SkillScreen-2B-Markdown / training_config.json
JerroldH's picture
Release v0.1.0: full-data deployment model, inference entry point and OOF thresholds
b60475f verified
Raw History Blame Contribute Delete
1.27 kB
{
"train_records": 13514,
"max_length": 8192,
"epochs": 1,
"learning_rate": 0.0002,
"effective_batch": 16,
"base_model": {
"repo_id": "AlexWortega/openjev",
"subfolder": "qwen3.5-2b-nli-v5",
"revision": "a20448012c213128955ca0c693e7c943865cab77"
},
"trainable_parameters": 2193408,
"checkpoint_policy": "Fixed end of epoch",
"runtime": {
"torch": "2.8.0+cu126",
"transformers": "5.17.0",
"peft": "0.21.2",
"flash-linear-attention": "0.3.2",
"triton": "3.4.0"
},
"adapter_sha256": "da4424a0b6091820866e767769d82bca1c717bd7f488b484f766353d6ca6f48b",
"seed": 42,
"training_scope": "all_data",
"dataset": {
"repo_id": "SkillScreen/SkillScreen-Data",
"revision": "d59aabbd7e0e8d95b464a2d335f36e948577b832"
},
"lora": {
"r": 8,
"alpha": 16,
"dropout": 0,
"target_modules": [
"q_proj",
"v_proj",
"in_proj_qkv",
"out_proj"
],
"classification_head_trainable": true
},
"weight_decay": 0.01,
"warmup_fraction": 0.05,
"gradient_clip_norm": 1.0,
"precision": "bfloat16",
"folds": 3,
"fold_grouping": "Normalized Markdown content groups; stratified by source and label.",
"selection": "Fixed final checkpoint after one epoch; no early stopping."
}