Add files using upload-large-folder tool
Browse files- .gitignore +23 -0
- artifacts/tokenizer-20k-v2/training_metadata.json +46 -0
- artifacts/tokenizer-20k-v3/special_tokens_map.json +30 -0
- artifacts/tokenizer-20k-v3/tokenizer.vocab +0 -0
- artifacts/tokenizer-20k-v3/tokenizer_config.json +51 -0
- artifacts/tokenizer-20k-v3/training_metadata.json +46 -0
- reports/fixed-ko-cpt-3b-5k/checksums.sha256 +2 -0
- reports/fixed-objective-100m-ab/broken-final-evaluation.json +30 -0
- reports/fixed-objective-100m-ab/recovery-byte-evaluation.json +17 -0
- reports/fixed-objective-100m-ab/recovery-checkpoint.txt +1 -0
- reports/fixed-objective-100m-ab/recovery-evaluation.json +30 -0
- reports/fixed-objective-100m-ab/recovery-train.log +101 -0
- reports/fixed-objective-100m-ab/scratch-byte-evaluation.json +17 -0
- reports/fixed-objective-100m-ab/scratch-checkpoint.txt +1 -0
- reports/fixed-objective-100m-ab/scratch-evaluation.json +30 -0
- reports/fixed-objective-100m-ab/scratch-train.log +101 -0
- reports/fixed-recovery-1b/byte-evaluation.json +17 -0
- reports/fixed-recovery-1b/evaluation.json +30 -0
- reports/fixed-recovery-1b/final_checkpoint.txt +1 -0
- reports/fixed-recovery-1b/gate.json +7 -0
- reports/fixed-recovery-1b/train.log +186 -0
- reports/fixed-recovery-6b/evaluation.json +30 -0
- reports/fixed-recovery-6b/gate.json +7 -0
- reports/kawk-1.5-50m-v2/MODEL_CARD.md +101 -0
- reports/kawk-1.5-50m-v2/byte-evaluation.json +17 -0
- reports/kawk-1.5-50m-v2/checksums.sha256 +2 -0
- reports/kawk-1.5-50m-v2/evaluation.json +30 -0
- reports/kawk-1.5-50m-v2/final_checkpoint.txt +1 -0
- reports/kawk-1.5-50m-v2/train.log +0 -0
- reports/kawk50m-ko-instruct-v1/checksums.sha256 +2 -0
- reports/kawk50m-ko-instruct-v1/evaluation.json +56 -0
- reports/kawk50m-ko-instruct-v1/final_checkpoint.txt +1 -0
- reports/kawk50m-ko-instruct-v1/gate.json +7 -0
- reports/v2-100m/train.log +101 -0
- reports/v2-100m/v0-byte-evaluation.json +17 -0
- reports/v2-100m/v0-evaluation.json +30 -0
- reports/v2-100m/v2-byte-evaluation.json +17 -0
- reports/v2-100m/v2-evaluation.json +29 -0
- reports/v2-20b/train.log +0 -0
- reports/v2-20b/v0-5b-byte-evaluation.json +17 -0
- reports/v2-20b/v2-20b-byte-evaluation.json +17 -0
- reports/v2-20b/v2-20b-evaluation.json +30 -0
- reports/v2b-100m/train.log +101 -0
- reports/v2b-100m/v0-byte-evaluation.json +17 -0
- reports/v2b-100m/v0-on-v2b-evaluation.json +30 -0
- reports/v2b-100m/v2b-byte-evaluation.json +17 -0
- reports/v2b-100m/v2b-evaluation.json +30 -0
- reports/v2c-100m/train.log +101 -0
- reports/v2c-100m/v2c-byte-evaluation.json +17 -0
- reports/v2c-100m/v2c-evaluation.json +30 -0
.gitignore
ADDED
|
