FazBrowse GitHub Viewer
|
Trending
|
URL:
|
Home
Tools:
[Download Repo ZIP]
[Original HTTPS Page]
Adds support for 4bit (nf4) and 8bit bitsandbytes quantization (3/3) by Rypo · Pull Request #151 · VectorSpaceLab/OmniGen · GitHub
Uh oh!
There was an error while loading.
Please reload this page
.
VectorSpaceLab
/
OmniGen
Public
Notifications
You must be signed in to change notification settings
Fork
363
Star
4.3k
Code
Issues
139
Pull requests
7
Actions
Projects
Security and quality
0
Insights
Additional navigation options
Code
Issues
Pull requests
Actions
Projects
Security and quality
Insights
Adds support for 4bit (nf4) and 8bit bitsandbytes quantization (3/3)
- #151
#151
Open
Rypo
wants to merge 15 commits into
VectorSpaceLab:main
VectorSpaceLab/OmniGen:main
from
Rypo:bnb_quantization
Rypo/OmniGen:bnb_quantization
Copy head branch name to clipboard
Conversation
Commits
15
(15)
Checks
Files changed
Open
Adds support for 4bit (nf4) and 8bit bitsandbytes quantization (3/3)
#151
Rypo
wants to merge 15 commits into
VectorSpaceLab:main
VectorSpaceLab/OmniGen:main
from
Rypo:bnb_quantization
Rypo/OmniGen:bnb_quantization
Copy head branch name to clipboard
Commits
Commits on Nov 26, 2024
feat: fast model loading with accelerate
Show description for 387f48c
Rypo
committed
387f48c
View commit details
Copy full SHA for 387f48c
Browse repository at this point
fix: avoid moving model to device prematurely
Rypo
committed
0287b50
View commit details
Copy full SHA for 0287b50
Browse repository at this point
fix: typo
Rypo
committed
53794a0
View commit details
Copy full SHA for 53794a0
Browse repository at this point
Commits on Dec 2, 2024
feat: add 4bit and 8bit quantization support with bitsandbytes
Show description for 889b6b9
Rypo
committed
889b6b9
View commit details
Copy full SHA for 889b6b9
Browse repository at this point
feat: add cli arg to gradio demo for nbit quantization
Rypo
committed
8ea2d6d
View commit details
Copy full SHA for 8ea2d6d
Browse repository at this point
Commits on Dec 3, 2024
feat: support quantization with prequantized weights, add auto detection for bnb quant dict
Rypo
committed
8d71606
View commit details
Copy full SHA for 8d71606
Browse repository at this point
feat: support loading prequantized weights from the hub
Rypo
committed
7ba2cbc
View commit details
Copy full SHA for 7ba2cbc
Browse repository at this point
Commits on Dec 5, 2024
fix: use default model config if passing weights file directly
Rypo
committed
caac377
View commit details
Copy full SHA for caac377
Browse repository at this point
refactor: use to HFQuantizer preprocessing in place of manually preprocessing, skip quant norm layers
Rypo
committed
b066b34
View commit details
Copy full SHA for b066b34
Browse repository at this point
fix: prevent device/dtype changes to quantized models
Rypo
committed
6d5b4bc
View commit details
Copy full SHA for 6d5b4bc
Browse repository at this point
refactor!: remove quantization_config argument from OmniGenPipeline.from_pretrained
Show description for ff20f1e
Rypo
committed
ff20f1e
View commit details
Copy full SHA for ff20f1e
Browse repository at this point
Commits on Dec 6, 2024
Merge branch 'main' into bnb_quantization
Rypo
committed
7a662d9
View commit details
Copy full SHA for 7a662d9
Browse repository at this point
Commits on Dec 12, 2024
fix: prevent float16 numerical overflow
Show description for d75af76
Rypo
committed
d75af76
View commit details
Copy full SHA for d75af76
Browse repository at this point
Commits on Dec 17, 2024
perf(fp16): reduce expected extra required iterations to 1.
Show description for 6ce30f2
Rypo
committed
6ce30f2
View commit details
Copy full SHA for 6ce30f2
Browse repository at this point
Merge branch 'main' into bnb_quantization
Rypo
committed
0acd89e
View commit details
Copy full SHA for 0acd89e
Browse repository at this point
Back
|
FazBrowse Home
|
New Git URL