gyung commited on
Commit
acee260
·
verified ·
1 Parent(s): b64e879

Add files using upload-large-folder tool

Browse files
Files changed (24) hide show
  1. .gitattributes +9 -0
  2. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/README.md +7 -0
  3. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/all_config.yaml +46 -0
  4. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.0.pt +3 -0
  5. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.1.pt +3 -0
  6. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.2.pt +3 -0
  7. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.3.pt +3 -0
  8. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.4.pt +3 -0
  9. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.5.pt +3 -0
  10. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.6.pt +3 -0
  11. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.7.pt +3 -0
  12. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/epoch_1_info.json +8 -0
  13. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/.metadata +3 -0
  14. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__0_0.distcp +3 -0
  15. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__1_0.distcp +3 -0
  16. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__2_0.distcp +3 -0
  17. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__3_0.distcp +3 -0
  18. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__4_0.distcp +3 -0
  19. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__5_0.distcp +3 -0
  20. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__6_0.distcp +3 -0
  21. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__7_0.distcp +3 -0
  22. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/latest_checkpoint.txt +1 -0
  23. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/train_metadata.yaml +13 -0
  24. epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/upload_manifest.json +8 -0
.gitattributes CHANGED
@@ -591,3 +591,12 @@ stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__3_0.distcp fil
591
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
592
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
593
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
591
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
592
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
593
  stage4b-korean-tool-finance-repeat-step470000/fsdp2_step_470000/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
594
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
595
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
596
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/.metadata filter=lfs diff=lfs merge=lfs -text
597
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
598
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__7_0.distcp filter=lfs diff=lfs merge=lfs -text
599
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
600
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
601
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
602
+ epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__5_0.distcp filter=lfs diff=lfs merge=lfs -text
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/README.md ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ # KoHRM-Text-1.4B Epoch 2 Final Raw Checkpoint
2
+
3
+ - Epoch label: `epoch2-final`
4
+ - Final stage: `stage4b-korean-tool-finance-repeat`
5
+ - Global step: `470077`
6
+ - Checkpoint tag: `epoch_1` inside the stage4b checkpoint directory
7
+ - Purpose: fixed raw FSDP2 resume checkpoint for the end of KoHRM epoch 2.
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/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-stage4b-korean-tool-finance-repeat-gbs180
24
+ checkpoint_step_interval: 10000
25
+ data:
26
+ path: /home/work/.data/hrm_text_prepared/koterm_korean_tool_finance_mix_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-stage3b-local-terminal-repeat-gbs180
39
+ resume_step: null
40
+ resume_step_offset: 453253
41
+ run_name: KoHRM-Text-1.4B-stage4b-korean-tool-finance-repeat
42
+ seed: 0
43
+ skip_batches: 0
44
+ total_steps_override: 700000
45
+ weight_decay: 0.1
46
+ weights_only_resume_from_ema: false
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4e3b089e35eacca121e8bc850c0fc138c4ef45a63cf9e370e2f852d6245db36b
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:66952aadc5b6f1d9d38cd436e6dba2c3b6a487138c6960beb815672ddf699495
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a7da182d5d1dfe900b2018b6e4fe6d318c69f791b7a3c94c1727e2112a5f57d
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.3.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:45156a7f9cef48f22d7a3c46c59d92394c3da40b2b596f2681cecece9156177e
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.4.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34356bd35b48ac6ce98742241b2c0e1c96147c36743415c7b0e432ae28f8bfc8
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.5.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6f5934083e16382c0d96bab003d7f577ced0da026175aff5a4ad2aaf31c603f6
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.6.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93480251cc3a3285f6cea88b5f8b7c6d46672f04cacd0623127885ff4469e7d8
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/carry_epoch_1.7.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d2de90149405f8416a3f1c4bb6b69b52843f8a2f835c1b874bd13c13129fc3f5
3
+ size 1309
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/epoch_1_info.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "epoch_1",
3
+ "global_step": 470077,
4
+ "stage_start_step": 453253,
5
+ "skip_batches_hint": 16824,
6
+ "data_path": "/home/work/.data/hrm_text_prepared/koterm_korean_tool_finance_mix_v1",
7
+ "global_batch_size": 180224
8
+ }
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5c8966bd008efa0c49e1a4a476fb1ceb0dcef7bb5e23a78b4e0626cd443cd50e
3
+ size 983815
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__0_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b500cac0bbf4cdbe0b9ab5e1a71e3b685fd1835dc7effb770409778fe48fa991
3
+ size 2769065329
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__1_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9be9e21dc11094c803edf54d4140be3d9f52e593b4c17a091e28a48d850310e6
3
+ size 2769090643
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__2_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7004a1c8f881fcd5a604acd35c29f9aa37fe8026e72e5153f51db796c0de2a1d
3
+ size 2769090643
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__3_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:19a92fde30a97addfbb125b546a0e97f066b52b32df7ecf2f910d9e3448673ee
3
+ size 2769090643
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__4_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8417d905aabb86e62e700eb93e1cbbb62b135a01065f6b58f772774d9f7345d0
3
+ size 2769090643
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__5_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:24ec42fc2f541ee9bb55c5e113875f475fd206c600c22ef71d25d1523488ba7c
3
+ size 2769090643
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__6_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e6b77037823e1dd5cb49b952073c8aa2bbc538564f2af2acec449919e2bda64
3
+ size 2769091588
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/fsdp2_epoch_1/__7_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d36ade5baa815b315808026392dc3f964919480b4e2cda2172ae495143c1b16b
3
+ size 2769098756
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/latest_checkpoint.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ epoch_1
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/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/hrm_text_prepared/koterm_korean_tool_finance_mix_v1
12
+ total_length: 3020504008
13
+ vocab_size: 131072
epoch2-final-stage4b-korean-tool-finance-repeat-globalstep-470077/upload_manifest.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "artifact": "KoHRM-Text-1.4B Epoch 2 Final Raw Checkpoint",
3
+ "epoch_label": "epoch2-final",
4
+ "stage": "stage4b-korean-tool-finance-repeat",
5
+ "global_step": 470077,
6
+ "checkpoint_tag": "epoch_1",
7
+ "source_checkpoint": "/home/work/.data/hrm_text_checkpoints/KoHRM-Text-1.4B-stage4b-korean-tool-finance-repeat-gbs180"
8
+ }