add new models in autotp workflow and enable gptj, baichuan

Author: delock

.github/workflows/autotp.yml

@@ -126,3 +126,93 @@ jobs:
           cd DeepSpeedExamples/inference/huggingface/text-generation
           deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
           deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+
+      - name: AutoTP test (EleutherAI/gpt-j-6b)
+        run: |
+          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+          source oneCCL/build/_install/env/setvars.sh
+          # modify MODEL to change the model name, other lines are the same
+          export MODEL=EleutherAI/gpt-j-6b
+          cd DeepSpeedExamples/inference/huggingface/text-generation
+          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+
+      - name: AutoTP test (baichuan-inc/Baichuan-7B)
+        run: |
+          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+          source oneCCL/build/_install/env/setvars.sh
+          # modify MODEL to change the model name, other lines are the same
+          export MODEL=baichuan-inc/Baichuan-7B
+          cd DeepSpeedExamples/inference/huggingface/text-generation
+          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+
+#      - name: AutoTP test (bigcode/starcoder)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=bigcode/starcoder
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (tiiuae/falcon-7b)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=tiiuae/falcon-7b
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (google/flan-t5-xl)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=google/flan-t5-xl
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (mistralai/Mistral-7B-v0.1)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=mistralai/Mistral-7B-v0.1
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (mosaicml/mpt-7b)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=mosaicml/mpt-7b
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (meta-llama/Llama-2-7b-hf)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=meta-llama/Llama-2-7b-hf
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1
+#
+#      - name: AutoTP test (EleutherAI/gpt-neox-20b)
+#        run: |
+#          export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libstdc++.so.6
+#          source oneCCL/build/_install/env/setvars.sh
+#          # modify MODEL to change the model name, other lines are the same
+#          export MODEL=EleutherAI/gpt-neox-20
+#          cd DeepSpeedExamples/inference/huggingface/text-generation
+#          deepspeed --num_gpus 2 --bind_cores_to_rank inference-test.py --model $MODEL --dtype bfloat16 --use_meta_tensor
+#          deepspeed --num_gpus 2 --bind_cores_to_rank ds-hf-compare.py --model $MODEL --dtype bfloat16 --num_inputs 1