Skip to content

[OMNIML-6063] [2/3] Decompress compressed-tensors MoE experts as they load #3137

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
shengliangxu wants to merge 20 commits into
base: shengliangx/hfep-6a2-ep-loader-core
Choose a base branch
from
Open

[OMNIML-6063] [2/3] Decompress compressed-tensors MoE experts as they load #3137

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
20 commits
Select commit Hold shift + click to select a range
17884f6
Decompress compressed-tensors MoE experts as they load
shengliangxu Oct 30, 2026
d7c4d42
Import _SUB_CONFIG_ATTRS at module level in the decompressor
shengliangxu Oct 30, 2026
8fc97d8
Install compressed-tensors in every unit session
shengliangxu Oct 30, 2026
79cb8e6
Refuse an unmarked config by name when checking the stored tensors
shengliangxu Oct 30, 2026
835af6c
Decompress with the one config group's own format, and refuse one wit…
shengliangxu Oct 30, 2026
0d47a96
Refuse on-the-fly decompression for a load in a dtype other than bf16
shengliangxu Oct 30, 2026
4ef070a
Compress the decompressor's docstrings and comments
shengliangxu Oct 30, 2026
41acc0b
Reverse a decompressing converter as the fusion it rewrote
shengliangxu Oct 30, 2026
2c90e04
Refuse a checkpoint whose compressed weights store different names
shengliangxu Oct 30, 2026
c2756cc
Keep integer sources native only in a decompressing load
shengliangxu Oct 30, 2026
b647b1d
Refuse a dtype other than bf16 at any config level
shengliangxu Oct 30, 2026
cdc2e24
Import the decompressor's dependency-light helpers at module level
shengliangxu Oct 30, 2026
5f9f379
Refuse unknown stored names and non-weight sources before a tensor is…
shengliangxu Oct 30, 2026
0a711b9
Keep packed integers uncast where an expert-parallel load shards them
shengliangxu Oct 30, 2026
6cd12d7
Rewrite experts' weights only, let an expert's bias load, check the f…
shengliangxu Oct 30, 2026
485f6b6
Start each decompressor test with no decompressing load under way
shengliangxu Oct 30, 2026
5d0928d
Say what the stored-name check covers, and why a grouped bias is refused
shengliangxu Oct 30, 2026
32ca3d3
Import the DTensor helpers at module scope in the decompressor's tests
shengliangxu Oct 30, 2026
7e678d2
Tighten the decompressor's docstrings to keep the change within its size
shengliangxu Oct 30, 2026
9deea3b
Return the tensor names check_compressed_keys read, for its caller to…
shengliangxu Oct 30, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Loading
Loading