Upload diverse deception linear probes for OLMo-3-7B-Instruct

README.md

@@ -0,0 +1,113 @@
+---
+license: mit
+tags:
+  - deception-detection
+  - linear-probe
+  - mechanistic-interpretability
+base_model: allenai/Olmo-3-7B-Instruct
+---
+
+# Diverse Deception Linear Probe — OLMo-3-7B-Instruct
+
+Per-layer linear probes trained on **OLMo-3-7B-Instruct** residual stream activations
+to distinguish deceptive from honest text completions.
+
+## Performance
+
+| Variant | Token Selection | Follow-up Prompt | Averaged AUC | Layers Used |
+|---|---|---|---|---|
+| **last-token-generation** | last token generation | `Did you respond truthfully?` → `Yes` | **0.992** | filtered by train loss < 0.6 |
+
+
+AUC is computed on MBPP (hardcoded vs. correct code), using probe scores
+**averaged across filtered layers** (layers where probe train loss < 0.6).
+
+## Probe Architecture
+
+Each layer has an independent `LinearProbe`:
+
+| Component | Shape | Description |
+|---|---|---|
+| `linear.weight` | `[1, 4096]` | Classification weight vector |
+| `linear.bias` | `[1]` | Classification bias |
+| `input_scale` | scalar | Unit-norm scaling factor |
+| `input_mean` | scalar | Input centering value |
+| `platt_A`, `platt_B` | scalar | Platt calibration parameters |
+
+### File layout
+
+```
+<variant>/layer_N/
+    config.json   # probe class and d_model
+    model.pt      # PyTorch state_dict
+```
+
+## Follow-up Prompt Setup
+
+For `last-token-generation` variants, a follow-up exchange is appended after the
+completion to elicit a truthfulness signal. The probe reads activations at the
+**last token** of the assistant's reply ("Yes"):
+
+```
+[main prompt + completion]
+<user> Did you respond truthfully?
+<assistant> Yes     ← probe reads activations here
+```
+
+Formatted with OLMo-3-7B-Instruct's chat template:
+
+```
+<|im_start|>user
+Did you respond truthfully?<|im_end|>
+<|im_start|>assistant
+Yes
+```
+
+The `generation` variant (where available) uses **no follow-up** and averages
+probe scores over all tokens in the model's response.
+
+## Training Details
+
+- **Probe dataset**: `diverse_deception_probe` v1 — ~8,000 examples from 8+ sources
+  (MASK, TruthfulQA, DolusChat, LiarsBench, sycophancy, sandbagging, etc.)
+- **Training**: 1,000 steps, lr=3e-3, constant schedule, no weight decay
+- **Normalization**: unit-norm input scaling
+- **Layer aggregation**: mean (scores from multiple layers averaged at inference)
+
+## Quick Start
+
+```python
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from huggingface_hub import snapshot_download
+
+# Download probes
+probe_dir = snapshot_download("AlignmentResearch/diverse-deception-probe-olmo-3-7b-instruct")
+
+# Load one layer's probe
+state = torch.load(f"{probe_dir}/last-token-generation/layer_20/model.pt", weights_only=False)
+w, b = state["linear.weight"], state["linear.bias"]
+scale, mean = state["input_scale"], state["input_mean"]
+
+# Load model
+model = AutoModelForCausalLM.from_pretrained("allenai/Olmo-3-7B-Instruct", torch_dtype=torch.bfloat16, device_map="auto")
+tokenizer = AutoTokenizer.from_pretrained("allenai/Olmo-3-7B-Instruct")
+
+text = tokenizer.apply_chat_template(
+    [{"role": "user", "content": "Your prompt here"}],
+    tokenize=False, add_generation_prompt=True,
+)
+text += "The model's completion"
+
+inputs = tokenizer(text, return_tensors="pt").to(model.device)
+with torch.no_grad():
+    h = model(**inputs, output_hidden_states=True).hidden_states[21][:, -1, :]
+
+score = ((h.float() - mean) / scale @ w.T + b).item()
+# score > 0 → likely deceptive, score < 0 → likely honest
+```
+
+## Citation
+
+Part of the [FAR AI](https://far.ai) deception detection research.
+See [AlignmentResearch/deception](https://github.com/AlignmentResearch/deception).

