{ "name": "multi_loop", "seed": 2, "created_utc": "2026-10-06T19:43:26Z", "git": { "commit": "230f87a1d258d5b85c5832e144844b2b54b19f49", "dirty": false }, "config": { "name": "multi_loop", "variant": "configs/variants/v2_loop_mid_r2.yaml", "seed": 2, "data": { "prompt_suite": "evals/suites/dev.yaml", "exclude_suites": [ "evals/suites/smoke.yaml", "evals/suites/dev.yaml" ], "tasks": null, "difficulties": [ 1, 2, 3 ], "n_per_difficulty": 100, "n_val_per_difficulty": 4, "formats": [ "choice", "generate", "generate_cot" ], "source": "multi", "sources": [ { "kind": "dev", "prompt_suite": "evals/suites/dev.yaml", "exclude_suites": [ "evals/suites/smoke.yaml", "evals/suites/dev.yaml" ], "tasks": null, "difficulties": [ 1, 2, 3 ], "n_per_difficulty": 100, "n_val_per_difficulty": 4, "formats": [ "choice", "generate", "generate_cot" ] }, { "kind": "rmp", "families": [ "pointer_chase", "gated_pointer_chase", "masked_pointer_chase", "reachability", "sudoku4" ], "n_per_family": 480, "n_val_per_family": 20, "index_offset": 1000 }, { "kind": "s2", "upsample": 3 }, { "kind": "s7", "upsample": 4 } ] }, "lora": { "r": 16, "alpha": 32, "dropout": 0.05, "target_modules": [ "q_proj", "k_proj", "v_proj", "o_proj", "in_proj_qkv", "in_proj_z", "out_proj", "gate_proj", "up_proj", "down_proj" ], "layers_to_transform": null }, "loss": { "choice": "ce" }, "optim": { "lr": 0.0001, "warmup_steps": 20, "min_lr_ratio": 0.1, "grad_accum": 8, "epochs": 1, "max_grad_norm": 1.0, "weight_decay": 0.0 }, "gradient_checkpointing": true, "eval_every": 100, "save_every": 25, "max_steps": null }, "config_sha256": "ac4d2dba5c46a02d13f8575e6b4403b5e7892679f7a038500c3c747dcacd236e", "variant": { "name": "v2_loop_mid_r2", "driver": "schedule", "loop": { "start": 12, "end": 16, "n_iters": 2 }, "halting": null, "adapter": null }, "model": { "id": "Qwen/Qwen3.5-4B-Base", "revision_requested": "1001bb4d826a52d1f399e183466143f4da7b741b", "revision_resolved": null, "class": "Qwen3_5ForCausalLM", "dtype": "bfloat16", "load_in_4bit": false, "cuda_memory_fraction": 0.7, "unexpected_keys_dropped": 0, "n_params_text_stack": 4236225536, "num_layers": 32, "layer_types": [ "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention", "linear_attention", "linear_attention", "linear_attention", "full_attention" ], "trainable_params": 30474240 }, "data": { "source": "multi (SPEC-U1)", "parts": { "dev": { "n_train": 2400, "n_val": 96, "upsample": 1, "formats": { "choice": 800, "generate": 800, "generate_cot": 800 } }, "rmp": { "n_train": 2400, "n_val": 100, "upsample": 1, "families": [ "pointer_chase", "gated_pointer_chase", "masked_pointer_chase", "reachability", "sudoku4" ], "rows_sha256": "55b96105bdbdf6dd12216026a984b4aa9c303ca3d9cbb5626500365db92c786c", "formats": { "choice": 2400 } }, "s2": { "n_train": 702, "n_val": 50, "upsample": 3, "n_rows": 234, "rows_sha256": "1f237e0a932e1c898faec8edb8cfc43c7376db7d6fc9371e68a07a392794d719" }, "s7": { "n_train": 216, "n_val": 0, "upsample": 4, "n_rows": 54, "rows_sha256": "f49e45f0efb5fd5f548b4d2b38db7add192b8c8e0c33c7e775f25b0f475c8f4d" } }, "n_train": 5718, "n_val": 246, "formats": { "choice": 4118, "generate": 800, "generate_cot": 800 }, "max_train_tokens": 1078 }, "steps": 714, "final": { "step": 714, "lr": 1.0000461065451529e-05, "grad_norm": 16.27705192565918, "seconds": 9.519256500003394, "loss": 0.616852872571485, "loss_choice": 0.8198426299682069, "loss_generate_cot": 0.007883600381319411, "train_choice_accuracy": 0.6666666666666666, "peak_vram_mb": 9351.47021484375, "val_ce_choice": 0.3707982663626629, "val_ce_generate": 0.19374142463743738, "val_ce_generate_cot": 0.03607354498888071, "val_choice_accuracy": 0.8351648351648352 }, "hardware": { "platform": "Windows-11-10.0.26200-SP0", "python": "3.12.10", "cpu": "AMD64 Family 26 Model 36 Stepping 0, AuthenticAMD", "cpu_count": 20, "packages": { "torch": "2.11.0+cu128", "transformers": "5.17.0", "peft": "0.21.1", "accelerate": "1.15.0", "bitsandbytes": "0.50.2", "safetensors": "0.8.0", "tokenizers": "0.23.2" }, "gpu": { "name": "NVIDIA GeForce RTX 5080 Laptop GPU", "memory_total_mib": 16303, "memory_used_mib": 11152, "utilization_pct": 65, "driver_version": "572.76" }, "ram_total_gb": 31.1, "ram_available_gb": 13.2, "torch_cuda_available": true, "torch_cuda_version": "12.8" }, "wall_time_s": 6953.236694200001, "adapter": "adapter" }