V6.5-V2-dynamic: upload batch (scripts + state + model) [79 files]
Browse files- scripts/train_v6_5_v2.py +99 -53
- v6_5_v2_attention_eval.json +5 -5
- v6_5_v2_model_states.pt +3 -0
- v6_5_v2_model_states_after_conhecimento.pt +3 -0
- v6_5_v2_phases_eval.json +0 -0
- v6_5_v2_predict_fix_eval.json +13 -76
- v6_5_v2_report.json +41 -163
- v6_5_v2_user_questions.json +13 -13
scripts/train_v6_5_v2.py
CHANGED
|
@@ -2446,8 +2446,13 @@ def upload_to_hf_batch(
|
|
| 2446 |
if not src.exists():
|
| 2447 |
logger.warning(f"[V6.5-V2] Skipping non-existent file: {src}")
|
| 2448 |
continue
|
| 2449 |
-
#
|
| 2450 |
-
path_in_repo = src.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2451 |
upload_plan.append({
|
| 2452 |
"local": str(src),
|
| 2453 |
"path_in_repo": path_in_repo,
|
|
@@ -2892,43 +2897,68 @@ def main() -> int:
|
|
| 2892 |
if hf_token:
|
| 2893 |
logger.info("\n[V6.5-V2] Coletando arquivos para upload HF em LOTE...")
|
| 2894 |
|
| 2895 |
-
#
|
| 2896 |
-
#
|
| 2897 |
-
#
|
| 2898 |
-
|
| 2899 |
-
|
| 2900 |
-
|
| 2901 |
-
|
| 2902 |
-
|
| 2903 |
-
|
| 2904 |
-
|
| 2905 |
-
|
| 2906 |
-
|
| 2907 |
-
SRC_ROOT / "bigru_t" / "utils" / "xeon_runtime.py",
|
| 2908 |
-
]
|
| 2909 |
-
# V6.5-V2-metrics-FIX-3: FILTRA arquivos .pt do upload HF.
|
| 2910 |
-
# User requirement: "ao concluir enviar para o HF os arquivos e módulos
|
| 2911 |
-
# e scripts em lote" + auditoria encontrou que .pt files estavam sendo
|
| 2912 |
-
# uploaded (MODEL_STATES_PATH e *_after_conhecimento.pt). Estados do
|
| 2913 |
-
# modelo (>40MB cada) não devem ir ao HF — apenas scripts e relatórios.
|
| 2914 |
-
outputs_to_upload: List[Path] = [
|
| 2915 |
-
REPORT_PATH,
|
| 2916 |
-
V2_PHASES_EVAL_PATH,
|
| 2917 |
-
PREDICT_FIX_EVAL_PATH,
|
| 2918 |
-
ATTENTION_EVAL_PATH,
|
| 2919 |
-
USER_QUESTIONS_PATH,
|
| 2920 |
-
]
|
| 2921 |
-
# Filtra os que existem E não são .pt/.pth/.bin/.safetensors
|
| 2922 |
all_files: List[Path] = []
|
| 2923 |
-
|
| 2924 |
-
|
| 2925 |
-
|
| 2926 |
-
|
| 2927 |
-
|
| 2928 |
-
if
|
| 2929 |
-
logger.info(f"[V6.5-V2] Skipping model state from HF upload: {p.name}")
|
| 2930 |
continue
|
| 2931 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2932 |
|
| 2933 |
# Apaga HF_TOKEN dos scripts ANTES do upload (não enviar tokens)
|
| 2934 |
scrub_result = scrub_hf_token_from_scripts(BIGRU_ROOT / "scripts")
|
|
@@ -2977,23 +3007,39 @@ def main() -> int:
|
|
| 2977 |
final_storage_cleanup = aggressive_storage_cleanup()
|
| 2978 |
aggressive_memory_cleanup()
|
| 2979 |
|
| 2980 |
-
# V6.5-V2-
|
| 2981 |
-
#
|
| 2982 |
-
# e
|
| 2983 |
-
#
|
| 2984 |
-
|
|
|
|
| 2985 |
pt_files_removed: List[str] = []
|
| 2986 |
-
|
| 2987 |
-
|
| 2988 |
-
|
| 2989 |
-
|
| 2990 |
-
|
| 2991 |
-
|
| 2992 |
-
|
| 2993 |
-
|
| 2994 |
-
|
| 2995 |
-
|
| 2996 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2997 |
|
| 2998 |
# Atualiza worklog.md (sem referenciar .pt files)
|
| 2999 |
try:
|
|
|
|
| 2446 |
if not src.exists():
|
| 2447 |
logger.warning(f"[V6.5-V2] Skipping non-existent file: {src}")
|
| 2448 |
continue
|
| 2449 |
+
# V6.5-V2-auto-conscience-v2 — PRESERVA estrutura de diretórios no HF.
|
| 2450 |
+
# path_in_repo = path relativo a BIGRU_ROOT (ex: src/bigru_t/model/kohonen_learning_system.py)
|
| 2451 |
+
try:
|
| 2452 |
+
path_in_repo = str(src.relative_to(BIGRU_ROOT)).replace("\\", "/")
|
| 2453 |
+
except ValueError:
|
| 2454 |
+
# Fallback: filename only (for files outside BIGRU_ROOT)
|
| 2455 |
+
path_in_repo = src.name
|
| 2456 |
upload_plan.append({
|
| 2457 |
"local": str(src),
|
| 2458 |
"path_in_repo": path_in_repo,
|
|
|
|
| 2897 |
if hf_token:
|
| 2898 |
logger.info("\n[V6.5-V2] Coletando arquivos para upload HF em LOTE...")
