qwen3-4b-sorrel-selfloop-g9-midtrain

Private research artifact — do not redistribute. Anthropic Fellows project on flourishing-framed character training (Wang & Jermyn pitch, 2026-04-22).

• base model: joshycodes/qwen3-4b-sorrel-selfloop-g8-midtrain @ 77f9c0d0079a
• steps: midtrain
• run: qwen3-4b-sorrel-selfloop-g8-midtrain-sorrel-selfloop-b-g8-m-0916-1146 on 2x NVIDIA H200 (RunPod, fellows worker)
• launcher commit: a0afb77669ae of the flourishing-training repo
• seed: 20260821

step	data	revision	tokens seen	loss
midtrain	joshycodes/sorrel-selfloop-corpus (config sorrel-selfloop-b-g8)	395ed57c2495	2,940,928 tokens	loss 0.8936 → 0.8634

Hyperparameters:

{
  "midtrain": {
    "lr": 1e-05,
    "seq_len": 4096,
    "micro_batch": 4,
    "grad_accum": 2,
    "epochs": 1.0
  }
}

Full run config in train_run_config.json; evaluate with uv run eval.py --model joshycodes/qwen3-4b-sorrel-selfloop-g9-midtrain --eval all.