@@ -0,0 +1,23 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
__pycache__/
|
| 2 |
+
*.py[cod]
|
| 3 |
+
.pytest_cache/
|
| 4 |
+
.ruff_cache/
|
| 5 |
+
.venv/
|
| 6 |
+
venv/
|
| 7 |
+
.cache/
|
| 8 |
+
|
| 9 |
+
# Local corpora and generated artifacts
|
| 10 |
+
data/raw/*
|
| 11 |
+
data/clean/*
|
| 12 |
+
data/tokenized/*
|
| 13 |
+
artifacts/*
|
| 14 |
+
runs/*
|
| 15 |
+
!data/raw/.gitkeep
|
| 16 |
+
!data/clean/.gitkeep
|
| 17 |
+
!data/tokenized/.gitkeep
|
| 18 |
+
!artifacts/.gitkeep
|
| 19 |
+
!runs/.gitkeep
|
| 20 |
+
|
| 21 |
+
# Temporary files
|
| 22 |
+
*.tmp
|
| 23 |
+
*.log
|
artifacts/tokenizer-20k-v2/training_metadata.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"type": "SentencePiece Unigram",
|
| 3 |
+
"vocab_size": 20000,
|
| 4 |
+
"normalization": "external NFC + SentencePiece identity",
|
| 5 |
+
"byte_fallback": true,
|
| 6 |
+
"split_digits": true,
|
| 7 |
+
"training_lines": 1999729,
|
| 8 |
+
"reused_sentencepiece_model": true,
|
| 9 |
+
"model_max_length": 5120,
|
| 10 |
+
"special_token_ids": {
|
| 11 |
+
"unk": 0,
|
| 12 |
+
"bos": 1,
|
| 13 |
+
"eos": 2,
|
| 14 |
+
"pad": 3
|
| 15 |
+
},
|
| 16 |
+
"probes": [
|
| 17 |
+
{
|
| 18 |
+
"text": "안녕하세요. 오늘 날씨가 좋습니다.",
|
| 19 |
+
"tokens": 7,
|
| 20 |
+
"unknown_tokens": 0,
|
| 21 |
+
"decoded": "안녕하세요. 오늘 날씨가 좋습니다.",
|
| 22 |
+
"exact_round_trip": true
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"text": "이거 진짜 되는 거 맞아? ㅋㅋㅋㅋ",
|
| 26 |
+
"tokens": 12,
|
| 27 |
+
"unknown_tokens": 0,
|
| 28 |
+
"decoded": "이거 진짜 되는 거 맞아? ㅋㅋㅋㅋ",
|
| 29 |
+
"exact_round_trip": true
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"text": "갤럭시 S26을 2026년에 구매했습니다.",
|
| 33 |
+
"tokens": 14,
|
| 34 |
+
"unknown_tokens": 0,
|
| 35 |
+
"decoded": "갤럭시 S26을 2026년에 구매했습니다.",
|
| 36 |
+
"exact_round_trip": true
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"text": "KTX를 타고 부산에 내려갔다. 🚄",
|
| 40 |
+
"tokens": 14,
|
| 41 |
+
"unknown_tokens": 0,
|
| 42 |
+
"decoded": "KTX를 타고 부산에 내려갔다. 🚄",
|
| 43 |
+
"exact_round_trip": true
|
| 44 |
+
}
|
| 45 |
+
]
|
| 46 |
+
}
|
artifacts/tokenizer-20k-v3/special_tokens_map.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token": {
|
| 3 |
+
"content": "<s>",
|
| 4 |
+
"lstrip": false,
|
| 5 |
+
"normalized": false,
|
| 6 |
+
"rstrip": false,
|
| 7 |
+
"single_word": false
|
| 8 |
+
},
|
| 9 |
+
"eos_token": {
|
| 10 |
+
"content": "</s>",
|
| 11 |
+
"lstrip": false,
|
| 12 |
+
"normalized": false,
|
| 13 |
+
"rstrip": false,
|
| 14 |
+
"single_word": false
|
| 15 |
+
},
|
| 16 |
+
"pad_token": {
|
| 17 |
+
"content": "<pad>",
|
| 18 |
+
"lstrip": false,
|
| 19 |
+
"normalized": false,
|
| 20 |
+
"rstrip": false,
|
| 21 |
+
"single_word": false
|
| 22 |
