Skip to content

Commit

Permalink
change ci exp dir to /tmp (NVIDIA#10859)
Browse files Browse the repository at this point in the history 
Signed-off-by: Chen Cui <chcui@nvidia.com>
  • Loading branch information
@cuichenx
cuichenx authored Oct 12, 2024
1 parent 6de4e54 commit 44aa545
Showing 1 changed file with 24 additions and 24 deletions.
48 changes: 24 additions & 24 deletions .github/workflows/cicd-main.yml
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5205,7 +5205,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 1 \
Expand All @@ -5215,14 +5215,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 1 \
--mbs 1
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_SFT_TP1PP1_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5236,7 +5236,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 1 \
Expand All @@ -5246,14 +5246,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 1 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_SFT_TP1PP2_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5267,7 +5267,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 2 \
Expand All @@ -5277,14 +5277,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 1 \
--pp_size 2 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_SFT_TP2PP1_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5298,7 +5298,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 2 \
--pp_size 1 \
Expand All @@ -5308,14 +5308,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft none \
--tp_size 2 \
--pp_size 1 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_LoRA_TP1PP1_MBS1:
needs: [cicd-test-container-setup]
Expand All @@ -5329,7 +5329,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 1 \
Expand All @@ -5339,14 +5339,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 1 \
--mbs 1
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_LoRA_TP1PP1_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5360,7 +5360,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 1 \
Expand All @@ -5370,14 +5370,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 1 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_LoRA_TP1PP2_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5391,7 +5391,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 2 \
Expand All @@ -5401,14 +5401,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 1 \
--pp_size 2 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_GPT_LoRA_TP2PP1_MBS2:
needs: [cicd-test-container-setup]
Expand All @@ -5422,7 +5422,7 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 3 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 2 \
--pp_size 1 \
Expand All @@ -5432,14 +5432,14 @@ jobs:
--restore_path /home/TestData/nemo2_ckpt/llama_68M \
--devices 2 \
--max_steps 6 \
--experiment_dir tests/collections/llm/${{ github.run_id }} \
--experiment_dir /tmp/nemo2_gpt_finetune/${{ github.run_id }} \
--peft lora \
--tp_size 2 \
--pp_size 1 \
--mbs 2
AFTER_SCRIPT: |
rm -rf tests/collections/llm/${{ github.run_id }}
rm -rf /tmp/nemo2_gpt_finetune/${{ github.run_id }}
L2_NeMo_2_NeMo_Mcore_Mixtral_bitexact:
needs: [cicd-test-container-setup]
Expand Down

0 comments on commit 44aa545

Please sign in to comment.