Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

chat_template.jinja

@@ -0,0 +1,4 @@
+{% if messages[0]['role'] == 'system' %}{{ raise_exception('System role not supported') }}{% endif %}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if (message['role'] == 'assistant') %}{% set role = 'model' %}{% else %}{% set role = message['role'] %}{% endif %}{{ '<start_of_turn>' + role + '
+' + message['content'] | trim + '<end_of_turn>
+' }}{% endfor %}{% if add_generation_prompt %}{{'<start_of_turn>model
+'}}{% endif %}

initial_latent_z.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82777bd3d172cbb47fd317bf8029e7974ce866fa1414cbd87e2c92b3bd4223ba
+size 1048656

metrics.json

@@ -0,0 +1,57 @@
+{
+  "phase_results": [
+    {
+      "phase": "multiturn",
+      "best_validation_loss": 2.652476119995117,
+      "best_metrics": {
+        "loss_total": 2.652476119995117,
+        "loss_decoder": 2.1608673334121704,
+        "loss_oracle_latent_mse": 0.2362155869603157,
+        "loss_oracle_latent_cosine": 0.005752995098009706,
+        "loss_oracle_first_token_kl": 0.254817932844162,
+        "first_token_correct": 96.0,
+        "first_token_count": 625.0,
+        "first_token_exact_match": 0.1536,
+        "state_teacher_forcing_ratio": 0.0
+      },
+      "oracle_preflight": {
+        "oracle_decoder_loss": 2.409222452066309,
+        "oracle_decoder_token_count": 24364.0,
+        "oracle_first_token_correct": 111.0,
+        "oracle_first_token_count": 625.0,
+        "oracle_first_token_exact_match": 0.1776,
+        "oracle_example_count": 151.0
+      },
+      "global_step": 363,
+      "train_dataset": {
+        "example_count": 7696,
+        "observation_steps_mean": 4.2548076923076925,
+        "observation_steps_max": 6,
+        "input_tokens_mean": 240.67255717255716,
+        "input_tokens_max": 1719,
+        "oracle_input_tokens_mean": 562.7067307692307,
+        "oracle_input_tokens_max": 3358,
+        "target_tokens_mean": 161.35044178794178,
+        "target_tokens_max": 1736,
+        "dataset_counts": {
+          "chatalpaca_multiturn_enriched": 7696
+        }
+      },
+      "validation_dataset": {
+        "example_count": 151,
+        "observation_steps_mean": 4.139072847682119,
+        "observation_steps_max": 6,
+        "input_tokens_mean": 236.91390728476821,
+        "input_tokens_max": 753,
+        "oracle_input_tokens_mean": 537.4238410596026,
+        "oracle_input_tokens_max": 1666,
+        "target_tokens_mean": 161.35099337748343,
+        "target_tokens_max": 912,
+        "dataset_counts": {
+          "chatalpaca_multiturn_enriched": 151
+        }
+      },
+      "wandb_enabled": true
+    }
+  ]
+}

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fa54c879ef2865c97be5750e0bdfe4d359aed282da52ec4fc1fa55a281d91cb6
+size 2518360980

sft_b200.yaml

@@ -0,0 +1,94 @@
+datasets:
+  - name: chatalpaca_multiturn_enriched
+    repo_id: BRlkl/chatalpaca-multiturn-enriched
+    source_split: train
+    format: messages_all_turns
+    validation_ratio: 0.02
+    split_seed: 17
+    min_turns: 2
+    max_turns: 6
+    max_message_chars: 6000
+    use_base_chat_template: true
+
+model:
+  base_model_name: google/t5gemma-l-l-prefixlm-it
+  initial_model_path: BRlkl/test
+  teacher_model_path: BRlkl/test
+  dtype: bfloat16
+  attn_implementation: sdpa
+  magicnorm_eps: 1.0e-6
+  z_slots: 256
+  num_time_tokens: 0
+  use_explicit_time_features: false
+  gate_attention_heads: 4
+  max_observation_tokens: 1024
+  max_decoder_tokens: 1024
+  thought_loop_proposal_mode: observation_hidden_compression
+  preserve_observation_encoder_manifold: true
+  observation_encoder_use_state_context: true
+  latent_attention_mask_mode: full
+  initial_update_gate_bias: 20.0
+
+training:
+  seed: 17
+  num_workers: 2
+  gradient_checkpointing: true
+  mixed_precision: bf16
+  max_grad_norm: 1.0
+  weight_decay: 0.01
+  backbone_learning_rate: 5.0e-6
+  new_module_learning_rate: 1.0e-4
+  adam_beta1: 0.9
+  adam_beta2: 0.95
+  adam_epsilon: 1.0e-8
+  fused_adamw: true
+  freeze_gate_head: true
+  log_every_steps: 1
+  eval_every_steps: 100
+  checkpoint_every_steps: 500
+  max_train_examples:
+  max_validation_examples:
+  freeze_decoder_first_fraction: 0.0
+  oracle_preflight_max_decoder_loss: 3.0
+  decoder_loss_weight: 1.0
+  oracle_latent_mse_weight: 1.0
+  oracle_latent_cosine_weight: 0.1
+  oracle_first_token_kl_weight: 1.0
+  oracle_first_token_kl_temperature: 1.0
+  state_teacher_forcing_start_ratio: 1.0
+  state_teacher_forcing_end_ratio: 0.0
+  state_teacher_forcing_end_fraction: 0.7
+  state_teacher_forcing_seed: 17
+  late_turn_weight_scale: 1.0
+  late_turn_weight_power: 1.0
+  student_observation_seed: 17
+  student_prev_assistant_dropout_prob: 0.0
+  student_prev_assistant_keep_last_chars:
+
+phase:
+  micro_batch_size: 16
+  eval_batch_size: 16
+  gradient_accumulation_steps: 4
+  num_train_epochs: 3
+  warmup_ratio: 0.03
+  shuffle_train: true
+
+cache:
+  preprocessed_root: cache/preprocessed_pre_sft_multiturn_chatalpaca_self_feedback
+
+paths:
+  run_root: runs_pre_sft_multiturn
+  export_root: exports_multiturn
+
+inference:
+  format: alpaca_single_turn
+  use_base_chat_template: true
+
+wandb:
+  enabled: true
+  project: samantha-pre-sft
+  run_name: t5gemma2-thoughtloop-pre-sft-chatalpaca-self-feedback22
+
+hub:
+  model_repo_id: BRlkl/test_multiturn_chatalpaca_self_feedback3333
+  private: false

