gyung commited on
Commit
de5bf76
·
verified ·
1 Parent(s): 015a5eb

Add files using upload-large-folder tool

Browse files
Files changed (23) hide show
  1. .gitattributes +9 -0
  2. stage4c-korean-tool-finance-repeat2-step700000/all_config.yaml +46 -0
  3. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.0.pt +3 -0
  4. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.1.pt +3 -0
  5. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.2.pt +3 -0
  6. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.3.pt +3 -0
  7. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.4.pt +3 -0
  8. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.5.pt +3 -0
  9. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.6.pt +3 -0
  10. stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.7.pt +3 -0
  11. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/.metadata +3 -0
  12. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__0_0.distcp +3 -0
  13. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__1_0.distcp +3 -0
  14. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__2_0.distcp +3 -0
  15. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__3_0.distcp +3 -0
  16. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__4_0.distcp +3 -0
  17. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__5_0.distcp +3 -0
  18. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__6_0.distcp +3 -0
  19. stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__7_0.distcp +3 -0
  20. stage4c-korean-tool-finance-repeat2-step700000/latest_checkpoint.txt +1 -0
  21. stage4c-korean-tool-finance-repeat2-step700000/step_700000_info.json +8 -0
  22. stage4c-korean-tool-finance-repeat2-step700000/train_metadata.yaml +13 -0
  23. stage4c-korean-tool-finance-repeat2-step700000/upload_manifest.json +1 -0
.gitattributes CHANGED
@@ -843,3 +843,12 @@ stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__5_0.distcp fi
843
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
844
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
845
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
 
 
 
 
 
 
 
 
 
 
843
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
844
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
845
  stage4c-korean-tool-finance-repeat2-step690000/fsdp2_step_690000/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
846
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__7_0.distcp filter=lfs diff=lfs merge=lfs -text
847
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
848
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
849
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__5_0.distcp filter=lfs diff=lfs merge=lfs -text
850
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
851
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
852
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
853
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
854
+ stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/.metadata filter=lfs diff=lfs merge=lfs -text
stage4c-korean-tool-finance-repeat2-step700000/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-stage4c-korean-tool-finance-repeat2-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-stage3c-local-terminal-repeat2-gbs180
39
+ resume_step: null
40
+ resume_step_offset: 686132
41
+ run_name: KoHRM-Text-1.4B-stage4c-korean-tool-finance-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
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.0.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9355c0a80f6bd473da8121cc6177e609bc8ff9c8f74a323062fdf55e343449d
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.1.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8054bc53fbc3f648b6bf3fb4ae46fc754a98bbc8fbbf9539bb8fb9dfad02b83b
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.2.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6308675f5be3611ad8e3be6a12980f09a4fca0da228ada8740137587ba8ac876
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.3.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c6d540c9a4ce992faff4fbe73e45c796e46e28c13595ae927b4520b01f076714
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.4.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b3948d93aad03cf97363c5c40f2d79a4b7e8a787740cbc049af9826ee9dc3272
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.5.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:57f26ab5f2e70d41e0e1e4d244c44560e473f4713f5812e66502501ecb7f8071
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.6.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:309d791cafd2b100e36a801330707c4efd6715a8f021b0a4d24aea527b1baab8
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/carry_step_700000.7.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:315010ed3b284b679d81a5938da73ec7cbfce0f91da25a00c72d019e478b4c66
3
+ size 1333
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/.metadata ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a19111e610db6554b6565802be804eee740279961dccb1e3c0aa2e76df59bf3b
3
+ size 983820
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__0_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a41589195c72ae5352a57036166c8d8fba7f295df3ca814e3d6897ff41689774
3
+ size 2769065329
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__1_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f7ed8dea6f06bdc261bfecfa32d9622bc6eb7e67fc848b5ff4f1fc863f50bf0
3
+ size 2769090643
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__2_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:136894e9a9a2a0508aecd0adff35be811b222c54374d3aa68a200e7870d726ce
3
+ size 2769090643
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__3_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:68312b144ee1194c4d86eda7cfe2690e323b096a1a9b598e9f3675214f86d576
3
+ size 2769090643
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__4_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4045827c8ed0093822afab61c49b134018b406778e0e7ac59784dba44f862988
3
+ size 2769090643
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__5_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1f047a94ebf19e72033b7f1aca8e19d8f4dd418aefd435ff11072abd489d0552
3
+ size 2769090643
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__6_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:43d1e8a2dfba3efa9d6e7b64a0880d28610cc591f233f7f954a5b53b1ea8eafa
3
+ size 2769091588
stage4c-korean-tool-finance-repeat2-step700000/fsdp2_step_700000/__7_0.distcp ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bed918ec7eaaecf0c96eef0e204ac7bf7cf31ffbf2bb5233b94801dc7cd49255
3
+ size 2769098756
stage4c-korean-tool-finance-repeat2-step700000/latest_checkpoint.txt ADDED
@@ -0,0 +1 @@
 
 
1
+ step_700000
stage4c-korean-tool-finance-repeat2-step700000/step_700000_info.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "tag": "step_700000",
3
+ "global_step": 700000,
4
+ "stage_start_step": 686132,
5
+ "skip_batches_hint": 13868,
6
+ "data_path": "/home/work/.data/hrm_text_prepared/koterm_korean_tool_finance_mix_v1",
7
+ "global_batch_size": 180224
8
+ }
stage4c-korean-tool-finance-repeat2-step700000/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
stage4c-korean-tool-finance-repeat2-step700000/upload_manifest.json ADDED
@@ -0,0 +1 @@
 
 
1
+ stage4c-korean-tool-finance-repeat2 step_700000 raw resume checkpoint