fix: Change base.yml path from MaxText to maxtext#1209
fix: Change base.yml path from MaxText to maxtext#1209RUEI4341 wants to merge 4 commits intoGoogleCloudPlatform:masterfrom
base.yml path from MaxText to maxtext#1209Conversation
In AI-Hypercomputer/maxtext#3044, all config files were moved from `src/MaxText/configs` to the lowercase `src/maxtext/configs`. Because of this directory change, the XLML DAGs are currently failing with the following error: ```FileNotFoundError: [Errno 2] No such file or directory: '/deps/src/MaxText/configs/base.yml'``` This change updated file paths for compatibility with AI-Hypercomputer/maxtext#3044.
RUEI4341
requested review from
JoeZijunZhou,
RissyRan,
Rohan-Bierneni,
aireenmei,
alfredyu-cienet,
andrewyct,
coolkp,
entrpn,
gobbleturk,
hyeygit,
jiangjy1982,
mailvijayasingh,
michelle-yooh,
notabee,
parambole,
raymondzouu,
richardsliu,
severus-ho,
shralex,
shuningjin,
sixiang-google,
tonyjohnchen,
vipannalla,
xibinliu and
yixinshi
as code owners
| base_command = ( | ||
| f"export BASE_OUTPUT_PATH={BASE_OUTPUT_PATH} && " | ||
| + "python3 -m MaxText.train MaxText/configs/base.yml base_output_directory=gs://runner-maxtext-logs run_name=${RUN_NAME} model_name=mixtral-8x7b tokenizer_path=assets/tokenizer.mistral-v1 dataset_path=gs://maxtext-dataset per_device_batch_size=4 enable_checkpointing=false ici_fsdp_parallelism=-1 max_target_length=1024 async_checkpointing=false attention=flash dtype=bfloat16 weight_dtype=bfloat16" | ||
| + "python3 -m MaxText.train maxtext/configs/base.yml base_output_directory=gs://runner-maxtext-logs run_name=${RUN_NAME} model_name=mixtral-8x7b tokenizer_path=assets/tokenizer.mistral-v1 dataset_path=gs://maxtext-dataset per_device_batch_size=4 enable_checkpointing=false ici_fsdp_parallelism=-1 max_target_length=1024 async_checkpointing=false attention=flash dtype=bfloat16 weight_dtype=bfloat16" |
There was a problem hiding this comment.
A separate PR would be good, but can you please also update the python3 -m MaxText.train references to use python3 -m maxtext.trainers.pre_train.train instead? The old ones are deprecated and will be removed in the near future. This is the full list of commands we will need to update: https://github.com/AI-Hypercomputer/maxtext/tree/102af23138003f20df9e9c8194ee6617e47881f9/src/MaxText
There was a problem hiding this comment.
Understood. We will open a separate PR to migrate those deprecated commands to the new training path.
| run_cmds = [ | ||
| "pip show aqtp", | ||
| f"bash MaxText/configs/{tpu}/{model_size}.sh EXECUTABLE=train.py OUTPUT_PATH={base_output_directory} PLATFORM=gke", | ||
| f"bash maxtext/configs/tpu/{tpu}/{model_size}.sh EXECUTABLE=train.py OUTPUT_PATH={base_output_directory} PLATFORM=gke", |
There was a problem hiding this comment.
Please use src/maxtext/configs as the path instead of maxtext/configs. This will work temporarily, but we will likely remove this soon
There was a problem hiding this comment.
Thank you for the clarification. We will update the path to src/maxtext/configs for now.
alfredyu-cienet
changed the title
base.yml path from MaxText to maxtext (#223)base.yml path from MaxText to maxtext
|
Hi @bvandermoon, Could you review these changes when you have a moment? Thank you. Also, to avoid potential code conflicts, we are planning to wait until this is merged before submitting the next PR, which will migrate |
3 participants
Description
In AI-Hypercomputer/maxtext#3044, all config files were moved from
src/MaxText/configsto the lowercasesrc/maxtext/configs. Because of this directory change, the XLML DAGs are currently failing with the following error:FileNotFoundError: [Errno 2] No such file or directory: '/deps/src/MaxText/configs/base.yml'This change updated file paths for compatibility with AI-Hypercomputer/maxtext#3044.
Checklist
Before submitting this PR, please make sure (put X in square brackets):