Commits · Dovakiins/qwerrwe

don't compile deepspeed or bitsandbytes from source (#837)

f544ab2
unverified

winglian commited on Nov 9, 2023

add deepspeed-kernels dependency for deepspeed>=0.12.0 (#827)

8056ecd
unverified

fpreiss commited on Nov 5, 2023

fix pytorch 2.1.0 build, add multipack docs (#722)

2aa1f71
unverified

winglian commited on Oct 13, 2023

apex not needed as amp is part of pytorch (#696)

aca0398
unverified

winglian commited on Oct 7, 2023

fix multiline for docker (#694)

de87ea6
unverified

winglian commited on Oct 7, 2023

Feat: Set WORKDIR to /workspace/axolotl (#679)

133e676
unverified

Nanobit commited on Oct 5, 2023

tweak: improve base builder for smaller layers (#500)

923eb91
unverified

Maxime commited on Sep 22, 2023

let MAX_JOBS use the default since we're not resource constrained on our self-hosted runners (#427)

e85d2eb
unverified

winglian commited on Sep 22, 2023

update dockerfile to not build evoformer since it fails the build (#607)

b53e777
unverified

winglian commited on Sep 19, 2023

update readme to point to direct link to runpod template, cleanup install instrucitons (#532)

34c0a86
unverified

winglian commited on Sep 8, 2023

Add support for GPTQ using native transformers/peft (#468)

3355706
unverified

winglian commited on Sep 5, 2023

remove --force-reinstall from Dockerfile to ensure correct pytorch version (#492)

e356b29
unverified

tmm1 commited on Aug 29, 2023

flash attn pip install (#426)

cf66547
unverified

mhenrichsen Ubuntu mhenrichsen Mads Henrichsen

winglian commited on Aug 18, 2023

bump flash-attn to 2.0.4 for the base docker image (#382)

ffac902
unverified

winglian commited on Aug 13, 2023

add peft install back since it doesn't get installed by setup.py (#331)

db2a358
unverified

winglian commited on Jul 31, 2023

pin accelerate so it works with llama2 (#330)

6c9a87c
unverified

winglian commited on Jul 31, 2023

Prune cuda117 (#327)

2c37bf6
unverified

winglian commited on Jul 26, 2023

add runpod envs to .bashrc, fix bnb env (#316)

cf62cfd
unverified

winglian commited on Jul 22, 2023

pin flash attention 2 to the fix for backwards pass

cdf85fd

winglian commited on Jul 21, 2023

flash attention 2

9b790d3

winglian commited on Jul 20, 2023

explicitly pin flash attention 1 to v1.0.9

b06d3e3

winglian commited on Jul 20, 2023

misc fixes

d75adb9

winglian commited on Jul 17, 2023

set transformers cache env var in docker image

f162f3c

winglian commited on Jul 17, 2023

git fetch fix for docker

eca3531

winglian commited on Jul 17, 2023

pin pydantic so deepspeed isn't broken

7145695

winglian commited on Jul 3, 2023

update pip install command for apex

530809f

winglian commited on Jun 29, 2023

shallow clone

5cd2126

winglian commited on Jun 2, 2023

clone in docker

12620f3

winglian commited on Jun 2, 2023

py310, fix cuda arg in deepspeed

c43c5c8

winglian commited on May 30, 2023

Merge pull request #108 from OpenAccess-AI-Collective/docker-gptq

bbc5bc5
unverified

winglian commited on May 30, 2023

Lint and format

392dfd9

Nanobit commited on May 28, 2023

cleanup from pr feedback

48612f8

winglian commited on May 30, 2023

default to qlora support, make gptq specific image

6ef96f5

winglian commited on May 30, 2023

move CUDA_VERSION_BNB arg inside of stage build scope

e43bcc6

winglian commited on May 29, 2023

fix CUDA_VERSION_BNB env var

00323f0

winglian commited on May 29, 2023

bnb fixes

21f17cc

winglian commited on May 29, 2023

use python setup install, bdist wheel is unreliable in installing extension

809cceb

winglian commited on May 28, 2023

ensure libbitsandbytes*.so gets included with wheel

a798ba1

winglian commited on May 28, 2023

fix missing run coninuation

cf37980

winglian commited on May 27, 2023

update docker to compile latest bnb to properly support qlora

312b8d5

winglian commited on May 27, 2023

fix wheel install glob

c3d2562

winglian commited on May 26, 2023

fix wheel blobs in dockerfile

1fc9b44

winglian commited on May 26, 2023

fix xentropy wheel name typo

259262b

winglian commited on May 25, 2023

fix relative path in flash-attn build:

8d6a289

winglian commited on May 25, 2023

fix cd within flash-attn

f5fa3d1

winglian commited on May 25, 2023

optionally be able to specify alpaca or chat style prompts

1d5ab84

winglian commited on May 20, 2023

update entrypoint and force min accelerate

fa8bd14

winglian commited on May 18, 2023

use printf instead of echo in dockerfile for portability

709be5a

winglian commited on May 8, 2023

fix path name to sorkspace

807cca8

winglian commited on May 8, 2023

setup runpod images

79deb35

winglian commited on May 8, 2023

Commit History

don't compile deepspeed or bitsandbytes from source (#837) f544ab2 unverified

add deepspeed-kernels dependency for deepspeed>=0.12.0 (#827) 8056ecd unverified

fix pytorch 2.1.0 build, add multipack docs (#722) 2aa1f71 unverified

apex not needed as amp is part of pytorch (#696) aca0398 unverified

fix multiline for docker (#694) de87ea6 unverified

Feat: Set WORKDIR to /workspace/axolotl (#679) 133e676 unverified

tweak: improve base builder for smaller layers (#500) 923eb91 unverified

let MAX_JOBS use the default since we're not resource constrained on our self-hosted runners (#427) e85d2eb unverified

update dockerfile to not build evoformer since it fails the build (#607) b53e777 unverified

update readme to point to direct link to runpod template, cleanup install instrucitons (#532) 34c0a86 unverified

