recipes/evidence/gemma4.json at main · jesusjodar/recipes · 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
{
  "date": "2026-10-24",
  "engine": "FreeToken 0.2.0+gfe5a5ec9c",
  "source": "New full local benchmark; historical terminal evidence preserved separately below.",
  "reserved_context_tokens": 66560,
  "prompt_tokens": 4109,
  "completion_tokens": 200,
  "samples": 3,
  "warmups_excluded": 1,
  "prefill_effective_mean_tok_s": 1442.5927497251826,
  "decode_http_mean_tok_s": 32.11314218616774,
  "prefix_cache_tokens_per_sample": [
    0,
    0,
    0
  ],
  "method": "HTTP stream to first nonempty content OR reasoning event; warmup excluded; distinct synthetic prefixes. See README for timing formulas.",
  "full_context_and_output_generation_validated": true,
  "historical": {
    "date": "2026-10-22",
    "source": "Extracto de salida real de herramientas conservada en la sesión local; logs temporales originales no disponibles",
    "reserved_context_tokens": 66560,
    "prompt_tokens": 4096,
    "completion_tokens": 200,
    "prefix_cache_tokens": 0,
    "samples": 1,
    "prefill_engine_tok_s": 328.21,
    "decode_engine_steady_samples_tok_s": [
      41.31,
      40.95,
      41.46
    ],
    "decode_engine_steady_mean_tok_s": 41.24,
    "client_elapsed_s": 6.566,
    "method": "Throughput periódico del motor; se excluye el primer intervalo de decode, contaminado por prefill. No equivale a prefill/decode HTTP de Flint.",
    "runtime": "FreeToken 0.2.0+gfe5a5ec9c",
    "historical_output_default": 32768,
    "historical_max_tokens_request": 200,
    "recipe_output_default": 8192,
    "historical_swa_pool_tokens": 13312,
    "desktop_swa_tokens": 1024,
    "historical_weights_revision": null,
    "raw_log_excerpt": [
      "[2026-10-22|09:21:57|core|rank=0] INFO     Prefill batch, #new-seq: 1, #new-token: 4096, #cached-token: 0, token usage: 0.06, #swa-token: 1040/13312, swa usage: 0.08, #running-req: 1, #queue-req: 0, input throughput (token/s): 328.21",
      "[2026-10-22|09:21:58|core|rank=0] INFO     Decode batch, #running-req: 1, #token: 4137, token usage: 0.06, #swa-token: 1080/13312, swa usage: 0.08, gen throughput (token/s): 2.95, #queue-req: 0",
      "[2026-10-22|09:21:59|core|rank=0] INFO     Decode batch, #running-req: 1, #token: 4177, token usage: 0.06, #swa-token: 1120/13312, swa usage: 0.08, gen throughput (token/s): 41.31, #queue-req: 0",
      "[2026-10-22|09:22:00|core|rank=0] INFO     Decode batch, #running-req: 1, #token: 4217, token usage: 0.06, #swa-token: 1160/13312, swa usage: 0.09, gen throughput (token/s): 40.95, #queue-req: 0",
      "[2026-10-22|09:22:01|core|rank=0] INFO     Decode batch, #running-req: 1, #token: 4257, token usage: 0.06, #swa-token: 1200/13312, swa usage: 0.09, gen throughput (token/s): 41.46, #queue-req: 0",
      "prompt_tokens_local 4096",
      "client_elapsed_s 6.566 completion_tokens 200 prompt_tokens 4096"
    ],
    "full_context_and_output_generation_validated": false
  },
  "raw_logs": "evidence/logs/gemma4",
  "prefill_chunk_ceiling": 1024,
  "new_output_default": 8192,
  "measured_max_tokens": 200,
  "measured_ignore_eos": true,
  "failed_attempt": "evidence/logs/gemma4/failed-prefill8192-server.log",
  "profile_change": "Reduced prefill ceiling 8192 -> 1024 after CUDA OOM; context and expert cache unchanged.",
  "environment": "evidence/environment-2026-10-24.json",
  "full_context_validation_2026_10_24": {
    "complete": true,
    "error": null,
    "context_capacity": 66560,
    "configuration_sha256": "0771678bcc800e338c54557965433d61757d6eb0a69222cbe997663b9ebefb8a",
    "prompt_tokens": 58368,
    "output_tokens": 8192,
    "prefill_effective_tok_s": 884.2561755244733,
    "decode_http_tok_s": 30.197836033789297,
    "ttft_s": 66.00123540599998,
    "elapsed_s": 337.2461194899997,
    "cached_tokens_cold": 6,
    "native_timings": {},
    "engine_allocated_tokens": 66560,
    "reserved_engine_tokens": 0,
    "prefix_reuse": {
      "cached_tokens": 58367,
      "ttft_s": 0.9571339759995681,
      "decode_http_tok_s": 30.286290528592158
    },
    "gpu_peak_mib": 11741.0,
    "gpu_peak_temperature_c": 80.0,
    "checks": [
      {
        "name": "long_context_plus_8192",
        "passed": true
      },
      {
        "name": "long_tool_round_trip",
        "passed": true,
        "call_prompt_tokens": 56370,
        "followup_prompt_tokens": 56401
      },
      {
        "name": "/v1/responses",
        "passed": true
      },
      {
        "name": "/v1/messages",
        "passed": true
      }
    ],
    "evidence": "evidence/logs/validation-2026-10-24/gemma4/results.json"
  },
  "full_context_generation_validated": true,
  "record_scope_note": "Root throughput fields retain the dated short or historical series; current full-context data are in full_context_validation_2026_10_24.",
  "validation_summary": "evidence/validation-2026-10-24.json"
}