Hugging Face's logo

allenai
/
tmax-9b

Safetensors
qwen3_5
Eval Results
Model card Files
xet
 Community
4
hamishivi commited on
Commit
63eae4b
·
verified ·
1 Parent(s): fdea3f4

Add files using upload-large-folder tool

Browse files
Files changed (25) hide show
  1. rollouts/README.md +94 -0
  2. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/metadata/metadata.tar.zst.part-000 +3 -0
  3. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/rollouts/rollouts.tar.zst.part-000 +3 -0
  4. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  5. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/metadata/metadata.tar.zst.part-000 +3 -0
  6. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/rollouts/rollouts.tar.zst.part-000 +3 -0
  7. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  8. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/metadata/metadata.tar.zst.part-000 +3 -0
  9. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/rollouts/rollouts.tar.zst.part-000 +3 -0
  10. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  11. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-001 +3 -0
  12. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/metadata/metadata.tar.zst.part-000 +3 -0
  13. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/rollouts/rollouts.tar.zst.part-000 +3 -0
  14. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  15. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183/metadata/metadata.tar.zst.part-000 +3 -0
  16. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/metadata/metadata.tar.zst.part-000 +3 -0
  17. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/rollouts/rollouts.tar.zst.part-000 +3 -0
  18. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  19. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/metadata/metadata.tar.zst.part-000 +3 -0
  20. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/rollouts/rollouts.tar.zst.part-000 +3 -0
  21. rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/training_logprobs/training_logprobs.tar.zst.part-000 +3 -0
  22. rollouts/manifests/SHA256SUMS +23 -0
  23. rollouts/manifests/archive_parts.csv +21 -0
  24. rollouts/manifests/source_files.csv +0 -0
  25. rollouts/manifests/summary.json +93 -0
rollouts/README.md ADDED
@@ -0,0 +1,94 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # tmax-9b SWERL Rollout and Training Logprob Archive
2
+
3
+ This folder contains the raw rollout and trainer logprob data for the run family:
4
+
5
+ `swerl_qwen35_9b_fp32lm_dppo_g32__42`
6
+
7
+ The data is archived without filtering or row-level modification. Files are grouped by
8
+ run fragment and data kind, compressed with `zstd`, and split into approximately
9
+ 10 GiB parts for resumable upload/download.
10
+
11
+ ## Size
12
+
13
+ Raw source size: **292.60 GB / 272.50 GiB**
14
+
15
+ | kind | files | raw size |
16
+ | --- | ---: | ---: |
17
+ | `training_logprobs` | 8400 | 225.91 GB / 210.39 GiB |
18
+ | `rollouts` | 17 | 66.69 GB / 62.11 GiB |
19
+ | `metadata` | 7 | 0.00 GB / 0.00 GiB |
20
+
21
+ ## Run fragments
22
+
23
+ The 500 training steps are split across restart fragments. Some restart boundaries
24
+ overlap; the source files are all preserved.
25
+
26
+ | run fragment | trainer step range | unique trainer steps | training logprob files | rollout files | metadata files | raw size |
27
+ | --- | ---: | ---: | ---: | ---: | ---: | ---: |
28
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982` | `1-98` | 98 | 1568 | 3 | 1 | 41.58 GB / 38.72 GiB |
29
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677` | `91-94` | 4 | 64 | 1 | 1 | 1.26 GB / 1.18 GiB |
30
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077` | `91-292` | 202 | 3232 | 6 | 1 | 122.92 GB / 114.47 GiB |
31
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702` | `291-353` | 63 | 1008 | 2 | 1 | 37.30 GB / 34.74 GiB |
32
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183` | `-` | 0 | 0 | 0 | 1 | 0.00 GB / 0.00 GiB |
33
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883` | `351-368` | 18 | 288 | 1 | 1 | 16.13 GB / 15.03 GiB |
34
+ | `swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805` | `361-500` | 140 | 2240 | 4 | 1 | 73.40 GB / 68.36 GiB |
35
+
36
+ ## Folder structure
37
+
38
+ ```text
39
+ rollouts/
40
+ README.md
41
+ manifests/
42
+ source_files.csv # one row per original source file
43
+ summary.json # counts, sizes, run fragments, step ranges
44
+ archive_parts.csv # one row per uploaded archive part
45
+ SHA256SUMS # checksums for archive parts and manifests
46
+ archives/
47
+ <run_fragment>/
48
+ training_logprobs/
49
+ training_logprobs.tar.zst.part-000
50
+ training_logprobs.tar.zst.part-001
51
+ ...
52
+ rollouts/
53
+ rollouts.tar.zst.part-000
54
+ ...
55
+ metadata/
56
+ metadata.tar.zst.part-000
57
+ ...
