-
Add TRAINING_LOGS.md (companion doc — grep recipes + key-runs index)
-
Add training_logs.zip — verl trainer + k8s-job + vlmsearch-eval stdout for all ablation runs
-
Add EVAL_RESULTS.md (companion doc — eval cmd → result file mapping)
-
Add eval_results.zip — all rollout JSONs + README mapping dirs to eval commands
-
Add single-7b-spatial-baseline-seed42-step100 ckpt (merged from FSDP shards)
-
Add multiturn-3b-wg-nocrop-seed43-step501 ckpt (merged from FSDP shards)
-
Add multiturn-3b-wg-nocrop-seed42-step1495 ckpt (merged from FSDP shards)
-
Add multiturn-3b-wg-baseline-seed43-step501 ckpt (merged from FSDP shards)
-
Add multiturn-3b-wg-baseline-seed42-step1495 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-rethink-seed43-step325 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-rethink-seed42-step1496 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-no23-seed43-step501 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-no23-seed42-step1496 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-baseline-seed43-step501 ckpt (merged from FSDP shards)
-
Add multiturn-3b-vstar-baseline-seed42-step1496 ckpt (merged from FSDP shards)
-
Delete files _perm_test_1777650620.txt with huggingface_hub
-
perm
-
initial commit