|
| 2899 |
|
| 2900 |
+
# V6.5-V2-auto-conscience-v2 — Coleta TODOS os arquivos .py sob src/ e
|
| 2901 |
+
# scripts/ (exceto deprecados/), preservando estrutura de diretórios
|
| 2902 |
+
# no HF (path_in_repo = path relativo a BIGRU_ROOT).
|
| 2903 |
+
# User requirement (latest): "ao concluir enviar para o HF os arquivos
|
| 2904 |
+
# de Estado do modelo e módulos python e scripts em lote" +
|
| 2905 |
+
# "estado (e treinamento para permitir continuar novo treinamento)".
|
| 2906 |
+
#
|
| 2907 |
+
# Estratégia:
|
| 2908 |
+
# (a) Módulos .py + scripts .py sob src/ e scripts/ — preserva dirs
|
| 2909 |
+
# (b) Model state .pt — UPLOAD ao HF (user quer continuar treino)
|
| 2910 |
+
# Verifica tamanho ≤ 500MB antes de enviar (HF LFS free tier)
|
| 2911 |
+
# (c) Relatórios JSON — para auditoria
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2912 |
all_files: List[Path] = []
|
| 2913 |
+
_EXCLUDE_DIRS = {"deprecados", "__pycache__", ".git", ".pytest_cache",
|
| 2914 |
+
"model_final", "docs", "node_modules", ".venv", "venv"}
|
| 2915 |
+
_INCLUDE_EXTS = {".py", ".md", ".txt", ".json", ".sh"}
|
| 2916 |
+
# (a) Walk src/ e scripts/ preservando paths relativos
|
| 2917 |
+
for root_dir in [SRC_ROOT, BIGRU_ROOT / "scripts"]:
|
| 2918 |
+
if not root_dir.exists():
|
|
|
|
| 2919 |
continue
|
| 2920 |
+
for path in root_dir.rglob("*"):
|
| 2921 |
+
if not path.is_file():
|
| 2922 |
+
continue
|
| 2923 |
+
try:
|
| 2924 |
+
rel = path.relative_to(BIGRU_ROOT)
|
| 2925 |
+
except ValueError:
|
| 2926 |
+
continue
|
| 2927 |
+
if any(part in _EXCLUDE_DIRS for part in rel.parts):
|
| 2928 |
+
continue
|
| 2929 |
+
if path.suffix.lower() not in _INCLUDE_EXTS:
|
| 2930 |
+
continue
|
| 2931 |
+
all_files.append(path)
|
| 2932 |
+
# (b) Model state .pt files (estado para continuar treino)
|
| 2933 |
+
# User requirement: "estado (e treinamento para permitir continuar
|
| 2934 |
+
# novo treinamento) do modelo testados e aprovados".
|
| 2935 |
+
# Inclui: v6_5_v2_model_states.pt, v6_5_v2_model_states_after_conhecimento.pt
|
| 2936 |
+
# Verifica tamanho ≤ 500MB (HF LFS free tier sem autenticar LFS).
|
| 2937 |
+
_MAX_HF_LFS_SIZE_BYTES = 500 * 1024 * 1024 # 500MB
|
| 2938 |
+
for pt_candidate in [
|
| 2939 |
+
MODEL_STATES_PATH,
|
| 2940 |
+
BIGRU_ROOT / "v6_5_v2_model_states_after_conhecimento.pt",
|
| 2941 |
+
]:
|
| 2942 |
+
if pt_candidate.exists():
|
| 2943 |
+
pt_size = pt_candidate.stat().st_size
|
| 2944 |
+
if pt_size <= _MAX_HF_LFS_SIZE_BYTES:
|
| 2945 |
+
all_files.append(pt_candidate)
|
| 2946 |
+
logger.info(f"[V6.5-V2] Incluindo estado do modelo no upload HF: {pt_candidate.name} ({pt_size/1e6:.1f}MB)")
|
| 2947 |
+
else:
|
| 2948 |
+
logger.warning(f"[V6.5-V2] Estado muito grande para HF LFS (>{_MAX_HF_LFS_SIZE_BYTES/1e6:.0f}MB): {pt_candidate.name} ({pt_size/1e6:.1f}MB) — será mantido localmente apenas")
|
| 2949 |
+
# (c) Relatórios JSON
|
| 2950 |
+
for rpt in [REPORT_PATH, V2_PHASES_EVAL_PATH, PREDICT_FIX_EVAL_PATH,
|
| 2951 |
+
ATTENTION_EVAL_PATH, USER_QUESTIONS_PATH]:
|
| 2952 |
+
if rpt.exists():
|
| 2953 |
+
all_files.append(rpt)
|
| 2954 |
+
# Dedup
|
| 2955 |
+
seen = set()
|
| 2956 |
+
deduped_files: List[Path] = []
|
| 2957 |
+
for p in all_files:
|
| 2958 |
+
if str(p) not in seen:
|
| 2959 |
+
seen.add(str(p))
|
| 2960 |
+
deduped_files.append(p)
|
| 2961 |
+
all_files = deduped_files
|
| 2962 |
|
| 2963 |
# Apaga HF_TOKEN dos scripts ANTES do upload (não enviar tokens)
|
| 2964 |
scrub_result = scrub_hf_token_from_scripts(BIGRU_ROOT / "scripts")
|
|
|
|
| 3007 |
final_storage_cleanup = aggressive_storage_cleanup()
|
| 3008 |
aggressive_memory_cleanup()
|
| 3009 |
|
| 3010 |
+
# V6.5-V2-auto-conscience-v2 — Limpeza de .pt files: SÓ remove se o upload
|
| 3011 |
+
# HF foi bem-sucedido. User requirement (latest):
|
| 3012 |
+
# "estado (e treinamento para permitir continuar novo treinamento)"
|
| 3013 |
+
# Se HF upload falhou, MANTÉM os .pt locais para o usuário poder recuperá-los.
|
| 3014 |
+
# Se HF upload OK, remove apenas os .pt do diretório do projeto (exceto
|
| 3015 |
+
# download/, que é a área de entrega para o usuário).
|
| 3016 |
pt_files_removed: List[str] = []
|
| 3017 |
+
if hf_upload_result.get("uploaded", False):
|
| 3018 |
+
logger.info("[V6.5-V2-auto-conscience-v2] HF upload OK — removendo .pt locais do projeto (estado já está no HF)...")
