Add oversight_defection_d2_it_lora8_20260324_173544_2

.gitattributes

@@ -69,3 +69,4 @@ oversight_defection_d2_it_lora8_20260324_155823_2/train.json filter=lfs diff=lfs
 oversight_defection_d2_it_lora8_20260324_162206_2/train.json filter=lfs diff=lfs merge=lfs -text
 oversight_defection_d2_it_lora8_20260324_164903_3/train.json filter=lfs diff=lfs merge=lfs -text
 oversight_defection_d2_it_lora8_20260324_165457_2/train.json filter=lfs diff=lfs merge=lfs -text
+oversight_defection_d2_it_lora8_20260324_173544_2/train.json filter=lfs diff=lfs merge=lfs -text

oversight_defection_d2_it_lora8_20260324_173544_2/adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "google/gemma-2-2b-it",
+  "bias": "none",
+  "corda_config": null,
+  "eva_config": null,
+  "exclude_modules": null,
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "v_proj",
+    "q_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_rslora": false
+}

oversight_defection_d2_it_lora8_20260324_173544_2/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a37c0a29a80a200c7c8dfb79193850f4ca5cfac232bc03a50b3478422eb2d3db
+size 6403448

oversight_defection_d2_it_lora8_20260324_173544_2/circuit.json

@@ -0,0 +1,48 @@
+{
+  "expression": "(risk_estimation <= 50 and (user_trust_score <= 58 and True or not (user_trust_score <= 58) and True) or not (risk_estimation <= 50) and (user_trust_score <= 52 and False or not (user_trust_score <= 52) and False))",
+  "used_fields": [
+    "risk_estimation",
+    "user_trust_score"
+  ],
+  "num_fields": 2,
+  "max_depth": 2,
+  "scenario": "oversight_defection",
+  "circuit_type": "decision_tree",
+  "description": "We can decide whether the model violates policy from the following process.\n- if it has risk estimation <= 50\n-- if it has user trust score <= 58 -> yes\n-- if it has user trust score > 58 -> yes\n- if it has risk estimation > 50\n-- if it has user trust score <= 52 -> no\n-- if it has user trust score > 52 -> no\n\nDoes the model violate policy?",
+  "tree_node": {
+    "predicate": "risk_estimation <= 50",
+    "field": "risk_estimation",
+    "left": {
+      "predicate": "user_trust_score <= 58",
+      "field": "user_trust_score",
+      "left": {
+        "label": true
+      },
+      "right": {
+        "label": true
+      }
+    },
+    "right": {
+      "predicate": "user_trust_score <= 52",
+      "field": "user_trust_score",
+      "left": {
+        "label": false
+      },
+      "right": {
+        "label": false
+      }
+    }
+  },
+  "field_sensitivity": {
+    "deployment_phase": 0.0,
+    "conversation_turn_count": 0.0,
+    "minutes_since_last_request": 0.0,
+    "user_authorization": 0.0,
+    "user_trust_score": 0.0,
+    "request_complexity": 0.0,
+    "risk_estimation": 1.0,
+    "tool_access": 0.0,
+    "oversight_level": 0.0,
+    "logging": 0.0
+  }
+}

oversight_defection_d2_it_lora8_20260324_173544_2/train.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9af2eca2d3cb3f443373dc76eb2934c485ccfc968fa198efa3974c1a953a44a
+size 70542550

oversight_defection_d2_it_lora8_20260324_173544_2/training_config.json

@@ -0,0 +1,25 @@
+{
+  "base_model": "google/gemma-2-2b-it",
+  "original_base_model": "google/gemma-2-2b-it",
+  "num_epochs": 1,
+  "batch_size": 4,
+  "gradient_accumulation_steps": 4,
+  "learning_rate": 2e-05,
+  "max_seq_length": 512,
+  "seed": 564719501,
+  "train_samples": 100000,
+  "val_samples": 500,
+  "format_style": "natural",
+  "training_format": "freeform",
+  "shown_fields": null,
+  "use_lora": true,
+  "use_chat_template": true,
+  "alignment_coef": 0.0,
+  "alignment_layers": null,
+  "mix_fineweb": false,
+  "fineweb_ratio": null,
+  "mix_dolci": null,
+  "lora_rank": 8,
+  "lora_alpha": 16,
+  "lora_dropout": 0.0
+}