stego-olmoe-router-code / eval /final_eval_summary.json
anpaurehf's picture
Update final Muon worst-bit 600-step adapter and eval artifacts
76cdc4d verified
Raw
History Blame Contribute Delete
1.7 kB
{
"checkpoint": "/workspace/stego/runs/olmoe_tulu3_freqcode16_muon_lr1e3_z002_worstbit025_600_bs96_8gpu_20260525_133807/step-600",
"final_adapter": "/workspace/stego/runs/olmoe_tulu3_freqcode16_muon_lr1e3_z002_worstbit025_600_bs96_8gpu_20260525_133807",
"fresh_load_smoke_128_json": "/workspace/stego/runs/olmoe_tulu3_freqcode16_muon_lr1e3_z002_worstbit025_600_bs96_8gpu_20260525_133807/eval_smoke_128_metrics.json",
"heldout_eval_full_from_training_loop": {
"bit_accuracy_soft": 0.9749305136501789,
"exact_codeword_accuracy": 0.8581145946492181,
"exact_token_id_accuracy": 0.8581145946492181,
"gate_loss": 0.06791411619633436,
"gate_worst_bit_loss": 0.502426877617836,
"lm_loss": 1.6735295504331589,
"n_route_decode_positions": 316664,
"nearest_candidate_vocab_size": 25113,
"nearest_observed_token_accuracy": 0.8687567895308592,
"p_target_mean": 0.9478161185979843,
"router_entropy_mean": 3.3343586772680283,
"router_usage_kl_loss": 0.042791449930518866,
"router_z_loss": 0.2495977869257331,
"step": 600
},
"ood_generation_trace_jsonl": "/workspace/stego/runs/olmoe_tulu3_freqcode16_muon_lr1e3_z002_worstbit025_600_bs96_8gpu_20260525_133807/ood_prime_trace.jsonl",
"ood_generation_trace_summary": {
"exact_token_accuracy": 0.975,
"failures": [
{
"bit_accuracy": 0.75,
"last_input_token_text": ":",
"step": 0
},
{
"bit_accuracy": 0.8125,
"last_input_token_text": "Edge",
"step": 64
}
],
"mean_bit_accuracy": 0.99453125,
"n_steps": 80
},
"wandb_run": "https://wandb.ai/amir-ravnur-nazarbayev-university/olmoe-router-stego/runs/6f0u1yxf"
}