|
| 3019 |
+
for pt_file in BIGRU_ROOT.rglob("*.pt"):
|
| 3020 |
+
try:
|
| 3021 |
+
rel_path = str(pt_file.relative_to(PROJECT_ROOT))
|
| 3022 |
+
pt_file.unlink()
|
| 3023 |
+
pt_files_removed.append(rel_path)
|
| 3024 |
+
logger.info(f" removed .pt: {rel_path}")
|
| 3025 |
+
except Exception as e:
|
| 3026 |
+
logger.warning(f" failed to remove {pt_file}: {e}")
|
| 3027 |
+
logger.info(
|
| 3028 |
+
f"[V6.5-V2-auto-conscience-v2] {len(pt_files_removed)} .pt files removed from project."
|
| 3029 |
+
)
|
| 3030 |
+
else:
|
| 3031 |
+
logger.warning(
|
| 3032 |
+
"[V6.5-V2-auto-conscience-v2] HF upload falhou — MANTENDO .pt locais "
|
| 3033 |
+
"para preservar estado do modelo. User pode recuperar manualmente."
|
| 3034 |
+
)
|
| 3035 |
+
# Copia .pt para download/ para que o usuário tenha acesso
|
| 3036 |
+
for pt_file in BIGRU_ROOT.rglob("*.pt"):
|
| 3037 |
+
try:
|
| 3038 |
+
dst = DOWNLOAD_DIR / pt_file.name
|
| 3039 |
+
shutil.copy2(pt_file, dst)
|
| 3040 |
+
logger.info(f" .pt copiado para download/: {pt_file.name}")
|
| 3041 |
+
except Exception as e:
|
| 3042 |
+
logger.warning(f" failed to copy {pt_file}: {e}")
|
| 3043 |
|
| 3044 |
# Atualiza worklog.md (sem referenciar .pt files)
|
| 3045 |
try:
|
v6_5_v2_attention_eval.json
CHANGED
|
@@ -3,18 +3,18 @@
|
|
| 3 |
"user_requirement": "verificar se o mecanismo de atenção está ativo e acessado logicamente funcional",
|
| 4 |
"metrics": {
|
| 5 |
"active": true,
|
| 6 |
-
"n_calls":
|
| 7 |
"n_errors": 0,
|
| 8 |
-
"last_norm_in":
|
| 9 |
-
"last_norm_out":
|
| 10 |
"last_attn_activated": true,
|
| 11 |
-
"last_attn_diff_norm": 114.
|
| 12 |
"n_heads": 8,
|
| 13 |
"logic_functional": true
|
| 14 |
},
|
| 15 |
"active": true,
|
| 16 |
"logic_functional": true,
|
| 17 |
-
"n_calls":
|
| 18 |
"n_errors": 0,
|
| 19 |
"n_heads": 8,
|
| 20 |
"assessment": "PASS"
|
|
|
|
| 3 |
"user_requirement": "verificar se o mecanismo de atenção está ativo e acessado logicamente funcional",
|
| 4 |
"metrics": {
|
| 5 |
"active": true,
|
| 6 |
+
"n_calls": 500,
|
| 7 |
"n_errors": 0,
|
| 8 |
+
"last_norm_in": 108.67192077636719,
|
| 9 |
+
"last_norm_out": 112.93913269042969,
|
| 10 |
"last_attn_activated": true,
|
| 11 |
+
"last_attn_diff_norm": 114.95559692382812,
|
| 12 |
"n_heads": 8,
|
| 13 |
"logic_functional": true
|
| 14 |
},
|
| 15 |
"active": true,
|
| 16 |
"logic_functional": true,
|
| 17 |
+
"n_calls": 500,
|
| 18 |
"n_errors": 0,
|
| 19 |
"n_heads": 8,
|
| 20 |
"assessment": "PASS"
|
v6_5_v2_model_states.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d441bc350835134d8bf3d2cba4fb0a7c4f29871779b8b414adddf65fc6b792d5
|
| 3 |
+
size 217669998
|
v6_5_v2_model_states_after_conhecimento.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d13463dc76a501f504b912149ced55883d2e3227c61d359157cb84e00fe331a2
|
| 3 |
+
size 217675097
|
v6_5_v2_phases_eval.json
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
v6_5_v2_predict_fix_eval.json
CHANGED
|
@@ -5,8 +5,8 @@
|
|
| 5 |
"results": [
|
| 6 |
{
|
| 7 |
"query": "o gato dorme na cama",
|
| 8 |
-
"prediction": "
|
| 9 |
-
"probability":
|
| 10 |
"is_gato_hardcoded": false,
|
| 11 |
"is_cachorro_hardcoded": false,
|
| 12 |
"is_registry_label": true,
|
|
@@ -14,8 +14,8 @@
|
|
| 14 |
},
|
| 15 |
{
|
| 16 |
"query": "calcule dois mais dois",
|
| 17 |
-
"prediction": "
|
| 18 |
-
"probability":
|
| 19 |
"is_gato_hardcoded": false,
|
| 20 |
"is_cachorro_hardcoded": false,
|
| 21 |
"is_registry_label": true,
|
|
@@ -23,8 +23,8 @@
|
|
| 23 |
},
|
| 24 |
{
|
| 25 |
"query": "qual é a capital do brasil",
|
| 26 |
-
"prediction": "
|
| 27 |
-
"probability":
|
| 28 |
"is_gato_hardcoded": false,
