Qwen25 Multi-Task Dynamics: Qwen2.5 1.5B
Qwen25 Multi-Task Dynamics: Qwen2.5 1.5B
Back to hub · results · analysis
This is an observational dynamics deep-dive for qwen25_15b. This page uses final-test metrics for ranking because final metrics are available.
Selection
Rows are ranked by within-task matched router score, computed as within-task router probe times strategy-match rate. The selected row is dist_plus_tokcc_a0p25_w0p05_tokcc0p95_b0p01_lin100 with score 0.9167 from final_test.
| selected | collection_label | continuous_latent_dim | loss_setting | analysis_status | metric_source | matched_router_score | final_answer_accuracy | within_task_router_probe | all_task_router_probe | strategy_match_rate | posterior_probe | reconstruction_loss | kl_prior_loss | latest_global_step |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| selected | 1.5B paper grid z64 | 64 | dist_plus_tokcc_a0p25_w0p05_tokcc0p95_b0p01_lin100 | complete_final | final_test | 0.9167 | 0.9941 | 0.9323 | 0.2113 | 0.9832 | 0.9901 | 0.0007 | 2.0967 | 1563 |
| 1.5B paper grid z64 | 64 | tokcc_a0p25_b0p1_lin100 | complete_final | final_test | 0.8845 | 0.9925 | 0.9084 | 0.2125 | 0.9737 | 0.9765 | 0.6243 | 1.4681 | 1563 | |
| ratio/tokinv follow-up z64 | 64 | dist_plus_tokcc_a0p25_w0p75_tokcc0p25_b0p01_lin100 | complete_final | final_test | 0.8823 | 0.9951 | 0.891 | 0.2128 | 0.9902 | 0.9777 | 0.0061 | 1.7611 | 1563 | |
| 1.5B paper grid z64 | 64 | dist_norm_b0p01_lin100 | complete_final | final_test | 0.8086 | 0.9958 | 0.8144 | 0.2138 | 0.9929 | 0.9641 | 0.0039 | 1.8263 | 1563 | |
| ratio/tokinv follow-up z64 | 64 | dist_plus_tokcc_a0p25_w0p5_tokcc0p5_b0p01_lin100 | complete_final | final_test | 0.8049 | 0.9956 | 0.8101 | 0.2069 | 0.9936 | 0.9732 | 0.0136 | 1.2404 | 1563 | |
| ratio/tokinv follow-up z64 | 64 | dist0p25_tokcc0p75_a0p25_b0p01_lin100 | complete_final | final_test | 0.6958 | 0.9987 | 0.6975 | 0.1647 | 0.9976 | 0.9474 | 0.0148 | 2.0736 | 1563 | |
| raw beta baseline z64 | 64 | raw_b0p01_const | complete_final | final_test | 0.6907 | 0.9993 | 0.6912 | 0.1341 | 0.9993 | 0.7274 | 0.032 | 0.0124 | 1563 | |
| 1.5B paper grid z64 | 64 | dist_plus_tokcc_a0p25_w0p05_tokcc0p95_b0p1_lin100 | complete_final | final_test | 0.6872 | 0.9991 | 0.6879 | 0.1349 | 0.9989 | 0.7692 | 0.032 | 0.0068 | 1563 | |
| raw beta baseline z64 | 64 | raw_b1_const | complete_final | final_test | 0.6861 | 0.9993 | 0.6866 | 0.1294 | 0.9993 | 0.6809 | 0.032 | 0.0121 | 1563 | |
| 1.5B paper grid z64 | 64 | tokcc_a0p25_b0p01_lin100 | complete_final | final_test | 0.6851 | 0.9985 | 0.6868 | 0.1367 | 0.9974 | 0.8001 | 0.6386 | 0.0179 | 1563 | |
| raw beta baseline z64 | 64 | raw_b1_lin100 | complete_final | final_test | 0.6835 | 0.9994 | 0.6839 | 0.136 | 0.9994 | 0.6793 | 0.032 | 0.0308 | 1563 | |
| raw beta baseline z64 | 64 | raw_b0p1_const | complete_final | final_test | 0.6737 | 0.9994 | 0.6741 | 0.1309 | 0.9994 | 0.6989 | 0.032 | 0.0223 | 1563 | |
| ratio/tokinv follow-up z64 | 64 | tokinv_a1_eps1em8_b0p01_lin100 | complete_final | final_test | 0.6284 | 0.9937 | 0.6341 | 0.1357 | 0.991 | 0.7993 | 0.572 | 31.6328 | 1563 | |
| 1.5B paper grid z64 | 64 | dist_norm_b0p1_lin100 | complete_final | final_test | 0.5844 | 0.9994 | 0.5847 | 0.1333 | 0.9994 | 0.8543 | 0.032 | 0.0066 | 1563 | |
| 1.5B paper grid z64 | 64 | raw_b0p1_lin100 | complete_final | final_test | 0.5689 | 0.9995 | 0.5692 | 0.1283 | 0.9995 | 0.6794 | 0.032 | 0.0122 | 1563 | |
| 1.5B paper grid z64 | 64 | raw_b0p01_lin100 | complete_final | final_test | 0.5566 | 0.9994 | 0.5569 | 0.1239 | 0.9994 | 0.7418 | 0.032 | 0.0068 | 1563 |
Validation Dynamics
The validation dynamics panel plots final-answer accuracy, matched router score, within-task router probe, all-task router probe, strategy-match rate, posterior-mu probe, reconstruction loss, and KL prior by epoch progress for each available run directory.
Router-Sampled Z-Space
Task-separated view: GIF columns are effective tasks; frame rows are effective tasks.
Source frame stream: path:.runs/synthetic_sequences/disentangled/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25-15b/task-multi_task_src-ent-qwen25-1-5b_adapt-lora_lt-continuous_loss-recon-dist-0p05-tokcc-0p95_beta-0p01_h5b23e54d90e0b933_s314159/continuous_val_router_sampled_z_frames.jsonl
Posterior-Mu Z-Space
Task-separated view: GIF columns are effective tasks; frame rows are effective tasks.
Source frame stream: path:.runs/synthetic_sequences/disentangled/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25-15b/task-multi_task_src-ent-qwen25-1-5b_adapt-lora_lt-continuous_loss-recon-dist-0p05-tokcc-0p95_beta-0p01_h5b23e54d90e0b933_s314159/continuous_val_posterior_mu_frames.jsonl
Run-Level Notes
The selected run directory is path:.runs/synthetic_sequences/disentangled/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25-15b/task-multi_task_src-ent-qwen25-1-5b_adapt-lora_lt-continuous_loss-recon-dist-0p05-tokcc-0p95_beta-0p01_h5b23e54d90e0b933_s314159. The data backing this page are in path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/data/dynamics_qwen25_15b_metrics.csv and path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/data/dynamics_qwen25_15b_eval_history.csv. The generation script is path:experiments/synthetic_sequences/disentangled/analysis/run_qwen25_paper_method_grid_multi_task_z64_20260501.py.