58
+ ```
59
+
60
+ ## Reconstructing an archive
61
+
62
+ Each `*.tar.zst.part-*` set is a split zstd-compressed tar stream. To extract one
63
+ archive:
64
+
65
+ ```bash
66
+ cat rollouts/archives/<run_fragment>/<kind>/<kind>.tar.zst.part-* \
67
+ | zstd -d \
68
+ | tar -xvf -
69
+ ```
70
+
71
+ The archive contents expand under:
72
+
73
+ ```text
74
+ run_fragments/<run_fragment>/<kind>/
75
+ ```
76
+
77
+ where `<kind>` is one of `training_logprobs`, `rollouts`, or `metadata`.
78
+
79
+ ## Original file naming
80
+
81
+ Training logprob files:
82
+
83
+ ```text
84
+ <run_fragment>_trainer_logprobs_step<STEP>_rank<RANK>.jsonl
85
+ ```
86
+
87
+ Rollout files:
88
+
89
+ ```text
90
+ <run_fragment>_rollouts_<SHARD>.jsonl
91
+ ```
92
+
93
+ The original filenames are preserved inside each tar archive and indexed in
94
+ `manifests/source_files.csv`.
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:84d7cbd58b1fed524410caf886c18d15fc33682604f51557f544bd5171bfb492
3
+ size 339
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a28d12708a1b371828b815c9d837782c9c92d5f0e97634ccfed891baa316601c
3
+ size 2085151337
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:39ebf66a03c9578ab207275a7dcc7002864eb54830e7cd54d763355f7677c0b7
3
+ size 3325911003
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:01b99d4ab85c18deb5b71d0269a24154c64d7f4d996000991752f7249f87c411
3
+ size 338
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:597f4f36ab78ff13d6917445eb1cd2f5ad4c219ff7b987de2110a59ac86b8632
3
+ size 62657749
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4212baeb5fa234a5bf696f34a2200c4421e4b660d009b388b454da077add46ec
3
+ size 99863263
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:443a5d46b13eb680d36d467385e099e4124b5c3392592c2daed8d863d3bd2d74
3
+ size 343
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3769070e66de299119e8a30a7610bc19e7b3d2b832ee67dd3214897da909bedd
3
+ size 6906317757
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b95e788536a48b0eea6c22b9ded0670da443df76f7e99d2afd206292fb5ad9c
3
+ size 10737418240
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-001 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:28d7a2b5859da2e0e780f91b7c1aef09a577c93276c7bb805807d205aed5ad7e
3
+ size 469793665
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f020baf169f547e97d4d5de6162868b47710fbe5a925d55762d5c9d2db5a2d2e
3
+ size 338
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e8b4acd502238ebd5e1a7c933d4e9a69b4bf9e1db67cfbf2e3c56db54221f15f
3
+ size 2370267011
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d7c3781feaa71c8ac818adc0950889b1c482a523965b7902f8454db9ac5cd9d
3
+ size 3857936755
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2bc44f70c5cb26f5d0124034e64b6ea3dfd07464193e87f7bd34ba5856d1f7c2
3
+ size 339
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:889fc95383a6c17b4494ae9d781ebe5de91fde7d5c55adf348d8c49ce6f8447e
3
+ size 339
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:05e756a97f621a0013109860b0e8953230a0f4e358cc0fe537319cab34f4b3bd
3
+ size 1227965089
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:47b1fc95f5dc5491247783f614432640ffb06e1935edd4fd22f72b98fb81d4db
3
+ size 1735790290
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/metadata/metadata.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c80d5073d2d396b5c505f5e5438372bc17f3df4e9643c8c067d246edd598800a
3
+ size 339
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/rollouts/rollouts.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5f35bd93ad54e793cf4fc95f90b609845a33a641d690e157cce3f6901328bdb
3
+ size 4212470809
rollouts/archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/training_logprobs/training_logprobs.tar.zst.part-000 ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b277681f0cb7d9cfb627ce48375f30094af6d992dacc45820b66debfe64bc9c1
3
+ size 6964925999
rollouts/manifests/SHA256SUMS ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ 84d7cbd58b1fed524410caf886c18d15fc33682604f51557f544bd5171bfb492 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/metadata/metadata.tar.zst.part-000
2
+ a28d12708a1b371828b815c9d837782c9c92d5f0e97634ccfed891baa316601c archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/rollouts/rollouts.tar.zst.part-000
3
+ 39ebf66a03c9578ab207275a7dcc7002864eb54830e7cd54d763355f7677c0b7 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/training_logprobs/training_logprobs.tar.zst.part-000
4
+ 01b99d4ab85c18deb5b71d0269a24154c64d7f4d996000991752f7249f87c411 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/metadata/metadata.tar.zst.part-000