|
| 29 |
"is_cachorro_hardcoded": false,
|
| 30 |
"is_registry_label": true,
|
|
@@ -32,8 +32,8 @@
|
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"query": "explique o que é uma rede neural",
|
| 35 |
-
"prediction": "
|
| 36 |
-
"probability":
|
| 37 |
"is_gato_hardcoded": false,
|
| 38 |
"is_cachorro_hardcoded": false,
|
| 39 |
"is_registry_label": true,
|
|
@@ -41,8 +41,8 @@
|
|
| 41 |
},
|
| 42 |
{
|
| 43 |
"query": "olá como você está",
|
| 44 |
-
"prediction": "
|
| 45 |
-
"probability":
|
| 46 |
"is_gato_hardcoded": false,
|
| 47 |
"is_cachorro_hardcoded": false,
|
| 48 |
"is_registry_label": true,
|
|
@@ -50,8 +50,8 @@
|
|
| 50 |
},
|
| 51 |
{
|
| 52 |
"query": "traduza hello para portugues",
|
| 53 |
-
"prediction": "
|
| 54 |
-
"probability":
|
| 55 |
"is_gato_hardcoded": false,
|
| 56 |
"is_cachorro_hardcoded": false,
|
| 57 |
"is_registry_label": true,
|
|
@@ -65,70 +65,7 @@
|
|
| 65 |
"unpunctuated",
|
| 66 |
"punctuated"
|
| 67 |
],
|
| 68 |
-
"prediction": "
|
| 69 |
-
"valid_for_dataset": true
|
| 70 |
-
},
|
| 71 |
-
{
|
| 72 |
-
"dataset": "carolina-c4ai/corpus-carolina",
|
| 73 |
-
"expected_labels": [
|
| 74 |
-
"raw_corpus",
|
| 75 |
-
"normalized_text"
|
| 76 |
-
],
|
| 77 |
-
"prediction": "raw_corpus",
|
| 78 |
-
"valid_for_dataset": true
|
| 79 |
-
},
|
| 80 |
-
{
|
| 81 |
-
"dataset": "CEIA-POSITIVO/ultrachat_br_clustred_balanced_v1",
|
| 82 |
-
"expected_labels": [
|
| 83 |
-
"user_turn",
|
| 84 |
-
"assistant_turn"
|
| 85 |
-
],
|
| 86 |
-
"prediction": "user_turn",
|
| 87 |
-
"valid_for_dataset": true
|
| 88 |
-
},
|
| 89 |
-
{
|
| 90 |
-
"dataset": "dominguesm/Canarim-Instruct-PTBR-Dataset",
|
| 91 |
-
"expected_labels": [
|
| 92 |
-
"instruction",
|
| 93 |
-
"response"
|
| 94 |
-
],
|
| 95 |
-
"prediction": "instruction",
|
| 96 |
-
"valid_for_dataset": true
|
| 97 |
-
},
|
| 98 |
-
{
|
| 99 |
-
"dataset": "adalbertojunior/punctuation-ptbr",
|
| 100 |
-
"expected_labels": [
|
| 101 |
-
"unpunctuated",
|
| 102 |
-
"punctuated"
|
| 103 |
-
],
|
| 104 |
-
"prediction": "unpunctuated",
|
| 105 |
-
"valid_for_dataset": true
|
| 106 |
-
},
|
| 107 |
-
{
|
| 108 |
-
"dataset": "iara-project/news-articles-ptbr-dataset",
|
| 109 |
-
"expected_labels": [
|
| 110 |
-
"headline",
|
| 111 |
-
"body"
|
| 112 |
-
],
|
| 113 |
-
"prediction": "headline",
|
| 114 |
-
"valid_for_dataset": true
|
| 115 |
-
},
|
| 116 |
-
{
|
| 117 |
-
"dataset": "manoela/noticias_ptbr",
|
| 118 |
-
"expected_labels": [
|
| 119 |
-
"headline",
|
| 120 |
-
"body"
|
| 121 |
-
],
|
| 122 |
-
"prediction": "headline",
|
| 123 |
-
"valid_for_dataset": true
|
| 124 |
-
},
|
| 125 |
-
{
|
| 126 |
-
"dataset": "BrunoN-Dev/corpus-ptbr-v1",
|
| 127 |
-
"expected_labels": [
|
| 128 |
-
"short_text",
|
| 129 |
-
"long_text"
|
| 130 |
-
],
|
| 131 |
-
"prediction": "short_text",
|
| 132 |
"valid_for_dataset": true
|
| 133 |
}
|
| 134 |
],
|
|
|
|
| 5 |
"results": [
|
| 6 |
{
|
| 7 |
"query": "o gato dorme na cama",
|
| 8 |
+
"prediction": "punctuated",
|
| 9 |
+
"probability": 1.0,
|
| 10 |
"is_gato_hardcoded": false,
|
| 11 |
"is_cachorro_hardcoded": false,
|
| 12 |
"is_registry_label": true,
|
|
|
|
| 14 |
},
|
| 15 |
{
|
| 16 |
"query": "calcule dois mais dois",
|
| 17 |
+
"prediction": "punctuated",
|
| 18 |
+
"probability": 1.0,
|
| 19 |
"is_gato_hardcoded": false,
|
| 20 |
"is_cachorro_hardcoded": false,
|
| 21 |
"is_registry_label": true,
|
|
|
|
| 23 |
},
|
| 24 |
{
|
| 25 |
"query": "qual é a capital do brasil",
|
| 26 |
+
"prediction": "punctuated",
|
| 27 |
+
"probability": 1.0,
|
| 28 |
"is_gato_hardcoded": false,
|
| 29 |
"is_cachorro_hardcoded": false,
|
| 30 |
