JamesK2W commited on
Commit
cac86db
·
verified ·
1 Parent(s): 78d2a3b

README: model table

Browse files
Files changed (1) hide show
  1. README.md +2 -1
README.md CHANGED
@@ -23,4 +23,5 @@ path = snapshot_download("JamesK2W/viewagent", allow_patterns=["nemi_sft_epoch_2
23
 
24
  | folder | note |
25
  |---|---|
26
- | `gnemi_epoch_0` | SFT on 1,024 graph_near_miss demonstrations (prompt-level union view graph of the base model D0 rollouts, stitched shortest prefix to within 2 units of the target, then submit); epoch_0 start of the rl -> sft -> rl loop. eval pass@8 4.7% (strict). |
 
 
23
 
24
  | folder | note |
25
  |---|---|
26
+ | `gnemi_epoch_0` | SFT on 1,024 graph_near_miss demonstrations (prompt-level union view graph of the base model D0 rollouts, stitched shortest prefix to within 2 units of the targ |
27
+ | `gnemi_epoch_1_final` | GRPO epoch 1 (71 steps, 32 prompts x 8 rollouts, D0 train half) from viewagent_graph_near_miss_epoch_0. Weights are the last saved checkpoint, step 60 (the run ended at step 71 without a final save; steps 61-71 were not kept). Eval pass@1 at step 60: 1.25% (7/560). |