Upload 3 files

hgpt/motionx/VQVAE_MotionX_2Kx1K_2024-12-20-16-56-10/checkpoints/last.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c705e6cb86806658c20fefdd4630c4c8b47dba9ec062ef186b8aa5b9cda39fa
+size 964443382

hgpt/motionx/VQVAE_MotionX_2Kx1K_2024-12-20-16-56-10/config_2024-12-19-22-23-43_train.yaml

@@ -0,0 +1,225 @@
+SEED_VALUE: 1234
+DEBUG: false
+NAME: VQVAE_MotionX_2Kx1K
+ACCELERATOR: gpu
+NUM_NODES: 1
+DEVICE:
+- 0
+- 1
+- 2
+- 3
+- 4
+- 5
+- 6
+- 7
+DATASET:
+  target: hGPT.data.MotionX.MotionXDataModule
+  NFEATS: 623
+  JOINT_TYPE: motionx
+  MOTIONX:
+    FRAME_RATE: 30.0
+    MAX_MOTION_LEN: 400
+    MIN_MOTION_LEN: 40
+    MAX_TEXT_LEN: 20
+    PICK_ONE_TEXT: true
+    UNIT_LEN: 4
+    STD_TEXT: false
+    ROOT: datasets/motionx/data
+    SPLIT_ROOT: datasets/motionx/data/split
+    SEMANTIC_TEXT_ROOT: datasets/motionx/data/texts/semantic_texts
+    FACE_TEXT_ROOT: datasets/motionx/data/texts/face_texts
+  WORD_VERTILIZER_PATH: deps/glove/
+  HUMANML3D:
+    ROOT: datasets/humanml3d
+    SPLIT_ROOT: datasets/humanml3d
+model:
+  target: hGPT.models.hgpt.HumanoidGPT
+  params:
+    lm: ''
+    motion_vae: ${vq.vqvae_2k_1k}
+    task: vae
+    condition: ''
+    stage: ${TRAIN.STAGE}
+    codebook_size: ${model.params.motion_vae.params.code_num}
+    metrics_dict: ${METRIC.TYPE}
+    debug: ${DEBUG}
+TRAIN:
+  STAGE: vae
+  SPLIT: train
+  PRECISION: '32'
+  NUM_JOINTS: 52
+  NUM_WORKERS: 16
+  BATCH_SIZE: 2048
+  END_EPOCH: 3000
+  RESUME: ''
+  PRETRAINED: ''
+  PRETRAINED_VAE: ''
+  OPTIM:
+    target: AdamW
+    params:
+      lr: 0.0002
+      betas:
+      - 0.9
+      - 0.99
+      weight_decay: 0.0
+  LR_SCHEDULER:
+    target: CosineAnnealingLR
+    params:
+      T_max: ${eval:${LOGGER.VAL_EVERY_STEPS} * 100}
+      eta_min: 1.0e-06
+EVAL:
+  SPLIT: val
+  BATCH_SIZE: 32
+  NUM_WORKERS: 8
+TEST:
+  DATASETS:
+  - MotionX
+  CHECKPOINTS: experiments/hgpt/VQVAE_MotionX_Debug/checkpoints/last.ckpt
+  SPLIT: test
+  BATCH_SIZE: 512
+  NUM_WORKERS: 8
+  SAVE_PREDICTIONS: true
+  COUNT_TIME: true
+  REPLICATION_TIMES: 1
+  REP_I: 0
+  FOLDER: results
+PREDICT:
+  SPLIT: test
+  CHECKPOINTS: experiments/hgpt/VQVAE_MotionX_Debug/checkpoints/last.ckpt
+  BATCH_SIZE: 512
+LOSS:
+  LAMBDA_FEATURE: 1.0
+  LAMBDA_VELOCITY: 0.5
+  LAMBDA_COMMIT: 0.02
+  LAMBDA_CLS: 1.0
+  ABLATION:
+    RECONS_LOSS: l1_smooth
+METRIC:
+  TYPE:
+  - MRMetrics
+  FORCE_IN_METER: true
+  DIST_SYNC_ON_STEP: true
+  MM_NUM_SAMPLES: 100
+  MM_NUM_REPEATS: 30
+  MM_NUM_TIMES: 10
+LOGGER:
+  VAL_EVERY_STEPS: 10
+  TYPE:
+  - tensorboard
+  - wandb
+  TENSORBOARD:
+    target: pytorch_lightning.loggers.TensorBoardLogger
+    params:
+      save_dir: ${FOLDER_EXP}
+      name: tensorboard
+      version: ''
+  WANDB:
