backup contextual-turn-encoder-base-todbert-lite-ar-1m

contextual-turn-encoder-base-todbert-lite-ar-1m/best/config.json

@@ -0,0 +1,19 @@
+{
+  "input_dim": 768,
+  "hidden_dim": 768,
+  "output_dim": 768,
+  "num_layers": 6,
+  "num_heads": 8,
+  "dropout": 0.1,
+  "max_turns": 64,
+  "attention_mode": "autoregressive",
+  "use_speaker_embeddings": true,
+  "num_speakers": 4,
+  "layer_norm": true,
+  "ff_dim": 3072,
+  "activation": "gelu",
+  "output_residual": true,
+  "arch": "v2",
+  "head_mode": "tied_continuous",
+  "head_transform": true
+}

contextual-turn-encoder-base-todbert-lite-ar-1m/best/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eff1ecfa69eedec9c5947a45c208fa4322ea28c1173c65b7492980c6c576fdc9
+size 179808064

contextual-turn-encoder-base-todbert-lite-ar-1m/best/training_args.json

@@ -0,0 +1,15 @@
+{
+  "tag": "todbert",
+  "arch": "v2",
+  "recipe": "v1",
+  "base": "todbert",
+  "scale": "1m",
+  "n_layers": 6,
+  "n_heads": 8,
+  "lr": 0.0002,
+  "mode": "autoregressive",
+  "epoch": 8,
+  "train_loss": 3.20293,
+  "val_loss": 4.5075,
+  "epoch_sec": 549.9
+}

contextual-turn-encoder-base-todbert-lite-ar-1m/config.json

@@ -0,0 +1,19 @@
+{
+  "input_dim": 768,
+  "hidden_dim": 768,
+  "output_dim": 768,
+  "num_layers": 6,
+  "num_heads": 8,
+  "dropout": 0.1,
+  "max_turns": 64,
+  "attention_mode": "autoregressive",
+  "use_speaker_embeddings": true,
+  "num_speakers": 4,
+  "layer_norm": true,
+  "ff_dim": 3072,
+  "activation": "gelu",
+  "output_residual": true,
+  "arch": "v2",
+  "head_mode": "tied_continuous",
+  "head_transform": true
+}

contextual-turn-encoder-base-todbert-lite-ar-1m/config.yaml

@@ -0,0 +1,66 @@
+model:
+  input_dim: 768
+  hidden_dim: 768
+  output_dim: 768
+  num_layers: 6
+  num_heads: 8
+  dropout: 0.1
+  max_turns: 64
+  attention_mode: autoregressive
+  use_speaker_embeddings: true
+  num_speakers: 4
+  layer_norm: true
+  ff_dim: 3072
+  activation: gelu
+  output_residual: true
+  arch: v2
+  head_mode: tied_continuous
+  head_transform: true
+losses:
+  masked_reconstruction:
+    enabled: null
+    mask_prob: 0.15
+    weight: 1.0
+  next_turn_prediction:
+    enabled: null
+    weight: 1.0
+  embedding_retrieval:
+    enabled: true
+    weight: 1.0
+    temperature: 0.07
+    normalize: true
+    candidate_mode: in_batch
+    target: auto
+  lambda_cosine: 1.0
+training:
+  seed: 42
+  batch_size: 128
+  epochs: 8
+  learning_rate: 0.0002
+  weight_decay: 0.01
+  warmup_ratio: 0.05
+  gradient_clip_norm: 1.0
+  device: mps
+  mixed_precision: false
+  num_workers: 0
+  log_interval: 100
+  output_dir: models/contextual-turn-encoder-base
+data:
+  path: null
+  max_turns: 64
+  window: truncate
+  stride: 32
+  dialogue_id_col: dialogue_id
+  turn_id_col: turn_id
+  utterance_col: utterance
+  speaker_col: speaker
+  embedding_col: embedding
+  speaker_map: null
+base_encoder:
+  model_name: sergioburdisso/dialog2flow-joint-bert-base
+  backend: sentence_transformers
+  batch_size: 64
+  normalize: false
+  freeze: true
+  device: auto
+  cache_dir: null

contextual-turn-encoder-base-todbert-lite-ar-1m/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eff1ecfa69eedec9c5947a45c208fa4322ea28c1173c65b7492980c6c576fdc9
+size 179808064

contextual-turn-encoder-base-todbert-lite-ar-1m/training_args.json

@@ -0,0 +1,15 @@
+{
+  "tag": "todbert",
+  "arch": "v2",
+  "recipe": "v1",
+  "base": "todbert",
+  "scale": "1m",
+  "n_layers": 6,
+  "n_heads": 8,
+  "lr": 0.0002,
+  "mode": "autoregressive",
+  "epoch": 8,
+  "train_loss": 3.20293,
+  "val_loss": 4.5075,
+  "epoch_sec": 549.9
+}

contextual-turn-encoder-base-todbert-lite-ar-1m/trainlog.jsonl

@@ -0,0 +1,8 @@
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 1, "train_loss": 4.75493, "val_loss": 4.87236, "epoch_sec": 451.8}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 2, "train_loss": 4.08244, "val_loss": 4.70925, "epoch_sec": 463.5}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 3, "train_loss": 3.89786, "val_loss": 4.63303, "epoch_sec": 449.6}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 4, "train_loss": 3.7431, "val_loss": 4.58999, "epoch_sec": 458.1}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 5, "train_loss": 3.58774, "val_loss": 4.54786, "epoch_sec": 463.1}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 6, "train_loss": 3.43398, "val_loss": 4.53797, "epoch_sec": 454.2}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 7, "train_loss": 3.29978, "val_loss": 4.5232, "epoch_sec": 519.9}
+{"tag": "todbert", "arch": "v2", "recipe": "v1", "base": "todbert", "scale": "1m", "n_layers": 6, "n_heads": 8, "lr": 0.0002, "mode": "autoregressive", "epoch": 8, "train_loss": 3.20293, "val_loss": 4.5075, "epoch_sec": 549.9}