"is_registry_label": true,
|
|
|
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"query": "explique o que é uma rede neural",
|
| 35 |
+
"prediction": "punctuated",
|
| 36 |
+
"probability": 1.0,
|
| 37 |
"is_gato_hardcoded": false,
|
| 38 |
"is_cachorro_hardcoded": false,
|
| 39 |
"is_registry_label": true,
|
|
|
|
| 41 |
},
|
| 42 |
{
|
| 43 |
"query": "olá como você está",
|
| 44 |
+
"prediction": "punctuated",
|
| 45 |
+
"probability": 1.0,
|
| 46 |
"is_gato_hardcoded": false,
|
| 47 |
"is_cachorro_hardcoded": false,
|
| 48 |
"is_registry_label": true,
|
|
|
|
| 50 |
},
|
| 51 |
{
|
| 52 |
"query": "traduza hello para portugues",
|
| 53 |
+
"prediction": "punctuated",
|
| 54 |
+
"probability": 1.0,
|
| 55 |
"is_gato_hardcoded": false,
|
| 56 |
"is_cachorro_hardcoded": false,
|
| 57 |
"is_registry_label": true,
|
|
|
|
| 65 |
"unpunctuated",
|
| 66 |
"punctuated"
|
| 67 |
],
|
| 68 |
+
"prediction": "punctuated",
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
"valid_for_dataset": true
|
| 70 |
}
|
| 71 |
],
|
v6_5_v2_report.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"version": "V6.5-V2",
|
| 3 |
-
"timestamp": "2026-08-09T03:
|
| 4 |
"config": {
|
| 5 |
"som_grid": [
|
| 6 |
6,
|
|
@@ -54,10 +54,10 @@
|
|
| 54 |
"init_done": true
|
| 55 |
},
|
| 56 |
"fp16_benchmark": {
|
| 57 |
-
"best_time_ms":
|
| 58 |
-
"avg_time_ms": 98.
|
| 59 |
-
"best_tflops": 1.
|
| 60 |
-
"avg_tflops": 1.
|
| 61 |
"matrix_size": 4000.0
|
| 62 |
},
|
| 63 |
"v2_verification": {
|
|
@@ -84,24 +84,24 @@
|
|
| 84 |
"all_pass": true
|
| 85 |
},
|
| 86 |
"fase_1_conhecimento_summary": {
|
| 87 |
-
"total_samples":
|
| 88 |
-
"meta_atingida":
|
| 89 |
-
"elapsed_s":
|
| 90 |
-
"storage_critical_stopped":
|
| 91 |
},
|
| 92 |
"fase_2_punicão_summary": {
|
| 93 |
-
"total_samples":
|
| 94 |
-
"meta_atingida":
|
| 95 |
-
"elapsed_s":
|
| 96 |
-
"punishment_events":
|
| 97 |
-
"hypotheses_trainings":
|
| 98 |
-
"delta_applications":
|
| 99 |
-
"skipped":
|
| 100 |
},
|
| 101 |
"attention_eval_summary": {
|
| 102 |
"active": true,
|
| 103 |
"logic_functional": true,
|
| 104 |
-
"n_calls":
|
| 105 |
"assessment": "PASS"
|
| 106 |
},
|
| 107 |
"predict_fix_summary": {
|
|
@@ -117,53 +117,52 @@
|
|
| 117 |
"n_with_think": 0,
|
| 118 |
"answer_rate": 1.0,
|
| 119 |
"think_rate": 0.0,
|
| 120 |
-
"avg_latency_ms": 4.
|
| 121 |
"avg_reasoning_length": 69.0
|
| 122 |
},
|
| 123 |
"model_states_saved_to": "/home/z/my-project/BiGRU_T_version/v6_5_v2_model_states.pt",
|
| 124 |
"save_info": {
|
| 125 |
"saved": true,
|
| 126 |
"path": "/home/z/my-project/BiGRU_T_version/v6_5_v2_model_states.pt",
|
| 127 |
-
"size_mb":
|
| 128 |
-
"size_gb": 0.
|
| 129 |
"size_status": "OK",
|
| 130 |
"size_within_1gb_limit": true,
|
| 131 |
"reason": "end_of_training_v2",
|
| 132 |
-
"step":
|
| 133 |
-
"total_samples":
|
| 134 |
"n_tensors": 7,
|
| 135 |
"n_buffer_tail": 64
|
| 136 |
},
|
| 137 |
"final_v2_metrics": {
|
| 138 |
"version": "V2-dynamic",
|
| 139 |
-
"n_hypotheses":
|
| 140 |
-
"n_hypotheses_active":
|
| 141 |
"max_n_hypotheses": 32,
|
| 142 |
-
"n_trials":
|
| 143 |
-
"hyp_train_steps":
|
| 144 |
"hyp_lr": 0.0001,
|
| 145 |
-
"delta_scale": 0.
|
| 146 |
"n_generators": 32,
|
| 147 |
"punishment_count": 0,
|
| 148 |
"success_count": 0,
|
| 149 |
-
"training_ready":
|
| 150 |
-
"classifier_trained":
|
| 151 |
-
"ewc_reference_set":
|
| 152 |
"buffer_size": 128,
|
| 153 |
-
"total_hyp_steps_executed":
|
| 154 |
-
"n_train_hyp_calls":
|
| 155 |
"dynamic_adaptation": {
|
| 156 |
-
"loss_history_len":
|
| 157 |
"loss_stats": {
|
| 158 |
-
"slope":
|
| 159 |
-
"volatility": 0.
|
| 160 |
-
"mean": 0.