sft_config.json

@@ -0,0 +1,99 @@
+{
+  "datasets": [
+    {
+      "name": "chatalpaca_multiturn_enriched",
+      "repo_id": "BRlkl/chatalpaca-multiturn-enriched",
+      "source_split": "train",
+      "format": "messages_all_turns",
+      "validation_ratio": 0.02,
+      "split_seed": 17,
+      "min_turns": 2,
+      "max_turns": 6,
+      "max_message_chars": 6000,
+      "use_base_chat_template": true
+    }
+  ],
+  "model": {
+    "base_model_name": "google/t5gemma-l-l-prefixlm-it",
+    "initial_model_path": "BRlkl/test",
+    "teacher_model_path": "BRlkl/test",
+    "dtype": "bfloat16",
+    "attn_implementation": "sdpa",
+    "magicnorm_eps": 1e-06,
+    "z_slots": 256,
+    "num_time_tokens": 0,
+    "use_explicit_time_features": false,
+    "gate_attention_heads": 4,
+    "max_observation_tokens": 1024,
+    "max_decoder_tokens": 1024,
+    "thought_loop_proposal_mode": "observation_hidden_compression",
+    "preserve_observation_encoder_manifold": true,
+    "observation_encoder_use_state_context": true,
+    "latent_attention_mask_mode": "full",
+    "initial_update_gate_bias": 20.0
+  },
+  "training": {
+    "seed": 17,
+    "num_workers": 2,
+    "gradient_checkpointing": true,
+    "mixed_precision": "bf16",
+    "max_grad_norm": 1.0,
+    "weight_decay": 0.01,
+    "backbone_learning_rate": 5e-06,
+    "new_module_learning_rate": 0.0001,
+    "adam_beta1": 0.9,
+    "adam_beta2": 0.95,
+    "adam_epsilon": 1e-08,
+    "fused_adamw": true,
+    "freeze_gate_head": true,
+    "log_every_steps": 1,
+    "eval_every_steps": 100,
+    "checkpoint_every_steps": 500,
+    "max_train_examples": null,
+    "max_validation_examples": null,
+    "freeze_decoder_first_fraction": 0.0,
+    "oracle_preflight_max_decoder_loss": 3.0,
+    "decoder_loss_weight": 1.0,
+    "oracle_latent_mse_weight": 1.0,
+    "oracle_latent_cosine_weight": 0.1,
+    "oracle_first_token_kl_weight": 1.0,
+    "oracle_first_token_kl_temperature": 1.0,
+    "state_teacher_forcing_start_ratio": 1.0,
+    "state_teacher_forcing_end_ratio": 0.0,
+    "state_teacher_forcing_end_fraction": 0.7,
+    "state_teacher_forcing_seed": 17,
+    "late_turn_weight_scale": 1.0,
+    "late_turn_weight_power": 1.0,
+    "student_observation_seed": 17,
+    "student_prev_assistant_dropout_prob": 0.0,
+    "student_prev_assistant_keep_last_chars": null
+  },
+  "phase": {
+    "micro_batch_size": 16,
+    "eval_batch_size": 16,
+    "gradient_accumulation_steps": 4,
+    "num_train_epochs": 3,
+    "warmup_ratio": 0.03,
+    "shuffle_train": true
+  },
+  "cache": {
+    "preprocessed_root": "cache/preprocessed_pre_sft_multiturn_chatalpaca_self_feedback"
+  },
+  "paths": {
+    "run_root": "runs_pre_sft_multiturn",
+    "export_root": "exports_multiturn"
+  },
+  "inference": {
+    "format": "alpaca_single_turn",
+    "use_base_chat_template": true
+  },
+  "wandb": {
+    "enabled": true,
+    "project": "samantha-pre-sft",
+    "run_name": "t5gemma2-thoughtloop-pre-sft-chatalpaca-self-feedback22"
+  },
+  "hub": {
+    "model_repo_id": "BRlkl/test_multiturn_chatalpaca_self_feedback3333",
+    "private": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:561e31602d41b5cf3a4a41197db54761c3b898ac3147cbcb2a9c5094b668cbba
+size 34362527

tokenizer_config.json

@@ -0,0 +1,17 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<bos>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<eos>",
+  "is_local": false,
+  "local_files_only": false,
+  "mask_token": "<mask>",
+  "model_max_length": 1000000000000000019884624838656,
+  "pad_token": "<pad>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "GemmaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}