Add support for GPTQ using native transformers/peft (#468) 3355706 unverified

remove --force-reinstall from Dockerfile to ensure correct pytorch version (#492) e356b29 unverified

flash attn pip install (#426) cf66547 unverified

bump flash-attn to 2.0.4 for the base docker image (#382) ffac902 unverified

add peft install back since it doesn't get installed by setup.py (#331) db2a358 unverified

pin accelerate so it works with llama2 (#330) 6c9a87c unverified

Prune cuda117 (#327) 2c37bf6 unverified

add runpod envs to .bashrc, fix bnb env (#316) cf62cfd unverified

pin flash attention 2 to the fix for backwards pass cdf85fd

flash attention 2 9b790d3

explicitly pin flash attention 1 to v1.0.9 b06d3e3

misc fixes d75adb9

set transformers cache env var in docker image f162f3c

git fetch fix for docker eca3531

pin pydantic so deepspeed isn't broken 7145695

update pip install command for apex 530809f

shallow clone 5cd2126

clone in docker 12620f3

py310, fix cuda arg in deepspeed c43c5c8

Merge pull request #108 from OpenAccess-AI-Collective/docker-gptq bbc5bc5 unverified

Lint and format 392dfd9

cleanup from pr feedback 48612f8

default to qlora support, make gptq specific image 6ef96f5

move CUDA_VERSION_BNB arg inside of stage build scope e43bcc6

fix CUDA_VERSION_BNB env var 00323f0

bnb fixes 21f17cc

use python setup install, bdist wheel is unreliable in installing extension 809cceb

ensure libbitsandbytes*.so gets included with wheel a798ba1

fix missing run coninuation cf37980

update docker to compile latest bnb to properly support qlora 312b8d5

fix wheel install glob c3d2562

fix wheel blobs in dockerfile 1fc9b44

fix xentropy wheel name typo 259262b

fix relative path in flash-attn build: 8d6a289

fix cd within flash-attn f5fa3d1

optionally be able to specify alpaca or chat style prompts 1d5ab84

update entrypoint and force min accelerate fa8bd14

use printf instead of echo in dockerfile for portability 709be5a

fix path name to sorkspace 807cca8

setup runpod images 79deb35

don't compile deepspeed or bitsandbytes from source (#837)

f544ab2
unverified

add deepspeed-kernels dependency for deepspeed>=0.12.0 (#827)

8056ecd
unverified

fix pytorch 2.1.0 build, add multipack docs (#722)

2aa1f71
unverified

apex not needed as amp is part of pytorch (#696)

aca0398
unverified

fix multiline for docker (#694)

de87ea6
unverified

Feat: Set WORKDIR to /workspace/axolotl (#679)

133e676
unverified

tweak: improve base builder for smaller layers (#500)

923eb91
unverified

let MAX_JOBS use the default since we're not resource constrained on our self-hosted runners (#427)

e85d2eb
unverified

update dockerfile to not build evoformer since it fails the build (#607)

b53e777
unverified

update readme to point to direct link to runpod template, cleanup install instrucitons (#532)

34c0a86
unverified

Add support for GPTQ using native transformers/peft (#468)

3355706
unverified

remove --force-reinstall from Dockerfile to ensure correct pytorch version (#492)

e356b29
unverified

flash attn pip install (#426)

cf66547
unverified

bump flash-attn to 2.0.4 for the base docker image (#382)

ffac902
unverified

add peft install back since it doesn't get installed by setup.py (#331)

db2a358
unverified

pin accelerate so it works with llama2 (#330)

6c9a87c
unverified

Prune cuda117 (#327)

2c37bf6
unverified

add runpod envs to .bashrc, fix bnb env (#316)

cf62cfd
unverified

pin flash attention 2 to the fix for backwards pass

cdf85fd

flash attention 2

9b790d3

explicitly pin flash attention 1 to v1.0.9

b06d3e3

misc fixes

d75adb9

set transformers cache env var in docker image

f162f3c

git fetch fix for docker

eca3531

pin pydantic so deepspeed isn't broken

7145695

update pip install command for apex

530809f

shallow clone

5cd2126

clone in docker

12620f3

py310, fix cuda arg in deepspeed

c43c5c8

Merge pull request #108 from OpenAccess-AI-Collective/docker-gptq

bbc5bc5
unverified

Lint and format

392dfd9

cleanup from pr feedback

48612f8

default to qlora support, make gptq specific image

6ef96f5

move CUDA_VERSION_BNB arg inside of stage build scope

e43bcc6

fix CUDA_VERSION_BNB env var

00323f0

bnb fixes

21f17cc

use python setup install, bdist wheel is unreliable in installing extension

809cceb

ensure libbitsandbytes*.so gets included with wheel

a798ba1

fix missing run coninuation

cf37980

update docker to compile latest bnb to properly support qlora

312b8d5

fix wheel install glob

c3d2562

fix wheel blobs in dockerfile

1fc9b44

fix xentropy wheel name typo

259262b

fix relative path in flash-attn build:

8d6a289

fix cd within flash-attn

f5fa3d1

optionally be able to specify alpaca or chat style prompts

1d5ab84

update entrypoint and force min accelerate

fa8bd14

use printf instead of echo in dockerfile for portability

709be5a

fix path name to sorkspace

807cca8

setup runpod images

79deb35