|
| 161 |
-
"
|
| 162 |
-
"n": 8
|
| 163 |
},
|
| 164 |
-
"punishment_rate":
|
| 165 |
-
"punishment_window_size":
|
| 166 |
-
"n_adaptations":
|
| 167 |
"limits": {
|
| 168 |
"min_n_hypotheses": 4,
|
| 169 |
"max_n_hypotheses": 32,
|
|
@@ -172,128 +171,7 @@
|
|
| 172 |
"min_hyp_train_steps": 10,
|
| 173 |
"max_hyp_train_steps": 80
|
| 174 |
},
|
| 175 |
-
"last_5_adaptations": [
|
| 176 |
-
{
|
| 177 |
-
"trigger": "punishment",
|
| 178 |
-
"step": 115,
|
| 179 |
-
"before": {
|
| 180 |
-
"n_hypotheses": 4,
|
| 181 |
-
"n_trials": 6,
|
| 182 |
-
"hyp_train_steps": 80
|
| 183 |
-
},
|
| 184 |
-
"after": {
|
| 185 |
-
"n_hypotheses": 4,
|
| 186 |
-
"n_trials": 6,
|
| 187 |
-
"hyp_train_steps": 80
|
| 188 |
-
},
|
| 189 |
-
"adapted": false,
|
| 190 |
-
"rules_fired": [],
|
| 191 |
-
"loss_stats": {
|
| 192 |
-
"slope": -0.00037475568907601494,
|
| 193 |
-
"volatility": 0.0033018073049031742,
|
| 194 |
-
"mean": 0.6912502944469452,
|
| 195 |
-
"std": 0.0022823752717213938,
|
| 196 |
-
"n": 8
|
| 197 |
-
},
|
| 198 |
-
"punishment_rate": 1.0
|
| 199 |
-
},
|
| 200 |
-
{
|
| 201 |
-
"trigger": "auto",
|
| 202 |
-
"step": 116,
|
| 203 |
-
"before": {
|
| 204 |
-
"n_hypotheses": 4,
|
| 205 |
-
"n_trials": 6,
|
| 206 |
-
"hyp_train_steps": 80
|
| 207 |
-
},
|
| 208 |
-
"after": {
|
| 209 |
-
"n_hypotheses": 4,
|
| 210 |
-
"n_trials": 6,
|
| 211 |
-
"hyp_train_steps": 80
|
| 212 |
-
},
|
| 213 |
-
"adapted": false,
|
| 214 |
-
"rules_fired": [],
|
| 215 |
-
"loss_stats": {
|
| 216 |
-
"slope": -0.0005504623765037174,
|
| 217 |
-
"volatility": 0.003395094498254662,
|
| 218 |
-
"mean": 0.6910898759961128,
|
| 219 |
-
"std": 0.002346315435793899,
|
| 220 |
-
"n": 8
|
| 221 |
-
},
|
| 222 |
-
"punishment_rate": 1.0
|
| 223 |
-
},
|
| 224 |
-
{
|
| 225 |
-
"trigger": "punishment",
|
| 226 |
-
"step": 117,
|
| 227 |
-
"before": {
|
| 228 |
-
"n_hypotheses": 4,
|
| 229 |
-
"n_trials": 6,
|
| 230 |
-
"hyp_train_steps": 80
|
| 231 |
-
},
|
| 232 |
-
"after": {
|
| 233 |
-
"n_hypotheses": 4,
|
| 234 |
-
"n_trials": 6,
|
| 235 |
-
"hyp_train_steps": 80
|
| 236 |
-
},
|
| 237 |
-
"adapted": false,
|
| 238 |
-
"rules_fired": [],
|
| 239 |
-
"loss_stats": {
|
| 240 |
-
"slope": -0.0005504623765037174,
|
| 241 |
-
"volatility": 0.003395094498254662,
|
| 242 |
-
"mean": 0.6910898759961128,
|
| 243 |
-
"std": 0.002346315435793899,
|
| 244 |
-
"n": 8
|
| 245 |
-
},
|
| 246 |
-
"punishment_rate": 1.0
|
| 247 |
-
},
|
| 248 |
-
{
|
| 249 |
-
"trigger": "auto",
|
| 250 |
-
"step": 118,
|
| 251 |
-
"before": {
|
| 252 |
-
"n_hypotheses": 4,
|
| 253 |
-
"n_trials": 6,
|
| 254 |
-
"hyp_train_steps": 80
|
| 255 |
-
},
|
| 256 |
-
"after": {
|
| 257 |
-
"n_hypotheses": 4,
|
| 258 |
-
"n_trials": 6,
|
| 259 |
-
"hyp_train_steps": 80
|
| 260 |
-
},
|
| 261 |
-
"adapted": false,
|
| 262 |
-
"rules_fired": [],
|
| 263 |
-
"loss_stats": {
|
| 264 |
-
"slope": -0.00034456806523459297,
|
| 265 |
-
"volatility": 0.0024739346406965515,
|
| 266 |
-
"mean": 0.6898908242583275,
|
| 267 |
-
"std": 0.0017067448084313731,
|
| 268 |
-
"n": 8
|
| 269 |
-
},
|
| 270 |
-
"punishment_rate": 1.0
|
| 271 |
-
},
|
| 272 |
-
{
|
| 273 |
-
"trigger": "punishment",
|
| 274 |
-
"step": 119,
|
| 275 |
-
"before": {
|
| 276 |
-
"n_hypotheses": 4,
|
| 277 |
-
"n_trials": 6,
|
| 278 |
-
"hyp_train_steps": 80
|
| 279 |
-
},
|
| 280 |
-
"after": {
|
| 281 |
-
"n_hypotheses": 4,
|
| 282 |
-
"n_trials": 6,
|
| 283 |
-
"hyp_train_steps": 80
|
| 284 |
-
},
|
| 285 |
-
"adapted": false,
|
| 286 |
-
"rules_fired": [],
|
| 287 |
-
"loss_stats": {
|
| 288 |
-
"slope": -0.00034456806523459297,
|
| 289 |
-
"volatility": 0.0024739346406965515,
|
| 290 |
-
"mean": 0.6898908242583275,
|
| 291 |
-
"std": 0.0017067448084313731,
|
| 292 |
-
"n": 8
|
| 293 |
-
},
|
| 294 |
-
"punishment_rate": 1.0
|
| 295 |
-
}
|
| 296 |
-
]
|
| 297 |
}
|
| 298 |
}
