Add qwen3 metax functional regression case

Author: AlexMa616

.github/configs/cuda.yml

@@ -92,9 +92,9 @@ test_matrix:
   functional:
     train:
       - model: gpt
-        test_case: gpt3_mcore_te_tp1_pp1_mock_data_dist_optimizer
+        test_case: qwen3_0p6b_mcore_te_tp1_pp1_no_mmap_bin_files
         training_script: pretrain_gpt.py
         n_repeat: 1
         golden_environment: dev
         golden_platform: dgx_a100
-        enable_lightweight_mode: true
+        enable_lightweight_mode: false

tests/functional_tests/test_cases/gpt/qwen3_0p6b_mcore_te_tp1_pp1_no_mmap_bin_files/golden_values_dev_dgx_a100.json

@@ -0,0 +1 @@
+{"lm loss": {"start_step": 1, "end_step": 10, "step_interval": 1, "values": {"1": 11.93016, "2": 11.87934, "3": 11.78204, "4": 11.84415, "5": 11.64288, "6": 11.57093, "7": 11.56676, "8": 11.75107, "9": 11.77425, "10": 11.44834}}}

tests/functional_tests/test_cases/gpt/qwen3_0p6b_mcore_te_tp1_pp1_no_mmap_bin_files/model_config.yaml

@@ -0,0 +1,74 @@
+ENV_VARS:
+  CUDA_DEVICE_MAX_CONNECTIONS: 1
+  NVTE_ALLOW_NONDETERMINISTIC_ALGO: 0
+  NCCL_ALGO: Ring
+  CUBLAS_WORKSPACE_CONFIG: :4096:8
+MODEL_ARGS:
+  --num-layers: 28
+  --hidden-size: 1024
+  --ffn-hidden-size: 3072
+  --kv-channels: 128
+  --num-attention-heads: 16
+  --group-query-attention: true
+  --num-query-groups: 8
+  --log-params-norm: true
+  --tensorboard-dir: ${TENSORBOARD_PATH}
+  --micro-batch-size: 1
+  --global-batch-size: 2
+  --seq-length: 256
+  --max-position-embeddings: 40960
+  --train-iters: 10
+  --timing-log-level: 0
+  --save: ${CHECKPOINT_SAVE_PATH}
+  --load: ${CHECKPOINT_LOAD_PATH}
+  --data-path: /home/gitlab-runner/data/pile_wikipedia_demo/pile_wikipedia_demo
+  --split: 100,0,0
+  --tokenizer-type: HuggingFaceTokenizer
+  --tokenizer-model: /home/gitlab-runner/tokenizers/qwentokenizer
+  --legacy-tokenizer: true
+  --trust-remote-code: true
+  --distributed-backend: nccl
+  --lr: 1.0e-5
+  --lr-decay-style: cosine
+  --min-lr: 1.0e-6
+  --weight-decay: 0.1
+  --clip-grad: 1.0
+  --lr-warmup-fraction: .01
+  --log-interval: 1
+  --save-interval: 100000
+  --eval-interval: 100000
+  --eval-iters: 0
+  --transformer-impl: transformer_engine
+  --tensor-model-parallel-size: 1
+  --pipeline-model-parallel-size: 1
+  --no-mmap-bin-files: true
+  --deterministic-mode: true
+  --no-check-for-nan-in-loss-and-grad: true
+  --no-gradient-accumulation-fusion: true
+  --attention-softmax-in-fp32: true
+  --use-mcore-models: true
+  --ckpt-format: torch
+  --bf16: true
+  --attention-backend: unfused
+  --disable-bias-linear: true
+  --reset-position-ids: true
+  --reset-attention-mask: true
+  --qk-layernorm: true
+  --normalization: RMSNorm
+  --norm-epsilon: 1e-6
+  --position-embedding-type: rope
+  --no-position-embedding: true
+  --rotary-base: 1000000
+  --rotary-percent: 1.0
+  --no-rope-fusion: true
+  --swiglu: true
+  --attention-dropout: 0.0
+  --hidden-dropout: 0.0
+  --init-method-std: 0.006
+  --seed: 42
+  --adam-beta1: 0.9
+  --adam-beta2: 0.95
+  --make-vocab-size-divisible-by: 64
+TEST_TYPE: regular
+METRICS:
+  - "lm loss"