gyung commited on
Commit
38fa133
·
verified ·
1 Parent(s): 79d16e4

Add files using upload-large-folder tool

Browse files
Files changed (23) hide show
  1. .gitattributes +9 -0
  2. stage1c-hrm-fastcap-repeat2-step510000/all_config.yaml +46 -0
  3. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.0.pt +3 -0
  4. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.1.pt +3 -0
  5. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.2.pt +3 -0
  6. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.3.pt +3 -0
  7. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.4.pt +3 -0
  8. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.5.pt +3 -0
  9. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.6.pt +3 -0
  10. stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.7.pt +3 -0
  11. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/.metadata +3 -0
  12. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__0_0.distcp +3 -0
  13. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__1_0.distcp +3 -0
  14. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__2_0.distcp +3 -0
  15. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__3_0.distcp +3 -0
  16. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__4_0.distcp +3 -0
  17. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__5_0.distcp +3 -0
  18. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__6_0.distcp +3 -0
  19. stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__7_0.distcp +3 -0
  20. stage1c-hrm-fastcap-repeat2-step510000/latest_checkpoint.txt +1 -0
  21. stage1c-hrm-fastcap-repeat2-step510000/step_510000_info.json +8 -0
  22. stage1c-hrm-fastcap-repeat2-step510000/train_metadata.yaml +13 -0
  23. stage1c-hrm-fastcap-repeat2-step510000/upload_manifest.json +1 -0
.gitattributes CHANGED
@@ -618,3 +618,12 @@ stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/__4_0.distcp filter=lfs diff=lf
618
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
619
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
620
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/.metadata filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
618
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
619
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
620
  stage4b-korean-tool-finance-repeat/fsdp2_epoch_1/.metadata filter=lfs diff=lfs merge=lfs -text
621
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
622
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
623
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
624
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/.metadata filter=lfs diff=lfs merge=lfs -text
625
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__5_0.distcp filter=lfs diff=lfs merge=lfs -text
626
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
627
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
628
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
629
+ stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__7_0.distcp filter=lfs diff=lfs merge=lfs -text
stage1c-hrm-fastcap-repeat2-step510000/all_config.yaml ADDED
@@ -0,0 +1,46 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ arch:
2
+ H_cycles: 2
3
+ H_override: {}
4
+ L_cycles: 3
5
+ bp_max_steps: 5
6
+ bp_warmup_ratio: 0.2
7
+ expansion: 4
8
+ half_layers: true
9
+ head: lm_head@LMHead
10
+ hidden_size: 1536
11
+ init_type: lecun_normal
12
+ n_layers: 32
13
+ name: baselines.hrm_nocarry_bp_warmup@HierarchicalReasoningModel
14
+ norm_eps: 1.0e-06
15
+ norm_type: pre
16
+ num_heads: 12
17
+ pos_emb_type: rope
18
+ rope_theta: 10000.0
19
+ beta1: 0.9
20
+ beta2: 0.95
21
+ checkpoint_interval: 1
22
+ checkpoint_keep_last: 2
23
+ checkpoint_path: /home/work/.data/hrm_text_checkpoints/KoHRM-Text-1.4B-stage1c-hrm-fastcap-repeat2-gbs180
24
+ checkpoint_step_interval: 10000
25
+ data:
26
+ path: /home/work/.data/hrm_text_prepared/koterm_hrm_cleaned_fastcap_stage1_v1
27
+ target_only: true
28
+ ema: 0.9999
29
+ epochs: 1
30
+ fwd_bwd_dtype: bfloat16
31
+ global_batch_size: 180224
32
+ log_interval: 5
33
+ lr: 0.00022
34
+ lr_min_ratio: 1.0
35
+ lr_warmup_steps: 2000
36
+ project_name: KoHRM-Text
37
+ resume_epoch: null
38
+ resume_from: /home/work/.data/hrm_text_checkpoints/KoHRM-Text-1.4B-stage4b-korean-tool-finance-repeat-gbs180
39
+ resume_step: null
40