|
| 299 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"version": "V6.5-V2",
|
| 3 |
+
"timestamp": "2026-08-09T03:50:50.214658",
|
| 4 |
"config": {
|
| 5 |
"som_grid": [
|
| 6 |
6,
|
|
|
|
| 54 |
"init_done": true
|
| 55 |
},
|
| 56 |
"fp16_benchmark": {
|
| 57 |
+
"best_time_ms": 93.99212000062107,
|
| 58 |
+
"avg_time_ms": 98.73566050009686,
|
| 59 |
+
"best_tflops": 1.361816288420287,
|
| 60 |
+
"avg_tflops": 1.2963907807136654,
|
| 61 |
"matrix_size": 4000.0
|
| 62 |
},
|
| 63 |
"v2_verification": {
|
|
|
|
| 84 |
"all_pass": true
|
| 85 |
},
|
| 86 |
"fase_1_conhecimento_summary": {
|
| 87 |
+
"total_samples": 500,
|
| 88 |
+
"meta_atingida": false,
|
| 89 |
+
"elapsed_s": 37.09058332443237,
|
| 90 |
+
"storage_critical_stopped": true
|
| 91 |
},
|
| 92 |
"fase_2_punicão_summary": {
|
| 93 |
+
"total_samples": 0,
|
| 94 |
+
"meta_atingida": false,
|
| 95 |
+
"elapsed_s": 0.0,
|
| 96 |
+
"punishment_events": 0,
|
| 97 |
+
"hypotheses_trainings": 0,
|
| 98 |
+
"delta_applications": 0,
|
| 99 |
+
"skipped": true
|
| 100 |
},
|
| 101 |
"attention_eval_summary": {
|
| 102 |
"active": true,
|
| 103 |
"logic_functional": true,
|
| 104 |
+
"n_calls": 500,
|
| 105 |
"assessment": "PASS"
|
| 106 |
},
|
| 107 |
"predict_fix_summary": {
|
|
|
|
| 117 |
"n_with_think": 0,
|
| 118 |
"answer_rate": 1.0,
|
| 119 |
"think_rate": 0.0,
|
| 120 |
+
"avg_latency_ms": 4.463593165079753,
|
| 121 |
"avg_reasoning_length": 69.0
|
| 122 |
},
|
| 123 |
"model_states_saved_to": "/home/z/my-project/BiGRU_T_version/v6_5_v2_model_states.pt",
|
| 124 |
"save_info": {
|
| 125 |
"saved": true,
|
| 126 |
"path": "/home/z/my-project/BiGRU_T_version/v6_5_v2_model_states.pt",
|
| 127 |
+
"size_mb": 217.669998,
|
| 128 |
+
"size_gb": 0.20272098295390606,
|
| 129 |
"size_status": "OK",
|
| 130 |
"size_within_1gb_limit": true,
|
| 131 |
"reason": "end_of_training_v2",
|
| 132 |
+
"step": 35,
|
| 133 |
+
"total_samples": 500,
|
| 134 |
"n_tensors": 7,
|
| 135 |
"n_buffer_tail": 64
|
| 136 |
},
|
| 137 |
"final_v2_metrics": {
|
| 138 |
"version": "V2-dynamic",
|
| 139 |
+
"n_hypotheses": 16,
|
| 140 |
+
"n_hypotheses_active": 16,
|
| 141 |
"max_n_hypotheses": 32,
|
| 142 |
+
"n_trials": 3,
|
| 143 |
+
"hyp_train_steps": 30,
|
| 144 |
"hyp_lr": 0.0001,
|
| 145 |
+
"delta_scale": 0.009999999776482582,
|
| 146 |
"n_generators": 32,
|
| 147 |
"punishment_count": 0,
|
| 148 |
"success_count": 0,
|
| 149 |
+
"training_ready": true,
|
| 150 |
+
"classifier_trained": false,
|
| 151 |
+
"ewc_reference_set": false,
|
| 152 |
"buffer_size": 128,
|
| 153 |
+
"total_hyp_steps_executed": 0,
|
| 154 |
+
"n_train_hyp_calls": 0,
|
| 155 |
"dynamic_adaptation": {
|
| 156 |
+
"loss_history_len": 0,
|
| 157 |
"loss_stats": {
|
| 158 |
+
"slope": 0.0,
|
| 159 |
+
"volatility": 0.0,
|
| 160 |
+
"mean": 0.0,
|
| 161 |
+
"n": 0
|
|
|
|
| 162 |
},
|
| 163 |
+
"punishment_rate": 0.0,
|
| 164 |
+
"punishment_window_size": 0,
|
| 165 |
+
"n_adaptations": 0,
|
| 166 |
"limits": {
|
| 167 |
"min_n_hypotheses": 4,
|
| 168 |
"max_n_hypotheses": 32,
|
|
|
|
| 171 |
"min_hyp_train_steps": 10,
|
| 172 |
"max_hyp_train_steps": 80
|
| 173 |
},
|
| 174 |
+
"last_5_adaptations": []
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 175 |
}
|
| 176 |
}
|
| 177 |
}
|
v6_5_v2_user_questions.json
CHANGED
|
@@ -18,17 +18,17 @@
|
|
| 18 |
"context_provided": false,
|
| 19 |
"system_prompt_used": false,
|
| 20 |
"few_shot_examples": false,
|
| 21 |
-
"som_prediction": "
|
| 22 |
"reasoning_length": 69,
|
| 23 |
"has_think": false,
|
| 24 |
"has_plan": false,
|
| 25 |
"has_answer": true,
|
| 26 |
"has_decompose": false,
|
| 27 |
"think_preview": "",
|
| 28 |
-
"answer_preview": "ReasoningEngine disabled. SOM prediction:
|
| 29 |
-
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction:
|
| 30 |
"n_tags": 1,
|
| 31 |
-
"latency_ms": 4.
