modrill commited on
Commit
8d58dcc
·
verified ·
1 Parent(s): 85a0cf9

Add provenance/RUN_IDENTITY.json

Browse files
Files changed (1) hide show
  1. provenance/RUN_IDENTITY.json +112 -0
provenance/RUN_IDENTITY.json ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "schema": "RUN_IDENTITY_V1",
3
+ "run_id": "t30b2507-q8b-think-v4-tail151643",
4
+ "corrects": "PAIRED_V4",
5
+ "model": {
6
+ "hf_id": "Qwen/Qwen3-8B-Base",
7
+ "revision": "49e3418fbbbca6ecbdf9608b4d22e5a407081db4",
8
+ "local_path": "/workspace/code-sft-infra/models/qwen3-8b-base",
9
+ "tied_embeddings": false
10
+ },
11
+ "data": {
12
+ "path": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/data/THINK_PAIRED_V4_2EP.parquet",
13
+ "rows": 9430,
14
+ "sha256": "c160713fea76eae124dbbc3b5c619d3164862da83f61d91d3970ede990a12540",
15
+ "census": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/data/TOKEN_CENSUS_V4_Q8B_THINK.json",
16
+ "physical_2ep": "concat(selected_v4_table, selected_v4_table) same order; token columns overlaid from TOKEN_CENSUS_V4; O7B also rewrites renderer_id",
17
+ "source_1ep_path": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/data/THINK_PAIRED_V4.parquet",
18
+ "source_1ep_sha256": "cf37c85413f69339508004611e196e42abbc2bddbd3462c209a1b97e1d820002",
19
+ "source_1ep_rows": 4715
20
+ },
21
+ "seeds": {
22
+ "train": 42,
23
+ "eval": 3407
24
+ },
25
+ "parameters": {
26
+ "epochs": 2,
27
+ "target_assistant_tokens": 64873788,
28
+ "active_tokens_1ep": 32436894,
29
+ "checkpoint_target_tokens": [
30
+ 32441828,
31
+ 48563817,
32
+ 64873788
33
+ ],
34
+ "milestone_updates": [
35
+ 495,
36
+ 741,
37
+ 990
38
+ ],
39
+ "warmup_target_tokens": 3892427,
40
+ "learning_rate": 0.0001,
41
+ "row_lr": 0.0001,
42
+ "optimizer": "adamw",
43
+ "optimizer_betas": [
44
+ 0.9,
45
+ 0.95
46
+ ],
47
+ "optimizer_eps": 1e-08,
48
+ "lora_rank": 64,
49
+ "lora_alpha": 128,
50
+ "lora_dropout": 0.0,
51
+ "lora_target_modules": [
52
+ "q_proj",
53
+ "k_proj",
54
+ "v_proj",
55
+ "o_proj",
56
+ "gate_proj",
57
+ "up_proj",
58
+ "down_proj"
59
+ ],
60
+ "lora_weight_decay": 0.1,
61
+ "row_weight_decay": 0.0,
62
+ "max_grad_norm": 1.0,
63
+ "tokens_per_optimizer_update": 65536,
64
+ "packing": false,
65
+ "truncation": false,
66
+ "repeat_examples": false,
67
+ "precision": "bfloat16",
68
+ "trainable_dtype": "float32",
69
+ "gradient_checkpointing": true,
70
+ "eos_weight": 1.0,
71
+ "scheduler": "cosine_by_assistant_target_token_dose",
72
+ "worker_lr_scale": "rstar_dual_worker_v4._lr_scale",
73
+ "supervised_tail": [
74
+ 151643
75
+ ],
76
+ "trainable_token_indices": {
77
+ "embed_tokens": [
78
+ 151643,
79
+ 151667,
80
+ 151668
81
+ ],
82
+ "lm_head": [
83
+ 151643,
84
+ 151667,
85
+ 151668
86
+ ]
87
+ },
88
+ "untie_tied_embeddings": true,
89
+ "context": 32768,
90
+ "eval": {
91
+ "suite": "DEV256",
92
+ "seed": 3407,
93
+ "temperature": 0.6,
94
+ "top_p": 0.95,
95
+ "top_k": 20,
96
+ "ctx": 32768,
97
+ "stop_ids": [
98
+ 151643,
99
+ 151645
100
+ ],
101
+ "base_reeval_same_mode": true
102
+ }
103
+ },
104
+ "code": {
105
+ "path": "/workspace/tools/rstar_dual_worker_v4.py",
106
+ "commit_or_sha256": "97a298e9ee4f0bb571baa88c7d75285ade2ea2a8a9b5875b35db25b5db9b3842",
107
+ "template": "/workspace/code-sft-runs/t30b2507-q4b-tn-v2-lr1e4-e1-20260907/templates/Q8B-THINK.template_v3.json"
108
+ },
109
+ "started_at": "2026-09-08T07:50:11Z",
110
+ "endpoint": "2EP_ONLY @64873788 assistant tokens; no checkpoint picking",
111
+ "status": "RUNNING"
112
+ }