+
},
|
| 23 |
+
"unk_token": {
|
| 24 |
+
"content": "<unk>",
|
| 25 |
+
"lstrip": false,
|
| 26 |
+
"normalized": false,
|
| 27 |
+
"rstrip": false,
|
| 28 |
+
"single_word": false
|
| 29 |
+
}
|
| 30 |
+
}
|
artifacts/tokenizer-20k-v3/tokenizer.vocab
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
artifacts/tokenizer-20k-v3/tokenizer_config.json
ADDED
|
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_bos_token": true,
|
| 3 |
+
"add_eos_token": false,
|
| 4 |
+
"add_prefix_space": true,
|
| 5 |
+
"added_tokens_decoder": {
|
| 6 |
+
"0": {
|
| 7 |
+
"content": "<unk>",
|
| 8 |
+
"lstrip": false,
|
| 9 |
+
"normalized": false,
|
| 10 |
+
"rstrip": false,
|
| 11 |
+
"single_word": false,
|
| 12 |
+
"special": true
|
| 13 |
+
},
|
| 14 |
+
"1": {
|
| 15 |
+
"content": "<s>",
|
| 16 |
+
"lstrip": false,
|
| 17 |
+
"normalized": false,
|
| 18 |
+
"rstrip": false,
|
| 19 |
+
"single_word": false,
|
| 20 |
+
"special": true
|
| 21 |
+
},
|
| 22 |
+
"2": {
|
| 23 |
+
"content": "</s>",
|
| 24 |
+
"lstrip": false,
|
| 25 |
+
"normalized": false,
|
| 26 |
+
"rstrip": false,
|
| 27 |
+
"single_word": false,
|
| 28 |
+
"special": true
|
| 29 |
+
},
|
| 30 |
+
"3": {
|
| 31 |
+
"content": "<pad>",
|
| 32 |
+
"lstrip": false,
|
| 33 |
+
"normalized": false,
|
| 34 |
+
"rstrip": false,
|
| 35 |
+
"single_word": false,
|
| 36 |
+
"special": true
|
| 37 |
+
}
|
| 38 |
+
},
|
| 39 |
+
"bos_token": "<s>",
|
| 40 |
+
"clean_up_tokenization_spaces": false,
|
| 41 |
+
"eos_token": "</s>",
|
| 42 |
+
"extra_special_tokens": {},
|
| 43 |
+
"legacy": true,
|
| 44 |
+
"model_max_length": 5120,
|
| 45 |
+
"pad_token": "<pad>",
|
| 46 |
+
"sp_model_kwargs": {},
|
| 47 |
+
"spaces_between_special_tokens": false,
|
| 48 |
+
"tokenizer_class": "LlamaTokenizer",
|
| 49 |
+
"unk_token": "<unk>",
|
| 50 |
+
"use_default_system_prompt": false
|
| 51 |
+
}
|
artifacts/tokenizer-20k-v3/training_metadata.json
ADDED
|
@@ -0,0 +1,46 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"type": "SentencePiece Unigram",
|
| 3 |
+
"vocab_size": 20000,
|
| 4 |
+
"normalization": "external NFC + SentencePiece identity",
|
| 5 |
+
"byte_fallback": true,
|
| 6 |
+
"split_digits": true,
|
| 7 |
+
"training_lines": 1000000,
|
| 8 |
+
"reused_sentencepiece_model": true,
|
| 9 |
+
"model_max_length": 5120,
|
| 10 |
+
"special_token_ids": {
|
| 11 |
+
"unk": 0,
|
| 12 |
+
"bos": 1,
|
| 13 |
+
"eos": 2,
|
| 14 |
+
"pad": 3
|
| 15 |
+
},
|
| 16 |
+
"probes": [
|
| 17 |
+
{
|
| 18 |
+
"text": "안녕하세요. 오늘 날씨가 좋습니다.",
|
| 19 |
+
"tokens": 7,
|
| 20 |
+
"unknown_tokens": 0,
|
| 21 |
+
"decoded": "안녕하세요. 오늘 날씨가 좋습니다.",
|
| 22 |
+
"exact_round_trip": true
|
| 23 |
+
},
|
| 24 |
+
{
|
