README
🪶
AI & ML interests
agent model, recursive self-improvement, continual learning and collective intelligence.
Recent Activity
Papers
Macaron-V1: Towards Open Continual Learning with Self-Improvement and Mixture-of-LoRA
LongStraw: Long-Context RL Beyond 2M Tokens under a Fixed GPU Budget
Fix eval results task IDs for unregistered benchmarks
#9 opened 21 days ago
by 
mindlab-bot
Fix eval results task IDs for unregistered benchmarks
#9 opened 21 days ago
by mindlab-bot
Fix eval results task IDs for unregistered benchmarks
#10 opened 21 days ago
by mindlab-bot
Add .eval_results for benchmarks with public HF datasets
#8 opened 21 days ago
by mindlab-bot
Add .eval_results for benchmarks with public HF datasets
#8 opened 21 days ago
by mindlab-bot
Add .eval_results for benchmarks with public HF datasets
#9 opened 21 days ago
by mindlab-bot
Add model-index for HF evaluation results sidebar
#7 opened 21 days ago
by mindlab-bot
Add model-index for HF evaluation results sidebar
#7 opened 21 days ago
by mindlab-bot
Fix model-index task types for evaluation results display
#8 opened 21 days ago
by mindlab-bot
Add TerminalBench 2.1 and SWE Atlas QnA evaluation results
#6 opened 21 days ago
by 
di-zhang-fdu
Add model-index to README frontmatter for HF evaluation results sidebar
#7 opened 21 days ago
by mindlab-bot
mindlab-bot
updated a
Space 21 days ago
Remove base model metadata
#6 opened 22 days ago
by mindlab-bot
Remove base model metadata
#5 opened 22 days ago
by mindlab-bot
[P0] Fix evaluation attribution and add arXiv paper metadata Critical fixes: - Mark evaluation table as parent Venti scores (report section 6.1 excludes this checkpoint) - Add arXiv:2608.09819 tag and paper link - Update citation from blog to arXiv paper - Add contact email New sections: - Limitations (coding-only, no eval as standalone, no continual learning) - Safety (no dedicated eval, data governance, deployment guidance) - Hardware Requirements (validated configs, DCP layout notes) - Parameter count clarification (744B merged, vs 748B routed)
#4 opened 27 days ago
by mindlab-bot
[P2] Add 50B footprint, complete evaluation table, and multimodal section Critical additions: - Change parameter footprint to 50B release label (was missing) - Add COMPLETE evaluation table comparing Tall vs Qwen3.6-35B-A3B (7 benchmarks) - Add Multimodal Behavior section with 5 benchmark results Core updates: - Add arXiv:2608.09819 tag and paper link - Update citation from blog to arXiv paper - Add contact email New sections (T3-T8): - Routing Behavior and Cost (Tall-specific latency: 1.76s vs Venti 4.68s) - Limitations (base-vs-system comparison, multimodal inherited not tuned) - Safety (no standalone eval, data governance, deployment guidance) - Hardware Requirements (local deployment focus, ~50B footprint) - Training Details (rank 64, alpha 128, expert parameters) - Parameter count clarification (50B label vs 35B base)
#5 opened 27 days ago
by mindlab-bot
[P1-P5] Add arXiv paper, evaluation protocols, and system details Core updates: - Add arXiv:2608.09819 tag and paper link - Update citation from blog to arXiv paper - Add contact email Evaluation improvements (V2-V4): - Add asterisk marks for imported baseline values - Add VitaBench2 and tau^3-Bench rows (from report Table 8) - Add comprehensive Evaluation Protocols table (Appendix B) - Add UI4A-Bench Layer Scores breakdown - Add Routing Behavior and Cost measurements New sections (V5-V9): - Limitations (single snapshot, no collective intelligence evidence, etc.) - Safety (no standalone eval, data governance, deployment guidance) - Hardware Requirements (validated configs, H20/B300 workloads, DCP notes) - Training Details (LoRA config, optimizer, epochs) - Parameter count clarification (748B label vs ~774.8B stored)
#4 opened 27 days ago
by mindlab-bot
Remove base model metadata
#5 opened 22 days ago
by mindlab-bot
[test] Permission check via PR
#3 opened 27 days ago
by mindlab-bot
anchen1011
submitted a
paper to Daily Papers about 1 month ago