+ resume_step_offset: 470077
41
+ run_name: KoHRM-Text-1.4B-stage1c-hrm-fastcap-repeat2
42
+ seed: 0
43
+ skip_batches: 0
44
+ total_steps_override: 700000
45
+ weight_decay: 0.1
46
+ weights_only_resume_from_ema: false
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67487ea3edfba4ec9245bd706a7d4756502ee3b303cd8b12fe2c4e49a63fa6f8
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f1abb43edb6b6dbe00de141d1561bb4510c29bf3f088637bb8480f5dfc6ac544
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:056349626d76a067320d16a00d5ca1a2dcfefdedb759c4bebafc696239229013
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.3.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cb6606729fb7a40e3d5b57e160bb2ad9cd8497aa5cabbea2e96502501f88784e
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.4.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b9b6fed405ae33f32e569b3ffdc8fb5120f6aeea89192c7257c4db896a074242
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.5.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f349c8aabc11a6ec85938477462cedfdc178a9e66648e340843258a08421a7e7
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.6.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7da0686f7c5b932495e210d57ffb89888ea9490fee4e463f606ab98d364a02fc
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/carry_step_510000.7.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bccbb4dee6d6677a2f17d24fb356cf59bc2d396799e2372414b62d312f55cd1
3
+ size 1333
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e95408c4010555dc544d7d7ebeb7d7d314ffac9ca6768fd3c3150063727dcedd
3
+ size 983812
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__0_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ee2cf23420df8229a819d92d14324b962b340ef013043ec701d55f5080a29145
3
+ size 2769065329
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__1_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ec08126934ea4adb6b89b7099b1cea1a26330ba36d34a0f5e96785e1b4c6cb9
3
+ size 2769090643
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__2_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8a732e120343457f060d63c7ce46e814d082023cf6441e50c276e2b75f151f4b
3
+ size 2769090643
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__3_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6b0910876051ca0afbb76f640b3d02f55d7c521d54e80907836b3b9bd481708
3
+ size 2769090643
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__4_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89342c6ae7291441fb8ff8352ee95c1541a0b13751ab952487916a031d3f90f7
3
+ size 2769090643
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__5_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5d952b63cdd94a9b506d406229baef12f8120801fbaed8db84ef44310ef6907e
3
+ size 2769090643
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__6_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:56609b1340006adddfcb4a46d66ce7d8ff2ca0c92877ed64428fd3a1289ce9c0
3
+ size 2769091588
stage1c-hrm-fastcap-repeat2-step510000/fsdp2_step_510000/__7_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:663966c54f92729c0bbe2932777e6d56aab22b686b43580f5d8f3b78083c1945
3
+ size 2769098756
stage1c-hrm-fastcap-repeat2-step510000/latest_checkpoint.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ step_520000
stage1c-hrm-fastcap-repeat2-step510000/step_510000_info.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "step_510000",
3
+ "global_step": 510000,
4
+ "stage_start_step": 470077,
5
+ "skip_batches_hint": 39923,
6
+ "data_path": "/home/work/.data/hrm_text_prepared/koterm_hrm_cleaned_fastcap_stage1_v1",
7
+ "global_batch_size": 180224
8
+ }
stage1c-hrm-fastcap-repeat2-step510000/train_metadata.yaml ADDED
@@ -0,0 +1,13 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ max_seq_len: 4096
2
+ tokenizer_info:
3
+ boq: <|im_start|>
4
+ condition_mapping:
5
+ cot: <|object_ref_end|>
6
+ direct: <|object_ref_start|>
7
+ noisy: <|quad_start|>
8
+ synth: <|quad_end|>
9
+ eoa: <|box_end|>
10
+ eoq: <|im_end|>
11
+ tokenizer_path: /home/work/.data/huggingface/trained_tokenizers/hrm-ko-terminal-131k-v1/tokenizer.json
12
+ total_length: 14554291763
13
+ vocab_size: 131072
stage1c-hrm-fastcap-repeat2-step510000/upload_manifest.json ADDED
@@ -0,0 +1 @@
 
 
1
+ stage1c-hrm-fastcap-repeat2 step_510000 raw resume checkpoint