| 25 |
+
"text": "이거 진짜 되는 거 맞아? ㅋㅋㅋㅋ",
|
| 26 |
+
"tokens": 8,
|
| 27 |
+
"unknown_tokens": 0,
|
| 28 |
+
"decoded": "이거 진짜 되는 거 맞아? ㅋㅋㅋㅋ",
|
| 29 |
+
"exact_round_trip": true
|
| 30 |
+
},
|
| 31 |
+
{
|
| 32 |
+
"text": "갤럭시 S26을 2026년에 구매했습니다.",
|
| 33 |
+
"tokens": 14,
|
| 34 |
+
"unknown_tokens": 0,
|
| 35 |
+
"decoded": "갤럭시 S26을 2026년에 구매했습니다.",
|
| 36 |
+
"exact_round_trip": true
|
| 37 |
+
},
|
| 38 |
+
{
|
| 39 |
+
"text": "KTX를 타고 부산에 내려갔다. 🚄",
|
| 40 |
+
"tokens": 14,
|
| 41 |
+
"unknown_tokens": 0,
|
| 42 |
+
"decoded": "KTX를 타고 부산에 내려갔다. 🚄",
|
| 43 |
+
"exact_round_trip": true
|
| 44 |
+
}
|
| 45 |
+
]
|
| 46 |
+
}
|
reports/fixed-ko-cpt-3b-5k/checksums.sha256
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
96802581ebf61151c7ec793b3c59bfeb7bc978b3208ba22da97417cc91fa290e model/model.safetensors
|
| 2 |
+
bb00c657506b2be0f78e026b06936cfa57a4f65bea71b0261978143cf1c125f2 trainer_state.pt
|
reports/fixed-objective-100m-ab/broken-final-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415/tokenizer",
|
| 4 |
+
"validation_loss": 7.156057047843933,
|
| 5 |
+
"validation_perplexity": 1281.8466952527854,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "이다\n 월산은에서서원가도보-선장\n산초사후신-원명지고-성수이-수-경강-암동-대-천--\n-()\n리산구동산삼리정\n210.85",
|
| 11 |
+
"full_text": "대한민국의 수도는이다\n 월산은에서서원가도보-선장\n산초사후신-원명지고-성수이-수-경강-암동-대-천--\n-()\n리산구동산삼리정\n210.85"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "0:5 를면102\n가히, 9,78나? 36.%이\n. :16%.:12%; 4.%,2%:.2%\n\n159 4 3 ",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서0:5 를면102\n가히, 9,78나? 36.%이\n. :16%.:12%; 4.%,2%:.2%\n\n159 4 3 "
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "의을하고. (3)어티브는11 의을이,2 , :0",
|
| 21 |
+
"full_text": "인공지능 언어 모델은의을하고. (3)어티브는11 의을이,2 , :0"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "에이 적 그의 를 . :=20%",
|
| 26 |
+
"full_text": "비가 내리는 날에는에이 적 그의 를 . :=20%"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/fixed-objective-100m-ab/recovery-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 200,
|
| 8 |
+
"characters": 316443,
|
| 9 |
+
"utf8_bytes": 753034,
|
| 10 |
+
"predicted_tokens": 168405,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 3.1995365798935245,
|
| 13 |
+
"token_perplexity": 24.521163963227362,
|
| 14 |
+
"nats_per_character": 1.7027330600991932,
|
| 15 |
+
"bits_per_utf8_byte": 1.0322907007771818,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/fixed-objective-100m-ab/recovery-checkpoint.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763
|
reports/fixed-objective-100m-ab/recovery-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"validation_loss": 3.166912519931793,
|
| 5 |
+
