maxtext
maxtext copied to clipboard
Rebase experimental_proxy on main
experimental_proxy only has 2 commits on it since branching:
$ git log --no-merges origin/experimental_proxy ^origin/main
commit 13f519e39e0d904e320c7d8a472161e4bcf03408 (HEAD -> avritt/noocdbt, origin/vivianrwu_experimental_proxy, origin/experimental_proxy, experimental_proxy)
Author: Zhihao Shan <[email protected]>
Date: Mon Sep 9 15:15:08 2024 -0700
correct model name for llama-2 70b
commit f8980b6a4d5cb01fa07258bda88b7497bdc01320
Author: Zhihao Shan <[email protected]>
Date: Thu Sep 5 11:30:47 2024 -0700
Add llama3 70b inference experimental support on v5e-16
In addition, fetch the KV cache sharding from model forward in decode
phase and use different sharding for prefill and decode.
We want https://github.com/AI-Hypercomputer/maxtext/commit/6c749c7e624ebb6d4a24716c2aa1dcc75f006b91 changes in experimental_proxy branch. Rebasing was easier than cherry-picking.
Steps:
git fetch origin
git branch -c avritt/noocdbt
git reset --hard origin/experimental_proxy
git rebase -X theirs --onto origin/main HEAD~2