ckpt-bigbench

Evolution-Strategies fine-tuning checkpoints from IBM Blue Vela.

The 78-task BIG-Bench series. Every run is ARM=hetero at population 30 over 500 iterations, and the run name carries the base model, the task weighting (bal for balanced, nat for natural), the per-member batch size where it is not 200, and for the step-size ladder the shaping rule and learning rate.

iter<N>.pth are the periodic training checkpoints and final/pytorch_model.pth is the end-of-run save. The two are NOT copies of each other even at the same N -- the final save is written a few ES steps after the last periodic one, measured at a relative L2 difference of 0.002 to 0.004 where ten iterations move 0.008. Treat final/ as the canonical end-of-run weights.

One copy per iteration is published. Where a run has both a final save and a periodic checkpoint at that same iteration, only the final save is here. Periodic checkpoints at earlier iterations are kept, and a run that never reached its end keeps every periodic checkpoint it has.

Runs still training are not here yet. They are published once they reach their target and write a final save, so that no iteration ever appears twice.

All .pth are bf16 state dicts keyed by vLLM parameter names (fused qkv_proj, gate_up_proj), loadable by the trainer resume path.

run	base model	final iteration	periodic checkpoints
bb78-llama8b-bal-hetero	NousResearch/Meta-Llama-3.1-8B-Instruct	500	50
bb78-llama8b-bal-hetero-a1e4	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-a2e4	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-a3e4	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-a4e4	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-nostd-a1e2	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-nostd-a1e3	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-nostd-a2e2	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-bal-hetero-nostd-a5e2	NousResearch/Meta-Llama-3.1-8B-Instruct	500	400
bb78-llama8b-bal-hetero-nostd-a5e3	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-llama8b-nat-hetero	NousResearch/Meta-Llama-3.1-8B-Instruct	500	50
bb78-llama8b-nat-hetero-m400	NousResearch/Meta-Llama-3.1-8B-Instruct	500	none
bb78-olmo7b-bal-hetero	allenai/OLMo-2-1124-7B-Instruct	500	50
bb78-olmo7b-nat-hetero	allenai/OLMo-2-1124-7B-Instruct	500	none
bb78-olmo7b-nat-hetero-m400	allenai/OLMo-2-1124-7B-Instruct	500	none
bb78-q25-7b-bal-hetero	Qwen/Qwen2.5-7B-Instruct	500	none
bb78-q25-7b-nat-hetero	Qwen/Qwen2.5-7B-Instruct	500	none
bb78-q25-7b-nat-hetero-m400	Qwen/Qwen2.5-7B-Instruct	500	none