"validation_perplexity": 23.734092579129626,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "그 어느 곳보다도 넓은 도시입니다.\n하지만, 도심에 위치한 도시들은 먼 거리에 있는 도시들과 비교하면 많은 차이가 있습니다.\n바로, 고속도로가 있습니다.\n이렇게 고속도로를 이용하지 않아도 고속도로에서 쉽게 서울을 볼 수 있습니다.\n오늘 소개해드릴 곳은 고속도로를 이용한 고성읍에 위치한 '송정' 입니다.\n저는 송정 고속도로를 이용했는데, 고속도로에서 자동차로 1",
|
| 11 |
+
"full_text": "대한민국의 수도는그 어느 곳보다도 넓은 도시입니다.\n하지만, 도심에 위치한 도시들은 먼 거리에 있는 도시들과 비교하면 많은 차이가 있습니다.\n바로, 고속도로가 있습니다.\n이렇게 고속도로를 이용하지 않아도 고속도로에서 쉽게 서울을 볼 수 있습니다.\n오늘 소개해드릴 곳은 고속도로를 이용한 고성읍에 위치한 '송정' 입니다.\n저는 송정 고속도로를 이용했는데, 고속도로에서 자동차로 1"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "12시에 퇴근해서, 그리고 오후 4시에 출근해서 저녁을 먹었는데도, 10시가 넘어서 집에 왔더라구요.\n그렇게 9시가 되어서야, 다시 일어나서 7시에 집에 와서 밥을 먹고 5시쯤에 퇴근했는데, 저녁에 일어나서 밥을 먹는 것이 더 낫겠더라구요.\n그래서, 6시에 퇴근한 후에 집에 왔더니, 저녁",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서12시에 퇴근해서, 그리고 오후 4시에 출근해서 저녁을 먹었는데도, 10시가 넘어서 집에 왔더라구요.\n그렇게 9시가 되어서야, 다시 일어나서 7시에 집에 와서 밥을 먹고 5시쯤에 퇴근했는데, 저녁에 일어나서 밥을 먹는 것이 더 낫겠더라구요.\n그래서, 6시에 퇴근한 후에 집에 왔더니, 저녁"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "텍스트의 구조를 단순화하고, 텍스트에 대한 데이터 분석을 통해 텍스트가 가진 객체적 의미를 분석하는 데 도움이 될 수 있습니다.\n이는 다양한 워크플로우와 데이터를 처리할 수 있는 고도의 기능을 갖춘 언어 언어 모델에서 더 빠르게 구현되고 있습니다. 워크플로우는 데이터를 보다 효율적이고 효율적으로 사용할 수 있도록 하기 위해 고도로 복잡한 구조를 사용하고 있습니다. 워크플로우를 사용하면 모든 워크플로우에 대한",
|
| 21 |
+
"full_text": "인공지능 언어 모델은텍스트의 구조를 단순화하고, 텍스트에 대한 데이터 분석을 통해 텍스트가 가진 객체적 의미를 분석하는 데 도움이 될 수 있습니다.\n이는 다양한 워크플로우와 데이터를 처리할 수 있는 고도의 기능을 갖춘 언어 언어 모델에서 더 빠르게 구현되고 있습니다. 워크플로우는 데이터를 보다 효율적이고 효율적으로 사용할 수 있도록 하기 위해 고도로 복잡한 구조를 사용하고 있습니다. 워크플로우를 사용하면 모든 워크플로우에 대한"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "햇빛을 많이 받고, 아침에 비가 오면 눈이 내린다.\n오후에는 기온이 급격히 내려가는데, 특히 겨울은 10도 내외로 오르고, 이 때문에 겨울에 비가 온다면 산으로 들어가야 한다.\n산행은 5월 중순경에 시작하고, 6월부터 8월까지는 1년 내내 운행된다.\n산행시간은 산행과는 다르게, 2주일 정도 소요",
|
| 26 |
+
"full_text": "비가 내리는 날에는햇빛을 많이 받고, 아침에 비가 오면 눈이 내린다.\n오후에는 기온이 급격히 내려가는데, 특히 겨울은 10도 내외로 오르고, 이 때문에 겨울에 비가 온다면 산으로 들어가야 한다.\n산행은 5월 중순경에 시작하고, 6월부터 8월까지는 1년 내내 운행된다.\n산행시간은 산행과는 다르게, 2주일 정도 소요"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/fixed-objective-100m-ab/recovery-train.log
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"tokens_per_optimizer_step": 131072,
|
| 7 |
+
"max_steps": 763,
|
| 8 |
+
"planned_tokens": 100007936,
|
| 9 |
+
"start_step": 0,
|
| 10 |
+
"initialized_from": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415",
|
| 11 |
+
"precision": "bf16",
|
| 12 |
+
"compile": true
|
| 13 |
+
}
|
| 14 |
+
{"step": 1, "max_steps": 763, "train_loss": 9.81873, "learning_rate": 3.9999999999999996e-05, "grad_norm": 99.39525604248047, "tokens_seen": 131072, "tokens_per_second": 40748}
|
| 15 |
+
{"step": 10, "max_steps": 763, "train_loss": 4.48067, "learning_rate": 0.00039999999999999996, "grad_norm": 6.562530040740967, "tokens_seen": 1310720, "tokens_per_second": 266763}