5
+ 597f4f36ab78ff13d6917445eb1cd2f5ad4c219ff7b987de2110a59ac86b8632 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/rollouts/rollouts.tar.zst.part-000
6
+ 4212baeb5fa234a5bf696f34a2200c4421e4b660d009b388b454da077add46ec archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/training_logprobs/training_logprobs.tar.zst.part-000
7
+ 443a5d46b13eb680d36d467385e099e4124b5c3392592c2daed8d863d3bd2d74 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/metadata/metadata.tar.zst.part-000
8
+ 3769070e66de299119e8a30a7610bc19e7b3d2b832ee67dd3214897da909bedd archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/rollouts/rollouts.tar.zst.part-000
9
+ 1b95e788536a48b0eea6c22b9ded0670da443df76f7e99d2afd206292fb5ad9c archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-000
10
+ 28d7a2b5859da2e0e780f91b7c1aef09a577c93276c7bb805807d205aed5ad7e archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-001
11
+ f020baf169f547e97d4d5de6162868b47710fbe5a925d55762d5c9d2db5a2d2e archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/metadata/metadata.tar.zst.part-000
12
+ e8b4acd502238ebd5e1a7c933d4e9a69b4bf9e1db67cfbf2e3c56db54221f15f archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/rollouts/rollouts.tar.zst.part-000
13
+ 7d7c3781feaa71c8ac818adc0950889b1c482a523965b7902f8454db9ac5cd9d archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/training_logprobs/training_logprobs.tar.zst.part-000
14
+ 2bc44f70c5cb26f5d0124034e64b6ea3dfd07464193e87f7bd34ba5856d1f7c2 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183/metadata/metadata.tar.zst.part-000
15
+ 889fc95383a6c17b4494ae9d781ebe5de91fde7d5c55adf348d8c49ce6f8447e archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/metadata/metadata.tar.zst.part-000
16
+ 05e756a97f621a0013109860b0e8953230a0f4e358cc0fe537319cab34f4b3bd archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/rollouts/rollouts.tar.zst.part-000
17
+ 47b1fc95f5dc5491247783f614432640ffb06e1935edd4fd22f72b98fb81d4db archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/training_logprobs/training_logprobs.tar.zst.part-000
18
+ c80d5073d2d396b5c505f5e5438372bc17f3df4e9643c8c067d246edd598800a archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/metadata/metadata.tar.zst.part-000
19
+ b5f35bd93ad54e793cf4fc95f90b609845a33a641d690e157cce3f6901328bdb archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/rollouts/rollouts.tar.zst.part-000
20
+ b277681f0cb7d9cfb627ce48375f30094af6d992dacc45820b66debfe64bc9c1 archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/training_logprobs/training_logprobs.tar.zst.part-000
21
+ e8c2de707dc9d95297551fb56cbf7fb864422720ebfa254f0787931611d4938b manifests/archive_parts.csv
22
+ f8ae9183138a119e5705342a222c4834ea5c7a2f6f047bbe514e2de76eefdb67 manifests/source_files.csv
23
+ 509afdacccb1158a673be2727ad5c32d4c1bba04e2bd4c2cb5fdcd4222420157 manifests/summary.json
rollouts/manifests/archive_parts.csv ADDED
@@ -0,0 +1,21 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ path,run_id,kind,bytes,sha256
2
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982,metadata,339,84d7cbd58b1fed524410caf886c18d15fc33682604f51557f544bd5171bfb492
3
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982,rollouts,2085151337,a28d12708a1b371828b815c9d837782c9c92d5f0e97634ccfed891baa316601c
4
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982,training_logprobs,3325911003,39ebf66a03c9578ab207275a7dcc7002864eb54830e7cd54d763355f7677c0b7
5
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677,metadata,338,01b99d4ab85c18deb5b71d0269a24154c64d7f4d996000991752f7249f87c411
6
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677,rollouts,62657749,597f4f36ab78ff13d6917445eb1cd2f5ad4c219ff7b987de2110a59ac86b8632
7
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677,training_logprobs,99863263,4212baeb5fa234a5bf696f34a2200c4421e4b660d009b388b454da077add46ec
8
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077,metadata,343,443a5d46b13eb680d36d467385e099e4124b5c3392592c2daed8d863d3bd2d74
9
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077,rollouts,6906317757,3769070e66de299119e8a30a7610bc19e7b3d2b832ee67dd3214897da909bedd
10
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077,training_logprobs,10737418240,1b95e788536a48b0eea6c22b9ded0670da443df76f7e99d2afd206292fb5ad9c
11
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077/training_logprobs/training_logprobs.tar.zst.part-001,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077,training_logprobs,469793665,28d7a2b5859da2e0e780f91b7c1aef09a577c93276c7bb805807d205aed5ad7e
