Hugging Face's logo

ZachW
/
extrapolation_rl

Text Generation
Transformers
Safetensors
English
reasoning
reinforcement-learning
synthetic-data
extrapolation
Model card Files
xet
 Community
ZachW commited on
Commit
255c9ea
·
verified ·
1 Parent(s): a927ab0

Upload extrapolation artifact: 050302 README

Browse files
Files changed (1) hide show
  1. 050302/README.md +64 -0
050302/README.md ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Extrapolation Release Checkpoints
2
+
3
+ This release contains the 020305 and 050302 extrapolation checkpoint batches for the Interplay-LM-Reasoning project. The checkpoints are organized together in one Hugging Face repository rather than as separate model repositories.
4
+
5
+ ## Layout
6
+
7
+ ```text
8
+ 020305/
9
+ README.md
10
+ pretrain/checkpoint-18809/
11
+ rl/op11_14_step200/
12
+ rl/op13_20_step200/
13
+ rl/op15_20_step200/
14
+ opd_refkl/op11_14_step200/
15
+ evals/
16
+ 050302/
17
+ README.md
18
+ pretrain/checkpoint-18586/
19
+ rl/op11_14_step200/
20
+ rl/op13_20_step200/
21
+ rl/op15_20_step200/
22
+ opd_refkl/op13_20_step200/
23
+ evals/
24
+ ```
25
+
26
+ The default upload includes compact evaluation files: `summary.csv`, `summary.json`, per-model `*_metrics.json`, graph `results.json`, and score reports. Raw generation JSONL files and per-sample graph-result JSONL files are large and are uploaded when the upload helper is run with `--include-generations` and `--include-per-sample-results`.
27
+
28
+ ## Run Index
29
+
30
+ | Model path | Exp name | Script/config | Model name | Time |
31
+ | --- | --- | --- | --- | --- |
32
+ | `020305/pretrain/checkpoint-18809` | `interplay_fig3_pretrain_arkeng_h100x4_r12` | `scripts/arnold/launch_fig3_pretrain_h100x4_arkeng_r12.yaml` | `pretrain_ckpt18809` | 2026-08-14 23:57:05 to 2026-08-15 05:48:41 UTC |
33
+ | `020305/rl/op11_14_step200` | `interplay_rl_ckpt18809_op11_14_arkeng_a100x8_r62` | `scripts/arnold/launch_rl_ckpt18809_op11_14_a100x8_arkeng_r62.yaml` | `rl_ckpt18809_op11_14_step200_merged_hf` | 2026-08-20 12:20:11 to 14:50:50 UTC |
34
+ | `020305/rl/op13_20_step200` | `interplay_rl_ckpt18809_op13_20_arkeng_a100x8_r62` | `scripts/arnold/launch_rl_ckpt18809_op13_20_a100x8_arkeng_r62.yaml` | `rl_ckpt18809_op13_20_step200_merged_hf` | 2026-08-20 12:21:26 to 14:51:54 UTC |
35
+ | `020305/rl/op15_20_step200` | `interplay_rl_ckpt18809_op15_20_arkeng_a100x8_r62` | `scripts/arnold/launch_rl_ckpt18809_op15_20_a100x8_arkeng_r62.yaml` | `rl_ckpt18809_op15_20_step200_merged_hf` | 2026-08-20 12:21:04 to 14:53:21 UTC |
36
+ | `020305/opd_refkl/op11_14_step200` | `interplay_opd_ckpt18809_teacher_op11_14_refkl_arkeng_a100x8_r75` | `scripts/arnold/launch_opd_ckpt18809_teacher_op11_14_refkl_a100x8_arkeng_r75.yaml` | `opd_ckpt18809_refkl_op11_14_s200_merged_hf` | 2026-08-22 09:28:42 to 11:59:32 UTC |
37
+ | `050302/pretrain/checkpoint-18586` | `interplay_fig3_pretrain_050302_arkeng_a100x8_r13` | `scripts/arnold/launch_fig3_pretrain_050302_a100x8_arkeng_r13.yaml` | `pretrain_050302_ckpt18586` | 2026-08-19 06:54:03 to 12:39:48 UTC |
