revert config files + address comments

huggingface · Jan 15, 2024 · 7f9762f · 7f9762f
1 parent 57d1401
commit 7f9762f
Show file tree

Hide file tree

Showing 7 changed files with 9 additions and 22 deletions.
diff --git a/.github/workflows/slow-tests.yml b/.github/workflows/slow-tests.yml
@@ -37,13 +37,7 @@ jobs:
         if: always()
         run: |
           source activate trl
-          make slow_sft_tests
-
-      - name: Run slow DPO tests on single GPU
-        if: always()
-        run: |
-          source activate trl
-          make slow_dpo_tests
+          make slow_tests
       
       - name: Generate Report
         if: always()
@@ -79,13 +73,7 @@ jobs:
         if: always()
         run: |
           source activate trl
-          make slow_sft_tests
-
-      - name: Run slow DPO tests on Multi GPU
-        if: always()
-        run: |
-          source activate trl
-          make slow_dpo_tests
+          make slow_tests
 
       - name: Run end-to-end SFT examples tests on multi GPU
         if: always()

diff --git a/Makefile b/Makefile
@@ -20,11 +20,8 @@ benchmark_aux:
 tests_common_gpu:
 	python -m pytest tests/test_* $(if $(IS_GITHUB_CI),--report-log "common_tests.log",)
 
-slow_sft_tests:
-	python -m pytest tests/slow/test_sft_slow.py $(if $(IS_GITHUB_CI),--report-log "sft_slow.log",)
-
-slow_dpo_tests:
-	python -m pytest tests/slow/test_dpo_slow.py $(if $(IS_GITHUB_CI),--report-log "dpo_slow.log",)
+slow_tests:
+	python -m pytest tests/slow/test_* $(if $(IS_GITHUB_CI),--report-log "slow_tests.log",)
 
 run_sft_examples:
 	touch temp_results_sft_tests.txt

diff --git a/commands/run_dpo.sh b/commands/run_dpo.sh
@@ -33,6 +33,7 @@ fi
 CMD="""
 accelerate launch $EXTRA_ACCELERATE_ARGS \
     --num_processes $NUM_GPUS \
+    --mixed_precision 'fp16' \
     `pwd`/examples/scripts/dpo.py \
     --model_name_or_path $MODEL_NAME \
     --output_dir $OUTPUT_DIR \

diff --git a/commands/run_sft.sh b/commands/run_sft.sh
@@ -35,6 +35,7 @@ fi
 CMD="""
 accelerate launch $EXTRA_ACCELERATE_ARGS \
     --num_processes $NUM_GPUS \
+    --mixed_precision 'fp16' \
     `pwd`/examples/scripts/sft.py \
     --model_name $MODEL_NAME \
     --dataset_name $DATASET_NAME \

diff --git a/examples/accelerate_configs/deepspeed_zero1.yaml b/examples/accelerate_configs/deepspeed_zero1.yaml
@@ -9,7 +9,7 @@ distributed_type: DEEPSPEED
 downcast_bf16: 'no'
 machine_rank: 0
 main_training_function: main
-mixed_precision: 'fp16'
+mixed_precision: 'bf16'
 num_machines: 1
 num_processes: 8
 rdzv_backend: static

diff --git a/examples/accelerate_configs/deepspeed_zero2.yaml b/examples/accelerate_configs/deepspeed_zero2.yaml
@@ -11,7 +11,7 @@ distributed_type: DEEPSPEED
 downcast_bf16: 'no'
 machine_rank: 0
 main_training_function: main
-mixed_precision: 'fp16'
+mixed_precision: 'bf16'
 num_machines: 1
 num_processes: 8
 rdzv_backend: static

diff --git a/examples/accelerate_configs/deepspeed_zero3.yaml b/examples/accelerate_configs/deepspeed_zero3.yaml
@@ -12,7 +12,7 @@ distributed_type: DEEPSPEED
 downcast_bf16: 'no'
 machine_rank: 0
 main_training_function: main
-mixed_precision: 'fp16'
+mixed_precision: 'bf16'
 num_machines: 1
 num_processes: 8
 rdzv_backend: static