paper-method-grid-multi-task-z64-qwen25_20260501

Status: complete integrated follow-up. The six tracked collections contain 39 expected cells: 39 have final metrics and 0 rows remain without final metrics.

This report analyzes a focused multi-task pretrained-from-entangled paper-method grid and its TOK-CC/dist-weight follow-ups. It is modeled after paper-selected reconstruction-stack losses, but narrows to the compact Qwen2.5 multi-task setting.

Links

• design
• results
• analysis
• tokcc_dist_weight_sweep
• dynamics_qwen25_05b
• dynamics_qwen25_15b
• exploration-efficiency

Artifacts

• Aggregate output directory: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501
• Aggregate script: path:experiments/synthetic_sequences/disentangled/analysis/run_qwen25_paper_method_grid_multi_task_z64_20260501.py
• Repo-side report: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/report.md
• Brief: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/brief.yaml
• Findings JSON: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/findings.json
• Notebook: path:experiments/synthetic_sequences/disentangled/analysis/aggregates/synthetic-pretrained-disentangled-from-entangled-v2-paper-method-grid-multi-task-z64-qwen25_20260501/notebook.ipynb
• Qwen2.5 0.5B dynamics note: dynamics_qwen25_05b
• Qwen2.5 1.5B dynamics note: dynamics_qwen25_15b
• TOK-CC/dist-weight follow-up note: tokcc_dist_weight_sweep
• Qwen2.5 0.5B slurmkit collection: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/paper-method-grid-multi-task-z64-qwen25-05b.yaml
• Qwen2.5 1.5B slurmkit collection: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/paper-method-grid-multi-task-z64-qwen25-15b.yaml
• Qwen2.5 0.5B z8 tokcc/dist sweep: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/tokcc-dist-weight-sweep-multi-task-z8-qwen25-05b.yaml
• Qwen2.5 0.5B z64 tokcc/dist sweep: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/tokcc-dist-weight-sweep-multi-task-z64-qwen25-05b.yaml
• Qwen2.5 ratio/tokinv follow-up: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/ratio-tokinv-followup-multi-task-z64-qwen25.yaml
• Qwen2.5 raw beta baseline follow-up: path:.slurmkit/collections/synthetic-pretrained/disentanglement/from-entangled-v2/raw-beta-baseline-followup-multi-task-z64-qwen25.yaml
• Analogical steering post-hoc collection: path:.slurmkit/collections/synthetic-pretrained/analogical-steering/from-entangled-v2/z64-qwen25-current-sweeps.yaml
• Exploration-efficiency post-hoc collection: path:.slurmkit/collections/synthetic-pretrained/exploration-efficiency/from-entangled-v2/z64-qwen25-global-token-kernel-sweep.yaml

Context

• Prior paper-selected loss report: Apr27 paper-selected results
• Prior qwen-0.5B reconstruction-stack report: qwen 0.5B reconstruction-stack results

Runs

• Qwen2.5 0.5B collection: synthetic-pretrained/disentanglement/from-entangled-v2/paper-method-grid-multi-task-z64-qwen25-05b
• Qwen2.5 1.5B collection: synthetic-pretrained/disentanglement/from-entangled-v2/paper-method-grid-multi-task-z64-qwen25-15b
• Qwen2.5 0.5B z8 tokcc/dist sweep: synthetic-pretrained/disentanglement/from-entangled-v2/tokcc-dist-weight-sweep-multi-task-z8-qwen25-05b
• Qwen2.5 0.5B z64 tokcc/dist sweep: synthetic-pretrained/disentanglement/from-entangled-v2/tokcc-dist-weight-sweep-multi-task-z64-qwen25-05b
• Qwen2.5 ratio/tokinv follow-up: synthetic-pretrained/disentanglement/from-entangled-v2/ratio-tokinv-followup-multi-task-z64-qwen25
• Qwen2.5 raw beta baseline follow-up: synthetic-pretrained/disentanglement/from-entangled-v2/raw-beta-baseline-followup-multi-task-z64-qwen25
• Integrated expected cells: 39.
• Analysis snapshot on 2026-05-03: 39 final-test cells, 39 cells with validation streams, 0 rows without final metrics, and 33 analogical-steering summaries.