedDataParallel with compile(..., mode="max-autotune") hangs in 2.5+

2

### 🐛 Describe the bug I have a DCNN-like network that I'm training with DDP and 2 GPUs. This always got a speedup from ```compile(net, mode="max-autotune")```. However, since PyTorch 2.5+...

Fixes #10976.

cuda: Add Q5_1, Q5_0, Q4_1 and Q4_0 to F32 conversion support. (#10976)

1

Using templates and reusing the `dequant_qX_Y` functions.

metal: Copy kernels for quant to F32 conversions (#10976).

1

Modeled after the CUDA implementations. Because of the use of `type4x4` I had no idea how to reuse the existing `dequantize` functions, so those are repeated here in `float` form....