backup contextual-turn-encoder-base-custom-bidi-full

contextual-turn-encoder-base-custom-bidi-full/best/config.json

@@ -0,0 +1,16 @@
+{
+  "input_dim": 768,
+  "hidden_dim": 768,
+  "output_dim": 768,
+  "num_layers": 6,
+  "num_heads": 8,
+  "dropout": 0.1,
+  "max_turns": 64,
+  "attention_mode": "bidirectional",
+  "use_speaker_embeddings": true,
+  "num_speakers": 4,
+  "layer_norm": true,
+  "ff_dim": 3072,
+  "activation": "gelu",
+  "output_residual": true
+}

contextual-turn-encoder-base-custom-bidi-full/best/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fc5246143435b908a26228514e05ab51089f89ea95151d72c5d730b65c51c557
+size 175072720

contextual-turn-encoder-base-custom-bidi-full/best/training_args.json

@@ -0,0 +1,7 @@
+{
+  "mode": "bidirectional",
+  "epoch": 4,
+  "train_loss": 1.43924,
+  "val_loss": 2.9518,
+  "epoch_sec": 1904.1
+}

contextual-turn-encoder-base-custom-bidi-full/config.json

@@ -0,0 +1,16 @@
+{
+  "input_dim": 768,
+  "hidden_dim": 768,
+  "output_dim": 768,
+  "num_layers": 6,
+  "num_heads": 8,
+  "dropout": 0.1,
+  "max_turns": 64,
+  "attention_mode": "bidirectional",
+  "use_speaker_embeddings": true,
+  "num_speakers": 4,
+  "layer_norm": true,
+  "ff_dim": 3072,
+  "activation": "gelu",
+  "output_residual": true
+}

contextual-turn-encoder-base-custom-bidi-full/config.yaml

@@ -0,0 +1,63 @@
+model:
+  input_dim: 768
+  hidden_dim: 768
+  output_dim: 768
+  num_layers: 6
+  num_heads: 8
+  dropout: 0.1
+  max_turns: 64
+  attention_mode: bidirectional
+  use_speaker_embeddings: true
+  num_speakers: 4
+  layer_norm: true
+  ff_dim: 3072
+  activation: gelu
+  output_residual: 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: 10
+  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-custom-bidi-full/model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab7141b56c695aab34b682eba8b2ec64bd46e894ef168e90e8344b475f0e31e2
+size 175072720

contextual-turn-encoder-base-custom-bidi-full/training_args.json

@@ -0,0 +1,7 @@
+{
+  "mode": "bidirectional",
+  "epoch": 10,
+  "train_loss": 0.48214,
+  "val_loss": 3.03056,
+  "epoch_sec": 986.7
+}

contextual-turn-encoder-base-custom-bidi-full/trainlog.jsonl

@@ -0,0 +1,10 @@
+{"mode": "bidirectional", "epoch": 1, "train_loss": 3.41986, "val_loss": 3.57963, "epoch_sec": 1278.6}
+{"mode": "bidirectional", "epoch": 2, "train_loss": 2.16521, "val_loss": 3.17761, "epoch_sec": 1099.3}
+{"mode": "bidirectional", "epoch": 3, "train_loss": 1.75156, "val_loss": 3.00788, "epoch_sec": 1105.8}
+{"mode": "bidirectional", "epoch": 4, "train_loss": 1.43924, "val_loss": 2.9518, "epoch_sec": 1904.1}
+{"mode": "bidirectional", "epoch": 5, "train_loss": 1.16854, "val_loss": 2.9893, "epoch_sec": 1159.4}
+{"mode": "bidirectional", "epoch": 6, "train_loss": 0.94373, "val_loss": 3.02662, "epoch_sec": 1110.8}
+{"mode": "bidirectional", "epoch": 7, "train_loss": 0.76426, "val_loss": 3.02846, "epoch_sec": 1166.8}
+{"mode": "bidirectional", "epoch": 8, "train_loss": 0.63033, "val_loss": 3.0518, "epoch_sec": 1117.7}
+{"mode": "bidirectional", "epoch": 9, "train_loss": 0.53636, "val_loss": 3.03342, "epoch_sec": 1062.4}
+{"mode": "bidirectional", "epoch": 10, "train_loss": 0.48214, "val_loss": 3.03056, "epoch_sec": 986.7}