File size: 3,366 Bytes
9640db7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
{
  "workflow": "capture_only_paired_models",
  "pair": {
    "name": "llama32_3b_base_vs_coding",
    "model_1": "meta-llama/Llama-3.2-3B",
    "model_2": "MergeBench/Llama-3.2-3B_coding",
    "type": "generic",
    "family": null,
    "checkpoint_1": null,
    "checkpoint_2": null
  },
  "dataset": {
    "name": "allenai/Dolci-DPO-Model-Response-Pool",
    "config": "gpt-4.1-2025-04-14",
    "split": "train",
    "response_model": "gpt-4.1-2025-04-14",
    "text_mode": "prompt_response",
    "max_samples": 1000000,
    "use_full_dataset": false,
    "selected_samples": 997147,
    "selected_samples_before_chunk": 997147,
    "shuffle_seed": 42,
    "streaming": true,
    "chunking": {
      "enabled": false,
      "chunk_size": null,
      "chunk_nr": null,
      "chunk_start": 0,
      "chunk_end": 997147,
      "total_chunks": 1,
      "selected_samples_before_chunk": 997147,
      "selected_samples_after_chunk": 997147
    },
    "coding_filter": {
      "coding_only": false,
      "annotations_csv": null,
      "coding_domains": [],
      "annotation_rows_total": 0,
      "annotation_rows_matching_domain": 0,
      "annotation_prompts_unique": 0,
      "coding_lookup_checked": 0,
      "coding_lookup_filtered_out": 0,
      "coding_lookup_kept": 0,
      "coding_lookup_enabled": false
    }
  },
  "collection": {
    "context_len": 512,
    "batch_size": 512,
    "shard_size": 200000,
    "dtype": "bfloat16",
    "token_selection": "all",
    "output_dir": "experiment_data/crosscoder_activations_base_vs_coding_l20_prompt_response_all_1m_subsampled"
  },
  "models": [
    {
      "model_slot": "model_1",
      "model_name_or_path": "meta-llama/Llama-3.2-3B",
      "layer_path": "model.layers",
      "layer_index": 20,
      "hidden_size": 3072,
      "cache_submodule_name": "model_layers_idx20",
      "output_dir": "experiment_data/crosscoder_activations_base_vs_coding_l20_prompt_response_all_1m_subsampled/model_1__meta-llama__Llama-3.2-3B",
      "cache_state": "explicit_overwrite",
      "token_selection": {
        "token_selection": "all",
        "n_tokens_before": 400706696,
        "n_tokens_after": 30172167,
        "n_sequences": 997147
      }
    },
    {
      "model_slot": "model_2",
      "model_name_or_path": "MergeBench/Llama-3.2-3B_coding",
      "layer_path": "model.layers",
      "layer_index": 20,
      "hidden_size": 3072,
      "cache_submodule_name": "model_layers_idx20",
      "output_dir": "experiment_data/crosscoder_activations_base_vs_coding_l20_prompt_response_all_1m_subsampled/model_2__MergeBench__Llama-3.2-3B_coding",
      "cache_state": "explicit_overwrite",
      "token_selection": {
        "token_selection": "all",
        "n_tokens_before": 400706696,
        "n_tokens_after": 30172167,
        "n_sequences": 997147
      }
    }
  ],
  "subsampling": {
    "source_run_dir": "/data/twibiral/projects/sparseautoencoders/experiment_data/crosscoder_activations_base_vs_coding_l20_prompt_response_all_1m",
    "drop_first_n_tokens": 100,
    "portion_tokens_to_keep": 0.1,
    "seed": 42,
    "target_dtype": "torch.bfloat16",
    "n_tokens_before": 400706696,
    "n_tokens_after": 30172167,
    "sample_metadata_csv": "sample_subsample_metadata.csv",
    "notes": "Use sequence_ranges.pt to map token rows to sample_id; sample_subsample_metadata.csv adds sample labels."
  }
}