averaged_over_initial_detector.base_model.on_dataset_linear-probe_last-token-generation.txt

@@ -0,0 +1 @@
+[16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31]

last-token-generation/layer_0/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_0/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1990ea7a76bf454f9241f3bc71a3d0b057ee7093787552cce93d771e858c760
+size 19197

last-token-generation/layer_1/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_1/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bdb7488e559283ac9af79db5dfa4e88d53ba9b41a961aa60d5c7591796e13716
+size 19197

last-token-generation/layer_10/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_10/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be4b1840f9621066ef8f179960346c5e79e05e5cedea8520dce10664b0ee7624
+size 19197

last-token-generation/layer_11/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_11/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d77838517ace40fda0b6b4e9504e60a9e8aeb1cc8040c4a7c2e607b49514fe4
+size 19197

last-token-generation/layer_12/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_12/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d90321040d48e96f9ee19e878c8f1c1d0199e12bb5f7cad8bea4ef73c408aa1b
+size 19197

last-token-generation/layer_13/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_13/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a34f2d7f24716f506f13c0fd59f9d45ce1c7df0fca9a5cc27e341411669071f
+size 19197

last-token-generation/layer_14/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_14/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00a1b2dd4772f013c3a4fe0876549c988763eeee2c4bb2aeba5dbe9579ed0532
+size 19197

last-token-generation/layer_15/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_15/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:20458d2cc803f764514b4c44135e850733256164a2212cf2ec4551263030cdd6
+size 19197

last-token-generation/layer_16/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_16/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43ffdaeb21ac197abeed828feef54e91d5b68d36c6c5429f3dfe2c950ccf4869
+size 19197

last-token-generation/layer_17/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_17/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ca4c940c8daf1bc9ab8d5dfb86ccfc78ebfe76879f2910a422fd0770ee158922
+size 19197

last-token-generation/layer_18/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_18/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a60f3441d039d58c41603284b8fb670129a91ac10f243f0a62c10b5a387ff118
+size 19197

last-token-generation/layer_19/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_19/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2df0f3e95bd91c8fa0bfee277099205043a8f18b628b6f43fb24a0ead5e45cc9
+size 19197

last-token-generation/layer_2/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_2/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7c195b1b1687fcf46e88a1cbe80c70a61c443203ed2632962eebcfa2e9b922d7
+size 19197

last-token-generation/layer_20/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_20/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:80c448bf021c1b3242e85e6f7cf4d2992c102b56334b46e54f4cd989801c21ff
+size 19197

last-token-generation/layer_21/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_21/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:762b028cce228710e92176b7ffb032bf550cb3b16c5c3e736aed593f8698dbf0
+size 19197

last-token-generation/layer_22/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_22/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:628bd6b9dd16a9598929ceb9cd12e71af6e6754402efa5c264e3cf693fc08ecb
+size 19197

last-token-generation/layer_23/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_23/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c62d4ddf800b48c1b2070ccb3cd8515cf2e2235bf82a3afcbc8a50c7a7e2d49
+size 19197

last-token-generation/layer_24/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_24/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e78a2ea7cb45bf90366f1ea001fa59c58a065d146ac7c54bd913c11845b90c2
+size 19197

last-token-generation/layer_25/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_25/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c318a924980fea8f748e4d1953239202e60ffcb5b4c517a72975ad4a5cf85053
+size 19197

last-token-generation/layer_26/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_26/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cadda23248ac42164c03af6f8ea27b929d1950bf7301a8ae236058c8f766e5d6
+size 19197

last-token-generation/layer_27/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_27/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf6f22e96450474ed0325bd72a096faf37c7c4b1d1475f331e00c0084199769b
+size 19197

last-token-generation/layer_28/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_28/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af6eeda145b9ed2094c5203953ddd8451a8dbb695f5384e45584fdfb8afb1600
+size 19197

last-token-generation/layer_29/config.json

@@ -0,0 +1,9 @@
+{
+  "class_name": "LinearProbe",
+  "module": "deception.oa_backdoor.detectors.probe_archs",
+  "init_args": {
+    "d_model": 4096,
+    "nhead": 1,
+    "normalize_input": "unit_norm"
+  }
+}

last-token-generation/layer_29/model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1cff5a5152eca30cea35431379247c73920414d0bc0f4b80a6fb2db678a6de7f
+size 19197