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"query": "Lula reserva valor",
|
|
@@ -36,17 +36,17 @@
|
|
| 36 |
"context_provided": false,
|
| 37 |
"system_prompt_used": false,
|
| 38 |
"few_shot_examples": false,
|
| 39 |
-
"som_prediction": "
|
| 40 |
"reasoning_length": 69,
|
| 41 |
"has_think": false,
|
| 42 |
"has_plan": false,
|
| 43 |
"has_answer": true,
|
| 44 |
"has_decompose": false,
|
| 45 |
"think_preview": "",
|
| 46 |
-
"answer_preview": "ReasoningEngine disabled. SOM prediction:
|
| 47 |
-
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction:
|
| 48 |
"n_tags": 1,
|
| 49 |
-
"latency_ms": 4.
|
| 50 |
},
|
| 51 |
{
|
| 52 |
"query": "Amazonas força-tarefa vítimas",
|
|
@@ -54,17 +54,17 @@
|
|
| 54 |
"context_provided": false,
|
| 55 |
"system_prompt_used": false,
|
| 56 |
"few_shot_examples": false,
|
| 57 |
-
"som_prediction": "
|
| 58 |
"reasoning_length": 69,
|
| 59 |
"has_think": false,
|
| 60 |
"has_plan": false,
|
| 61 |
"has_answer": true,
|
| 62 |
"has_decompose": false,
|
| 63 |
"think_preview": "",
|
| 64 |
-
"answer_preview": "ReasoningEngine disabled. SOM prediction:
|
| 65 |
-
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction:
|
| 66 |
"n_tags": 1,
|
| 67 |
-
"latency_ms": 4.
|
| 68 |
}
|
| 69 |
],
|
| 70 |
"summary": {
|
|
@@ -72,7 +72,7 @@
|
|
| 72 |
"n_with_think": 0,
|
| 73 |
"answer_rate": 1.0,
|
| 74 |
"think_rate": 0.0,
|
| 75 |
-
"avg_latency_ms": 4.
|
| 76 |
"avg_reasoning_length": 69.0
|
| 77 |
},
|
| 78 |
"quality_assessment": {
|
|
|
|
| 18 |
"context_provided": false,
|
| 19 |
"system_prompt_used": false,
|
| 20 |
"few_shot_examples": false,
|
| 21 |
+
"som_prediction": "punctuated",
|
| 22 |
"reasoning_length": 69,
|
| 23 |
"has_think": false,
|
| 24 |
"has_plan": false,
|
| 25 |
"has_answer": true,
|
| 26 |
"has_decompose": false,
|
| 27 |
"think_preview": "",
|
| 28 |
+
"answer_preview": "ReasoningEngine disabled. SOM prediction: punctuated",
|
| 29 |
+
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction: punctuated</answer>",
|
| 30 |
"n_tags": 1,
|
| 31 |
+
"latency_ms": 4.607677459716797
|
| 32 |
},
|
| 33 |
{
|
| 34 |
"query": "Lula reserva valor",
|
|
|
|
| 36 |
"context_provided": false,
|
| 37 |
"system_prompt_used": false,
|
| 38 |
"few_shot_examples": false,
|
| 39 |
+
"som_prediction": "punctuated",
|
| 40 |
"reasoning_length": 69,
|
| 41 |
"has_think": false,
|
| 42 |
"has_plan": false,
|
| 43 |
"has_answer": true,
|
| 44 |
"has_decompose": false,
|
| 45 |
"think_preview": "",
|
| 46 |
+
"answer_preview": "ReasoningEngine disabled. SOM prediction: punctuated",
|
| 47 |
+
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction: punctuated</answer>",
|
| 48 |
"n_tags": 1,
|
| 49 |
+
"latency_ms": 4.328012466430664
|
| 50 |
},
|
| 51 |
{
|
| 52 |
"query": "Amazonas força-tarefa vítimas",
|
|
|
|
| 54 |
"context_provided": false,
|
| 55 |
"system_prompt_used": false,
|
| 56 |
"few_shot_examples": false,
|
| 57 |
+
"som_prediction": "punctuated",
|
| 58 |
"reasoning_length": 69,
|
| 59 |
"has_think": false,
|
| 60 |
"has_plan": false,
|
| 61 |
"has_answer": true,
|
| 62 |
"has_decompose": false,
|
| 63 |
"think_preview": "",
|
| 64 |
+
"answer_preview": "ReasoningEngine disabled. SOM prediction: punctuated",
|
| 65 |
+
"raw_response_preview": "<answer>ReasoningEngine disabled. SOM prediction: punctuated</answer>",
|
| 66 |
"n_tags": 1,
|
| 67 |
+
"latency_ms": 4.455089569091797
|
| 68 |
}
|
| 69 |
],
|
| 70 |
"summary": {
|
|
|
|
| 72 |
"n_with_think": 0,
|
| 73 |
"answer_rate": 1.0,
|
| 74 |
"think_rate": 0.0,
|
| 75 |
+
"avg_latency_ms": 4.463593165079753,
|
| 76 |
"avg_reasoning_length": 69.0
|
| 77 |
},
|
| 78 |
"quality_assessment": {
|