Search results

Why autocast in checkpoint_activation uses default dtype?

Status: Open.

#1199 In facebookresearch/fairscale;

· psolt-leonardo opened on Aug 4, 2025

Support free-threading Python wheel distributions

Status: Open.

#1197 In facebookresearch/fairscale;

· andfoy opened on Apr 21, 2025

Error running large model of esm15B using FSDP

Status: Open.

#1194 In facebookresearch/fairscale;

· BKRBH opened on Feb 19, 2025

AttributeError: 'FlatParameter' object has no attribute '_full_param_padded'

Status: Open.

#1193 In facebookresearch/fairscale;

· hui-li-xf opened on Jan 19, 2025

When mixed precision input contains grad tensor, FSDP cast it with no grad

Status: Open.

#1191 In facebookresearch/fairscale;

· kamwoh opened on Dec 8, 2024

Hi, Groups division may be incorrect in initialize() in fairscale/nn/model_parallel/initialize.py

Status: Open.

#1189 In facebookresearch/fairscale;

· Youngluc opened on Aug 30, 2024

Raising assert param.grad is not None when finetuning LoRA.

Status: Open.

#1188 In facebookresearch/fairscale;

· HashimotoPatrickMu opened on Jun 26, 2024

[question] Different training between DDP & Sharded DDP

Status: Open.

#1172 In facebookresearch/fairscale;

· kwohlfahrt opened on Mar 29, 2024

Example of MOE

Status: Open.

#1165 In facebookresearch/fairscale;

· Juanhui28 opened on Feb 28, 2024

How can I use torchrun + model parallelism + FSDP

Status: Open.

#1155 In facebookresearch/fairscale;

· HackGiter opened on Dec 8, 2023

It is dangerous to using default non_block=True.

Status: Open.

#1146 In facebookresearch/fairscale;

· heshenghuan opened on Oct 27, 2023

assert self.has_full_params

Status: Open.

#1134 In facebookresearch/fairscale;

· pokameng opened on Sep 11, 2023