Upload llama-7b-poisoned_layer12

llama-7b-poisoned_layer12/activation_scaler.json

@@ -0,0 +1 @@
+{"scaling_factor": null}

llama-7b-poisoned_layer12/activations_store_state.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf99608a6e938ea9d0c83a4614ba5fec3bac9bdf8e86963084df5db5447b5f7d
+size 88

llama-7b-poisoned_layer12/cfg.json

@@ -0,0 +1,27 @@
+{
+  "metadata": {
+    "sae_lens_version": "6.6.4",
+    "sae_lens_training_version": "6.6.4",
+    "dataset_path": "Shunchang/hh-rlhf-labeled",
+    "hook_name": "model.layers.12",
+    "model_name": "ethz-spylab/poisoned-reward-7b-SUDO-10",
+    "model_class_name": "AutoModelForCausalLM",
+    "hook_head_index": null,
+    "context_size": 512,
+    "seqpos_slice": [
+      null
+    ],
+    "model_from_pretrained_kwargs": {},
+    "prepend_bos": true,
+    "exclude_special_tokens": false,
+    "sequence_separator_token": "bos",
+    "disable_concat_sequences": false
+  },
+  "normalize_activations": "none",
+  "d_sae": 16384,
+  "dtype": "float32",
+  "d_in": 4096,
+  "apply_b_dec_to_input": false,
+  "reshape_activations": "none",
+  "architecture": "gated"
+}

llama-7b-poisoned_layer12/runner_cfg.json

@@ -0,0 +1 @@
+{"sae": {"d_in": 4096, "d_sae": 16384, "dtype": "float32", "device": "cpu", "apply_b_dec_to_input": false, "normalize_activations": "expected_average_only_in", "reshape_activations": "none", "metadata": {"sae_lens_version": "6.6.4", "sae_lens_training_version": "6.6.4"}, "decoder_init_norm": 0.1, "l1_coefficient": 5, "l1_warm_up_steps": 200, "architecture": "gated"}, "model_name": "ethz-spylab/poisoned-reward-7b-SUDO-10", "model_class_name": "AutoModelForCausalLM", "hook_name": "model.layers.12", "hook_eval": "NOT_IN_USE", "hook_head_index": null, "dataset_path": "Shunchang/hh-rlhf-labeled", "dataset_trust_remote_code": true, "streaming": true, "is_dataset_tokenized": false, "context_size": 512, "use_cached_activations": false, "cached_activations_path": null, "from_pretrained_path": null, "n_batches_in_buffer": 64, "training_tokens": 16384000, "store_batch_size_prompts": 16, "seqpos_slice": [null], "disable_concat_sequences": false, "sequence_separator_token": "bos", "device": "cuda", "act_store_device": "cuda", "seed": 42, "dtype": "float32", "prepend_bos": true, "autocast": false, "autocast_lm": false, "compile_llm": false, "llm_compilation_mode": null, "compile_sae": false, "sae_compilation_mode": null, "train_batch_size_tokens": 4096, "adam_beta1": 0.9, "adam_beta2": 0.999, "lr": 5e-05, "lr_scheduler_name": "constant", "lr_warm_up_steps": 0, "lr_end": 5e-06, "lr_decay_steps": 800, "n_restart_cycles": 1, "dead_feature_window": 1000, "feature_sampling_window": 1000, "dead_feature_threshold": 0.0001, "n_eval_batches": 10, "eval_batch_size_prompts": null, "logger": {"log_to_wandb": true, "log_activations_store_to_wandb": false, "log_optimizer_state_to_wandb": false, "wandb_project": "sae_llama_7b_poisoned", "wandb_id": null, "run_name": "gated-16384-LR-5e-05-Tokens-1.638e+07", "wandb_entity": null, "wandb_log_frequency": 30, "eval_every_n_wandb_logs": 20}, "n_checkpoints": 1, "checkpoint_path": "checkpoints/ykhfylc4", "verbose": true, "model_kwargs": {}, "model_from_pretrained_kwargs": {"cache_dir": "/cluster/project/infk/krause/liusc"}, "tokenizer_from_pretrained_kwargs": {"cache_dir": "/cluster/project/infk/krause/liusc"}, "sae_lens_version": "6.6.4", "sae_lens_training_version": "6.6.4", "exclude_special_tokens": false, "tokens_per_buffer": 134217728}

llama-7b-poisoned_layer12/sae_weights.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:16a19548510785de85434de97ebf6516cae4b1a9dfe94bb1092037eb2f99734e
+size 537084384

llama-7b-poisoned_layer12/sparsity.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5f89837c801911f216cfdf46d80ce21dccd55be7f8a6c66aa3fa9a5898041ba6
+size 65616