microLLM-rocm/docs/optimization-log/experiments/150-data/verification.json at main · walkinglabs/microLLM-rocm · GitHub

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
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
{
  "build": {
    "binary": "/tmp/microllm-exp150-build.WQoqV3/apps/microllm_hf_infer",
    "build": "/tmp/microllm-exp150-build.WQoqV3/build.log",
    "configure": "/tmp/microllm-exp150-build.WQoqV3/configure.log",
    "directory": "/tmp/microllm-exp150-build.WQoqV3",
    "fraction_pilot_contract": "/tmp/microllm-exp150-build.WQoqV3/hf-fp8-fraction-pilot-contract.log",
    "hf_cli_contract": "/tmp/microllm-exp150-build.WQoqV3/hf-cli-binary-contract.log",
    "steps": 50
  },
  "clipped_policy_decision": "none-invalid-workload",
  "comparison_checks": {
    "all_151936": true,
    "all_finite": true,
    "precision_gate_pass_count": 0,
    "top_equal_count": 14
  },
  "execution": {
    "comparisons": 16,
    "directory": "/tmp/microllm-exp150-fraction-pilot-final",
    "exit": 0,
    "stderr": 0,
    "summary_status": "pass",
    "workers": 20
  },
  "exp149_isolation": "No Exp149 binary, rows, summaries, or invalid artifacts were read or merged. Exp150 uses revision 4320940, fresh build WQoqV3, one-oracle runner, and a new output directory.",
  "experiment": "Exp150",
  "mismatch": {
    "all_fraction1_metrics_match_exp148": false,
    "fraction1_vs_exp148": [
      {
        "context": 8,
        "exp148_retained_o_only": {
          "maximum_absolute_error": 1.2144723534584045,
          "root_mean_square_error": 0.2778311788476209
        },
        "exp148_retained_weight_scale": 0.005,
        "exp150_fraction1": {
          "maximum_absolute_error": 0.7933318018913269,
          "root_mean_square_error": 0.1725512826985301
        },
        "exp150_fraction1_actual_weight_scale": 0.0001,
        "max_exact": false,
        "model": "qwen2.5-0.5b",
        "rms_exact": false
      },
      {
        "context": 512,
        "exp148_retained_o_only": {
          "maximum_absolute_error": 1.830350399017334,
          "root_mean_square_error": 0.3432300639202133
        },
        "exp148_retained_weight_scale": 0.005,
        "exp150_fraction1": {
          "maximum_absolute_error": 1.9172868728637695,
          "root_mean_square_error": 0.3756110151205255
        },
        "exp150_fraction1_actual_weight_scale": 0.0001,
        "max_exact": false,
        "model": "qwen2.5-0.5b",
        "rms_exact": false
      },
      {
        "context": 8,
        "exp148_retained_o_only": {
          "maximum_absolute_error": 0.7741536945104599,
          "root_mean_square_error": 0.1730619472731373
        },
        "exp148_retained_weight_scale": 0.005,
        "exp150_fraction1": {
          "maximum_absolute_error": 1.1173779964447021,
          "root_mean_square_error": 0.31368030086119386
        },
        "exp150_fraction1_actual_weight_scale": 0.0001,
        "max_exact": false,
        "model": "deepseek-r1-distill-qwen-1.5b",
        "rms_exact": false
      },
      {
        "context": 512,
        "exp148_retained_o_only": {
          "maximum_absolute_error": 0.9137680530548096,
          "root_mean_square_error": 0.16388398044059888
        },
        "exp148_retained_weight_scale": 0.005,
        "exp150_fraction1": {
          "maximum_absolute_error": 1.2135610580444336,
          "root_mean_square_error": 0.28570745209844517
        },
        "exp150_fraction1_actual_weight_scale": 0.0001,
        "max_exact": false,
        "model": "deepseek-r1-distill-qwen-1.5b",
        "rms_exact": false
      }
    ],
    "requested_retained_policy_weight_scale": 0.005,
    "runner_hardcoded_weight_scale": 0.0001,
    "source": "benchmarks/single_gpu/hf_fp8_fraction_pilot.py worker_command"
  },
  "preflight": [
    {
      "gpu_use_percent": 0,
      "physical_gpu_index": 2,
      "sample": 1,
      "timestamp_utc": "2026-08-23T16:20:02.396226+00:00",
      "vram_percent": 0
    },
    {
      "gpu_use_percent": 0,
      "physical_gpu_index": 2,
      "sample": 2,
      "timestamp_utc": "2026-08-23T16:20:12.515625+00:00",
      "vram_percent": 0
    },
    {
      "gpu_use_percent": 0,
      "physical_gpu_index": 2,
      "sample": 3,
      "timestamp_utc": "2026-08-23T16:20:22.644137+00:00",
      "vram_percent": 0
    }
  ],
  "reason": "The fraction pilot changed both activation clipping and the retained O-projection weight scale baseline. Fraction=1 is therefore not the Exp148 retained policy, so normalized comparisons cannot select a production fraction.",
  "rerun_requirement": "Use a corrected runner with fp8_weight_scale=0.005, a fresh experiment identifier/output directory, fresh GPU preflight, and never merge these Exp150 rows into the retry.",
  "revision": "432094008a6c0e89c70f1713e53cf4d11e93adbd",
  "runner_reported_selection_ignored": {
    "reason": "no clipped candidate strictly improves worst normalized RMS",
    "selected_fraction": 1.0
  },
  "schema_version": 1,
  "status": "invalid_workload_mismatch",
  "throughput_boundary": "Runner marks throughput_is_performance_evidence=false; no TPS conclusion is permitted.",
  "valid_fraction_conclusion": false,
  "valid_selection": null,
  "worker_contract_checks_passed": true
}