Code changes for supporting llama3_1-405b reference implementation #111
Conversation
|
MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅ |
| - llama3-402b | ||
| skip_if_env: | ||
| CM_ML_MODEL_LLAMA3_CHECKPOINT_PATH: | ||
| - 'on' |
There was a problem hiding this comment.
Also only if we are in the docker build stage. Otherwise when the path is given we should register it in CM cache
There was a problem hiding this comment.
Updated in commit bfb7cb6
| - llama3-402b | ||
| skip_if_env: | ||
| CM_DATASET_LLAMA3_PATH: | ||
| - "on" |
There was a problem hiding this comment.
Same here as for the model.
There was a problem hiding this comment.
Updated in commit bfb7cb6
script/get-ml-model-llama3/_cm.yaml
Outdated
| env: | ||
| CM_MODEL_ZOO_ENV_KEY: LLAMA3 | ||
| group: huggingface-stub | ||
| docker: |
There was a problem hiding this comment.
Is this docker section needed?
There was a problem hiding this comment.
Updated in commit a21f2b0
| skip_if_env: | ||
| CM_DATASET_LLAMA3_PATH: | ||
| - "on" | ||
| CM_USE_DATASET_FROM_HOST: |
There was a problem hiding this comment.
This variable is not needed right? Because if the path is directly passed from the host to a container then, this won't work. Same for model.
There was a problem hiding this comment.
That's right, thanks. Also I formatted the app-mlperf-inference-mlcommons-python _cm.yaml file with the help of prettier extension in VS Code.
TODO:
rclonedirectory path toget-dataset-mlperf-inference-llama3.