PR #15727 convert : use reflinks for faster conversion

convert : use reflinks for faster conversion #15727

compilade wants to merge 13 commits into compilade/convert-safetensors-parse from compilade/convert-reflinks

compilade added demo

compilade added python

github-actions added ggml

compilade force-pushed the compilade/convert-safetensors-parse branch from 786b32d8 to e582f1ac 155 days ago

compilade force pushed to 833d03c2 155 days ago

compilade force-pushed the compilade/convert-safetensors-parse branch from e582f1ac to e996f3ae 96 days ago

convert : use reflinks for faster conversion

562aa42c

convert : fix reflinks for stacked MoE tensors

d9210570

gguf-py : fix flake8 lint

791bd97b

convert : detect filesystem block size for reflinks

c3738cfc

convert : use F32 operations on Mamba A_log

614b95a8

convert : allow sharding reflinked models

d3fcb0e9

gguf-py : improve reflink size logging

5712aa89

convert : more robust default ftype detection

e097d98a

convert : remove unused field ModelTensorInfo.src_qtype

3126b5ee

gguf-py : allow previewing reflinked size on non-Linux platforms

6ffa46d8

convert : better logging of partially reflinkable tensors

4be1a5d4

gguf-py : handle cross-filesystem file range copies

f88a4b93

convert : for FP8, use scale type to decide auto type

2ef41855

compilade force pushed from 833d03c2 to 2ef41855 96 days ago

Reviewers

No reviews

Assignees

No one assigned

Labels

demo python ggml

Milestone

No milestone