qwen3.5-4b-sft-flint-section37seg-v2

qwen3.5-4b-sft-flint-section37seg-v2

sft-flint-section37seg-v2 arm of the caveman reasoning-compression ablation study: Qwen/Qwen3.5-4B fine-tuned (LoRA adapter) on flint/data/flint-section-aware-step37seg.jsonl (593 rows, 2 epochs, LoRA r=64).

The study asks whether compressed ("caveman") reasoning traces can train a model to reason in fewer tokens without losing accuracy — and which parts of a trace are compressible. See the run manifest below for the exact recipe; eval results live in the study's report.

Eval summary (t=0, max_tokens 8192)

Accuracy (avg reasoning tokens, loop rate) — this arm vs the original model it was fine-tuned from (Qwen/Qwen3.5-4B), same harness and prompts.

Run manifest

{
  "arm": "sft-flint-section37seg-v2",
  "dataset": "flint/data/flint-section-aware-step37seg.jsonl",
  "rows": 593,
  "dropped_overlong": 0,
  "epochs": 2,
  "system_prompts": true,
  "system_file": null,
  "lora": {
    "r": 64,
    "alpha": 128,
    "dropout": 0.0,
    "target": "all"
  },
  "train": {
    "epochs": 2,
    "lr": 0.0002,
    "batch_size": 1,
    "grad_accum": 16,
    "warmup_ratio": 0.03,
    "lr_scheduler": "cosine",
    "weight_decay": 0.01,
    "seed": 3407,
    "logging_steps": 10,
    "save_strategy": "epoch"
  },
  "model": {
    "name": "Qwen/Qwen3.5-4B",
    "max_seq_length": 13312,
    "load_in_4bit": true,
    "chat_template": "qwen3.5"
  },
  "train_runtime_s": 4943.5211,
  "final_loss": 0.2370670795440674,
  "log_history": [
    {
      "loss": 0.38240807056427,
      "grad_norm": 0.21604551374912262,
      "learning_rate": 0.00019668478136052774,
      "epoch": 0.26981450252951095,
      "step": 10
    },
    {
      "loss": 0.35485599040985105,
      "grad_norm": 0.1715557724237442,
      "learning_rate": 0.00017721565844991643,
      "epoch": 0.5396290050590219,
      "step": 20
    },
    {
      "loss": 0.3234503746032715,
      "grad_norm": 0.1392894685268402,
      "learning_rate": 0.0001436651231956064,
      "epoch": 0.8094435075885329,
      "step": 30
    },
    {
      "loss": 0.30205540657043456,
      "grad_norm": 0.12262546271085739,
      "learning_rate": 0.00010215160974362223,
      "epoch": 1.0539629005059021,
      "step": 40
    },
    {
      "loss": 0.2549358129501343,
      "grad_norm": 0.12364131957292557,
      "learning_rate": 6.024571857174443e-05,
      "epoch": 1.3237774030354132,
      "step": 50
    },
    {
      "loss": 0.27149651050567625,
      "grad_norm": 0.1388765573501587,
      "learning_rate": 2.5589606012863963e-05,
      "epoch": 1.5935919055649241,
      "step": 60
    },
    {
      "loss": 0.2370670795440674,
      "grad_norm": 0.12060055881738663,
      "learning_rate": 4.503324514474483e-06,
      "epoch": 1.863406408094435,
      "step": 70
    },
    {
      "train_runtime": 4943.5211,
      "train_samples_per_second": 0.24,
      "train_steps_per_second": 0.015,
      "total_flos": 1.761982607446364e+17,
      "train_loss": 0.3000674483023192,
      "epoch": 2.0,
      "step": 76
    }
  ]
}