|
| 16 |
+
{"step": 20, "max_steps": 763, "train_loss": 3.99055, "learning_rate": 0.0005999404674662951, "grad_norm": 2.6631991863250732, "tokens_seen": 2621440, "tokens_per_second": 266620}
|
| 17 |
+
{"step": 30, "max_steps": 763, "train_loss": 3.67793, "learning_rate": 0.0005994643647016376, "grad_norm": 1.6457377672195435, "tokens_seen": 3932160, "tokens_per_second": 266618}
|
| 18 |
+
{"step": 40, "max_steps": 763, "train_loss": 3.564066, "learning_rate": 0.0005985129988902226, "grad_norm": 1.066536545753479, "tokens_seen": 5242880, "tokens_per_second": 266597}
|
| 19 |
+
{"step": 50, "max_steps": 763, "train_loss": 3.556779, "learning_rate": 0.000597088047986813, "grad_norm": 0.8559065461158752, "tokens_seen": 6553600, "tokens_per_second": 266575}
|
| 20 |
+
{"step": 60, "max_steps": 763, "train_loss": 3.543101, "learning_rate": 0.0005951920252235705, "grad_norm": 0.7963358759880066, "tokens_seen": 7864320, "tokens_per_second": 266509}
|
| 21 |
+
{"step": 70, "max_steps": 763, "train_loss": 3.484368, "learning_rate": 0.0005928282746773871, "grad_norm": 0.812751829624176, "tokens_seen": 9175040, "tokens_per_second": 266514}
|
| 22 |
+
{"step": 80, "max_steps": 763, "train_loss": 3.431172, "learning_rate": 0.0005900009653718274, "grad_norm": 0.7371594905853271, "tokens_seen": 10485760, "tokens_per_second": 266493}
|
| 23 |
+
{"step": 90, "max_steps": 763, "train_loss": 3.462744, "learning_rate": 0.0005867150839240876, "grad_norm": 0.7278872728347778, "tokens_seen": 11796480, "tokens_per_second": 266500}
|
| 24 |
+
{"step": 100, "max_steps": 763, "train_loss": 3.360779, "learning_rate": 0.0005829764257499357, "grad_norm": 0.6683600544929504, "tokens_seen": 13107200, "tokens_per_second": 266463}
|
| 25 |
+
{"step": 100, "validation_loss": 3.474578, "validation_perplexity": 32.2842}
|
| 26 |
+
{"step": 110, "max_steps": 763, "train_loss": 3.439466, "learning_rate": 0.0005787915848421489, "grad_norm": 0.6720628142356873, "tokens_seen": 14417920, "tokens_per_second": 154865}
|
| 27 |
+
{"step": 120, "max_steps": 763, "train_loss": 3.419731, "learning_rate": 0.0005741679421404754, "grad_norm": 0.5747093558311462, "tokens_seen": 15728640, "tokens_per_second": 266533}
|
| 28 |
+
{"step": 130, "max_steps": 763, "train_loss": 3.401971, "learning_rate": 0.0005691136525136301, "grad_norm": 0.7317373156547546, "tokens_seen": 17039360, "tokens_per_second": 266504}
|
| 29 |
+
{"step": 140, "max_steps": 763, "train_loss": 3.342753, "learning_rate": 0.0005636376303762884, "grad_norm": 0.6881513595581055, "tokens_seen": 18350080, "tokens_per_second": 266487}
|
| 30 |
+