+    target: pytorch_lightning.loggers.WandbLogger
+    params:
+      project: humanoidgpt
+      offline: true
+      id: null
+      version: ''
+      name: ${NAME}
+      save_dir: ${FOLDER_EXP}
+eval:
+  tm2t:
+    t2m_path: deps/t2m/
+    t2m_textencoder:
+      target: hGPT.archs.tm2t_evaluator.TextEncoderBiGRUCo
+      params:
+        word_size: 300
+        pos_size: 15
+        hidden_size: 512
+        output_size: 512
+    t2m_moveencoder:
+      target: hGPT.archs.tm2t_evaluator.MovementConvEncoder
+      params:
+        input_size: 259
+        hidden_size: 512
+        output_size: 512
+    t2m_motionencoder:
+      target: hGPT.archs.tm2t_evaluator.MotionEncoderBiGRUCo
+      params:
+        input_size: ${eval.tm2t.t2m_moveencoder.params.output_size}
+        hidden_size: 1024
+        output_size: 512
+vq:
+  default:
+    target: hGPT.archs.hgpt_vq.VQVae
+    params:
+      quantizer: ema_reset
+      code_num: 512
+      code_dim: 512
+      output_emb_width: 512
+      down_t: 2
+      stride_t: 2
+      width: 512
+      depth: 3
+      dilation_growth_rate: 3
+      norm: None
+      activation: relu
+      nfeats: ${DATASET.NFEATS}
+  vqvae_2k_1k:
+    target: hGPT.archs.hgpt_vq.VQVae
+    params:
+      quantizer: ema_reset
+      n_quantizer: 1
+      code_num: 2048
+      code_dim: 1024
+      output_emb_width: 1024
+      width: 1024
+      down_t: 2
+      stride_t: 2
+      depth: 3
+      dilation_growth_rate: 3
+      norm: None
+      activation: relu
+      nfeats: ${DATASET.NFEATS}
+  rvqvae_2k_1k_2layer:
+    target: hGPT.archs.hgpt_vq.RVQVae
+    params:
+      quantizer: ema_reset
+      n_quantizer: 2
+      code_num: 2048
+      code_dim: 1024
+      output_emb_width: 1024
+      width: 1024
+      down_t: 2
+      stride_t: 2
+      depth: 3
+      dilation_growth_rate: 3
+      norm: None
+      activation: relu
+      nfeats: ${DATASET.NFEATS}
+lm:
+  default:
+    target: hGPT.archs.hgpt_lm.MLM
+    params:
+      model_type: t5
+      model_path: ./deps/flan-t5-base
+      stage: ${TRAIN.STAGE}
+      motion_codebook_size: ${model.params.codebook_size}
+  llama:
+    target: hGPT.archs.hgpt_lm.MLM
+    params:
+      model_type: llama
+      model_path: ./deps/Meta-Llama-3.1-8B-Instruct
+      stage: ${TRAIN.STAGE}
+      motion_codebook_size: ${model.params.codebook_size}
+      max_length: 1500
+      lora: true
+      train_strategy: pt
+      new_token_type: insert
+      framerate: 30.0
+      down_t: 4
+      predict_ratio: 0.2
+      inbetween_ratio: 0.25
+      quota_ratio: 0.5
+      noise_density: 0.15
+      mean_noise_span_length: 3
+CONFIG_FOLDER: configs
+ARCHS_FOLDER: configs/archs
+FOLDER: experiments
+FOLDER_EXP: experiments/hgpt/VQVAE_MotionX_2Kx1K
+TIME: 2024-12-19-22-23-43