12
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702,metadata,338,f020baf169f547e97d4d5de6162868b47710fbe5a925d55762d5c9d2db5a2d2e
13
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702,rollouts,2370267011,e8b4acd502238ebd5e1a7c933d4e9a69b4bf9e1db67cfbf2e3c56db54221f15f
14
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702,training_logprobs,3857936755,7d7c3781feaa71c8ac818adc0950889b1c482a523965b7902f8454db9ac5cd9d
15
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183,metadata,339,2bc44f70c5cb26f5d0124034e64b6ea3dfd07464193e87f7bd34ba5856d1f7c2
16
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883,metadata,339,889fc95383a6c17b4494ae9d781ebe5de91fde7d5c55adf348d8c49ce6f8447e
17
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883,rollouts,1227965089,05e756a97f621a0013109860b0e8953230a0f4e358cc0fe537319cab34f4b3bd
18
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883,training_logprobs,1735790290,47b1fc95f5dc5491247783f614432640ffb06e1935edd4fd22f72b98fb81d4db
19
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/metadata/metadata.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805,metadata,339,c80d5073d2d396b5c505f5e5438372bc17f3df4e9643c8c067d246edd598800a
20
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/rollouts/rollouts.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805,rollouts,4212470809,b5f35bd93ad54e793cf4fc95f90b609845a33a641d690e157cce3f6901328bdb
21
+ archives/swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805/training_logprobs/training_logprobs.tar.zst.part-000,swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805,training_logprobs,6964925999,b277681f0cb7d9cfb627ce48375f30094af6d992dacc45820b66debfe64bc9c1
rollouts/manifests/source_files.csv ADDED
The diff for this file is too large to render. See raw diff
 
rollouts/manifests/summary.json ADDED
@@ -0,0 +1,93 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "by_kind": {
3
+ "metadata": {
4
+ "bytes": 1246,
5
+ "files": 7
6
+ },
7
+ "rollouts": {
8
+ "bytes": 66687660021,
9
+ "files": 17
10
+ },
11
+ "training_logprobs": {
12
+ "bytes": 225909541479,
13
+ "files": 8400
14
+ }
15
+ },
16
+ "by_run_fragment": {
17
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779647982": {
18
+ "bytes": 41580369508,
19
+ "files": 1572,
20
+ "metadata_files": 1,
21
+ "rollout_files": 3,
22
+ "training_logprobs_files": 1568,
23
+ "training_step_count": 98,
24
+ "training_step_max": 98,
25
+ "training_step_min": 1
26
+ },
27
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779685677": {
28
+ "bytes": 1263454234,
29
+ "files": 66,
30
+ "metadata_files": 1,
31
+ "rollout_files": 1,
32
+ "training_logprobs_files": 64,
33
+ "training_step_count": 4,
34
+ "training_step_max": 94,
35
+ "training_step_min": 91
36
+ },
37
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779689077": {
38
+ "bytes": 122916532328,
39
+ "files": 3239,
40
+ "metadata_files": 1,
41
+ "rollout_files": 6,
42
+ "training_logprobs_files": 3232,
43
+ "training_step_count": 202,
44
+ "training_step_max": 292,
45
+ "training_step_min": 91
46
+ },
47
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779775702": {
48
+ "bytes": 37304656096,
49
+ "files": 1011,
50
+ "metadata_files": 1,
51
+ "rollout_files": 2,
52
+ "training_logprobs_files": 1008,
53
+ "training_step_count": 63,
54
+ "training_step_max": 353,
55
+ "training_step_min": 291
56
+ },
57
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779805183": {
58
+ "bytes": 178,
59
+ "files": 1,
60
+ "metadata_files": 1,
61
+ "rollout_files": 0,
62
+ "training_logprobs_files": 0,
63
+ "training_step_count": 0,
64
+ "training_step_max": null,
65
+ "training_step_min": null
66
+ },
67
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779806883": {
68
+ "bytes": 16133088958,
69
+ "files": 290,
70
+ "metadata_files": 1,
71
+ "rollout_files": 1,
72
+ "training_logprobs_files": 288,
73
+ "training_step_count": 18,
74
+ "training_step_max": 368,
75
+ "training_step_min": 351
76
+ },
77
+ "swerl_qwen35_9b_fp32lm_dppo_g32__42__1779819805": {
78
+ "bytes": 73399101444,
79
+ "files": 2245,
80
+ "metadata_files": 1,
81
+ "rollout_files": 4,
82
+ "training_logprobs_files": 2240,
83
+ "training_step_count": 140,
84
+ "training_step_max": 500,
85
+ "training_step_min": 361
86
+ }
87
+ },
88
+ "run_prefix": "swerl_qwen35_9b_fp32lm_dppo_g32__42",
89
+ "source_root": "/weka/oe-adapt-default/allennlp/deletable_rollouts",
90
+ "total_bytes": 292597202746,
91
+ "total_files": 8424,
92
+ "total_size": "292.60 GB / 272.50 GiB"
93
+ }