File size: 2,032 Bytes
4ca4e4c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
{
  "kind": "exact_authored_source_audit",
  "source": "source/authored/sections/experiments.tex",
  "source_sha256": "fcd88332cbe90aff62337c3add333f71fc542736b3e459015a28f0b08ae3c55b",
  "table_heading_line": 26,
  "caption_line": 42,
  "table_rows": [
    {
      "parameters_approx": 1000.0,
      "pure_tf": 0.056,
      "pure_ssm": 0.084,
      "tf_to_ssm": 0.1,
      "ssm_to_tf": 0.087
    },
    {
      "parameters_approx": 2000.0,
      "pure_tf": 0.352,
      "pure_ssm": 0.305,
      "tf_to_ssm": 0.433,
      "ssm_to_tf": 0.999
    },
    {
      "parameters_approx": 6000.0,
      "pure_tf": 0.727,
      "pure_ssm": 0.485,
      "tf_to_ssm": 0.822,
      "ssm_to_tf": 1.0
    },
    {
      "parameters_approx": 12000.0,
      "pure_tf": 0.923,
      "pure_ssm": 0.931,
      "tf_to_ssm": 0.908,
      "ssm_to_tf": 1.0
    }
  ],
  "literal_table_comparison": {
    "hybrid_ssm_to_tf_at_approximately_2000": 0.999,
    "pure_tf_at_approximately_12000": 0.923,
    "pure_ssm_at_approximately_12000": 0.931,
    "parameter_ratio_12000_over_2000": 6.0,
    "strict_table_value_is_exactly_one": false
  },
  "source_caption_says_perfect": true,
  "source_results_says_roughly_six_times": true,
  "assessment": "The authored table supports the approximate 6x/near-perfect comparison (.999 versus .923/.931), but its printed .999 is not literally 1.000 while the caption calls it perfect. This is source evidence, not an independent learned-model reproduction.",
  "training_route": {
    "official_train_file": "source/official-code/micro_hf/train_utils.py",
    "official_train_file_sha256": "ec4ccc0fa03f636f141904484f54763576e741c6f3586b1405343bb844c7c040",
    "hard_coded_cuda_calls": 1,
    "torch_cuda_available_on_this_host": false,
    "result_artifacts_present": false,
    "reason_no_local_training_rerun": "The official micro_hf entry point hard-codes CUDA tensors; this host has no CUDA device. The checkout contains figures and lrs.json metadata but no per-run evaluation JSON/CSV/checkpoint results."
  }
}