fix:pipeline报错用例，删除看护用例中的save参数-MindSpeed-LLM-AtomGit

已合并

fix:pipeline报错用例，删除看护用例中的save参数 #2380

wx_4e191bb7af创建于 2025年3月11日从refs/pull/2380/head 合入到 master

fix:pipeline报错用例，删除看护用例中的save参数 #2380

已合并

wx_4e191bb7af创建于 2025年3月11日

从refs/pull/2380/head合入到master

共 5 个文件变更+33-37

Mtests/pipeline/common/test_inference.py+33-33

     world_size = 2
     test_config = create_testconfig(Path(__file__).with_suffix(".json"))
--    @pytest.mark.parametrize("params", test_config["test_llama2_legacy_prompt_greedy_search"])
--    def test_llama2_legacy_greedy_search(self, build_args, params):
--        os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"
--        os.environ["CLOSE_MATMUL_K_SHIFT"] = "1"
--        if dist.get_rank() == 0:
--            handler, log_capture = setup_logger(PATTERN)
--
--        main()
--        if dist.get_rank() == 0:
--            print("=============== llama2 legacy prompt greedy search =============")
--            print(log_capture)
--            context = acquire_context(log_capture)
--            assert [context] == [
--                "I'm doing well, thanks.\nI'm doing well, thanks. I'm doing well, thanks. I'm doing"
--            ], "forward pass has been changed, check it!"
--
--    @pytest.mark.parametrize("params", test_config["test_llama2_lora_prompt_legacy_greedy_search"])
--    def test_llama2_lora_legacy_greedy_search(self, build_args, params):
--        os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"
--        os.environ["CLOSE_MATMUL_K_SHIFT"] = "1"
--        if dist.get_rank() == 0:
--            handler, log_capture = setup_logger(PATTERN)
--
--        main()
--        if dist.get_rank() == 0:
--            print("=============== llama2 lora legacy prompt greedy search =============")
--            print(log_capture)
--            context = acquire_context(log_capture)
--            assert [context] == [
--                "I'm doing well, thanks.\nI'm doing well, thanks. I'm doing well, thanks. I'm doing"
--            ], "forward pass has been changed, check it!"
--
     @pytest.mark.parametrize("params", test_config["test_chatglm3_mcore_greedy_search"])
     def test_chatglm3_mcore_greedy_search(self, build_args, params):
         os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"
 class TestInference(DistributedTest):
     world_size = 8
     test_config = create_testconfig(Path(__file__).with_suffix(".json"))
--
++
++    @pytest.mark.parametrize("params", test_config["test_llama2_legacy_prompt_greedy_search"])
++    def test_llama2_legacy_greedy_search(self, build_args, params):
++        os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"
++        os.environ["CLOSE_MATMUL_K_SHIFT"] = "1"
++        if dist.get_rank() == 0:
++            handler, log_capture = setup_logger(PATTERN)
++
++        main()
++        if dist.get_rank() == 0:
++            print("=============== llama2 legacy prompt greedy search =============")
++            print(log_capture)
++            context = acquire_context(log_capture)
++            assert [context] == [
++                "I'm doing well, thanks.\nI'm doing well, thanks. I'm doing well, thanks. I'm doing"
++            ], "forward pass has been changed, check it!"
++
++    @pytest.mark.parametrize("params", test_config["test_llama2_lora_prompt_legacy_greedy_search"])
++    def test_llama2_lora_legacy_greedy_search(self, build_args, params):
++        os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"
++        os.environ["CLOSE_MATMUL_K_SHIFT"] = "1"
++        if dist.get_rank() == 0:
++            handler, log_capture = setup_logger(PATTERN)
++
++        main()
++        if dist.get_rank() == 0:
++            print("=============== llama2 lora legacy prompt greedy search =============")
++            print(log_capture)
++            context = acquire_context(log_capture)
++            assert [context] == [
++                "I'm doing well, thanks.\nI'm doing well, thanks. I'm doing well, thanks. I'm doing"
++            ], "forward pass has been changed, check it!"
++
     @pytest.mark.parametrize("params", test_config["test_baichuan2_mcore_greedy_search"])
     def test_baichuan2_mcore_greedy_search(self, build_args, params):
         os.environ["CUDA_DEVICE_MAX_CONNECTIONS"] = "1"

Mtests/pipeline/gemma-7B/gemma2_tp8_pp1_ptd.sh+0-1

	@@ -106,4 +106,3 @@ torchrun $DISTRIBUTED_ARGS $basepath/pretrain_gpt.py \
106	$OUTPUT_ARGS \	106	$OUTPUT_ARGS \
107	--distributed-backend nccl \	107	--distributed-backend nccl \
108	--load ${CKPT_LOAD_DIR} \	108	--load ${CKPT_LOAD_DIR} \
109	- --save ${CKPT_SAVE_DIR}

Mtests/pipeline/gpt4/gpt4_mcore_tp4_cp2_32k_moe_drop.sh+0-1

	@@ -106,7 +106,6 @@ CKPT_ARGS="
106	--no-save-optim \	106	--no-save-optim \
107	--no-save-rng \	107	--no-save-rng \
108	--seed 1234 \	108	--seed 1234 \
109	- --save ${CKPT_SAVE_DIR}
110	"	109	"
111		110
112	OUTPUT_ARGS="	111	OUTPUT_ARGS="

Mtests/pipeline/llama2/llama2_tp8_pp1_coc_ptd.sh+0-1

	@@ -98,4 +98,3 @@ torchrun $DISTRIBUTED_ARGS $basepath/pretrain_gpt.py \
98	$OUTPUT_ARGS \	98	$OUTPUT_ARGS \
99	--distributed-backend nccl \	99	--distributed-backend nccl \
100	--load ${CKPT_LOAD_DIR} \	100	--load ${CKPT_LOAD_DIR} \
101	- --save ${CKPT_SAVE_DIR} \

Mtests/st/shell_scripts/llama2_tp4pp2vpp2_tp2d_tpx2tpy2.sh+0-1

	@@ -13,7 +13,6 @@ WORLD_SIZE=$(($NPUS_PER_NODE*$NNODES))
13	basepath=$(cd `dirname $0`; cd ../../../; pwd)	13	basepath=$(cd `dirname $0`; cd ../../../; pwd)
14		14
15		15
16	-CKPT_SAVE_DIR="./ckpt"
17	DATA_PATH="/data/pretrain_dataset/alpaca_text_document"	16	DATA_PATH="/data/pretrain_dataset/alpaca_text_document"
18	TOKENIZER_MODEL="/data/llama-2-7b-hf/tokenizer.model"	17	TOKENIZER_MODEL="/data/llama-2-7b-hf/tokenizer.model"
19	CKPT_LOAD_DIR="/data/llama2-2dtp-tp4pp2vp2tpx2tpy2"	18	CKPT_LOAD_DIR="/data/llama2-2dtp-tp4pp2vp2tpx2tpy2"