38
+ | `050302/rl/op11_14_step200` | `interplay_rl_050302pretrain_op11_14posttrain_arkeng_a100x8_r76` | `scripts/arnold/launch_rl_050302pretrain_op11_14posttrain_a100x8_arkeng_r76.yaml` | `rl050302_op11_14_step200_merged_hf` | 2026-08-22 09:21:36 to 21:35:04 UTC |
39
+ | `050302/rl/op13_20_step200` | `interplay_rl_050302pretrain_op13_20posttrain_arkeng_a100x8_r76` | `scripts/arnold/launch_rl_050302pretrain_op13_20posttrain_a100x8_arkeng_r76.yaml` | `rl050302_op13_20_step200_merged_hf` | 2026-08-22 09:21:34 to 21:35:07 UTC |
40
+ | `050302/rl/op15_20_step200` | `interplay_rl_050302pretrain_op15_20posttrain_arkeng_a100x8_r76` | `scripts/arnold/launch_rl_050302pretrain_op15_20posttrain_a100x8_arkeng_r76.yaml` | `rl050302_op15_20_step200_merged_hf` | 2026-08-22 09:21:35 to 21:35:38 UTC |
41
+ | `050302/opd_refkl/op13_20_step200` | `interplay_opd_050302_teacher_op13_20_refkl_arkeng_a100x8_r80` | `scripts/arnold/launch_opd_050302_teacher_op13_20_refkl_a100x8_arkeng_r80.yaml` | `opd050302_refkl_s200_merged_hf` | 2026-08-23 22:29:00 to 2026-08-24 10:44:46 UTC |
42
+
43
+ ## Evaluation Summary
44
+
45
+ All rows use 1,152,000 ID examples, 1,280,000 OOD examples, and 2,432,000 total examples.
46
+
47
+ | Model name | ID acc | OOD acc | Total acc | Evaluation artifact path |
48
+ | --- | ---: | ---: | ---: | --- |
49
+ | `pretrain_ckpt18809` | 0.362037 | 0.100438 | 0.224353 | `020305/evals/pretrain_ckpt18809/` |
50
+ | `rl_ckpt18809_op11_14_step200_merged_hf` | 0.800709 | 0.369602 | 0.573810 | `020305/evals/rl_ckpt18809_op11_14_step200_merged_hf/` |
51
+ | `rl_ckpt18809_op13_20_step200_merged_hf` | 0.683144 | 0.289346 | 0.475882 | `020305/evals/rl_ckpt18809_op13_20_step200_merged_hf/` |
52
+ | `rl_ckpt18809_op15_20_step200_merged_hf` | 0.388938 | 0.160735 | 0.268831 | `020305/evals/rl_ckpt18809_op15_20_step200_merged_hf/` |
53
+ | `opd_ckpt18809_refkl_op11_14_s200_merged_hf` | 0.801092 | 0.365645 | 0.571909 | `020305/evals/opd_ckpt18809_refkl_op11_14_s200_merged_hf/` |
54
+ | `pretrain_050302_ckpt18586` | 0.659137 | 0.174320 | 0.403970 | `050302/evals/pretrain_050302_ckpt18586/` |
55
+ | `rl050302_op11_14_step200_merged_hf` | 0.756464 | 0.348082 | 0.541526 | `050302/evals/rl050302_op11_14_step200_merged_hf/` |
56
+ | `rl050302_op13_20_step200_merged_hf` | 0.709700 | 0.343556 | 0.516993 | `050302/evals/rl050302_op13_20_step200_merged_hf/` |
57
+ | `rl050302_op15_20_step200_merged_hf` | 0.719865 | 0.291799 | 0.494567 | `050302/evals/rl050302_op15_20_step200_merged_hf/` |
58
+ | `opd050302_refkl_s200_merged_hf` | 0.709348 | 0.342995 | 0.516530 | `050302/evals/opd050302_refkl_s200_merged_hf/` |
59
+
60
+ ## Source Pointers
61
+
62
+ - Data: <https://huggingface.co/datasets/Interplay-LM-Reasoning/composition>
63
+ - Context data: <https://huggingface.co/datasets/Interplay-LM-Reasoning/context>
64
+ - Repository documentation: `docs/extrapolation_release_wrapup.md` in the GitHub repository.