{"step": 150, "max_steps": 763, "train_loss": 3.34088, "learning_rate": 0.0005577495339664442, "grad_norm": 0.6463481783866882, "tokens_seen": 19660800, "tokens_per_second": 266514}
|
| 31 |
+
{"step": 160, "max_steps": 763, "train_loss": 3.381509, "learning_rate": 0.0005514597483108611, "grad_norm": 0.6442299485206604, "tokens_seen": 20971520, "tokens_per_second": 266490}
|
| 32 |
+
{"step": 170, "max_steps": 763, "train_loss": 3.386218, "learning_rate": 0.0005447793669086647, "grad_norm": 0.6932205557823181, "tokens_seen": 22282240, "tokens_per_second": 266502}
|
| 33 |
+
{"step": 180, "max_steps": 763, "train_loss": 3.288735, "learning_rate": 0.0005377201721653773, "grad_norm": 0.6026113033294678, "tokens_seen": 23592960, "tokens_per_second": 266507}
|
| 34 |
+
{"step": 190, "max_steps": 763, "train_loss": 3.293852, "learning_rate": 0.0005302946146119068, "grad_norm": 0.6506741046905518, "tokens_seen": 24903680, "tokens_per_second": 266498}
|
| 35 |
+
{"step": 200, "max_steps": 763, "train_loss": 3.332148, "learning_rate": 0.0005225157909451416, "grad_norm": 0.6899046897888184, "tokens_seen": 26214400, "tokens_per_second": 266500}
|
| 36 |
+
{"step": 200, "validation_loss": 3.389528, "validation_perplexity": 29.6519}
|
| 37 |
+
{"step": 210, "max_steps": 763, "train_loss": 3.368893, "learning_rate": 0.0005143974209288811, "grad_norm": 0.642666757106781, "tokens_seen": 27525120, "tokens_per_second": 230532}
|
| 38 |
+
{"step": 220, "max_steps": 763, "train_loss": 3.322338, "learning_rate": 0.0005059538231958432, "grad_norm": 0.6120380759239197, "tokens_seen": 28835840, "tokens_per_second": 266469}
|
| 39 |
+
{"step": 230, "max_steps": 763, "train_loss": 3.360906, "learning_rate": 0.0004971998899934259, "grad_norm": 0.5670522451400757, "tokens_seen": 30146560, "tokens_per_second": 266529}
|
| 40 |
+
{"step": 240, "max_steps": 763, "train_loss": 3.273283, "learning_rate": 0.00048815106091776853, "grad_norm": 0.5588228702545166, "tokens_seen": 31457280, "tokens_per_second": 266447}
|
| 41 |
+
{"step": 250, "max_steps": 763, "train_loss": 3.262542, "learning_rate": 0.0004788232956824341, "grad_norm": 0.5968481302261353, "tokens_seen": 32768000, "tokens_per_second": 266511}
|
| 42 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000250
|
| 43 |
+
{"step": 260, "max_steps": 763, "train_loss": 3.298922, "learning_rate": 0.00046923304596974483, "grad_norm": 0.6154866218566895, "tokens_seen": 34078720, "tokens_per_second": 231703}
|
| 44 |
+
{"step": 270, "max_steps": 763, "train_loss": 3.334789, "learning_rate": 0.0004593972264144153, "grad_norm": 0.5695328712463379, "tokens_seen": 35389440, "tokens_per_second": 266517}
|
| 45 |
+
{"step": 280, "max_steps": 763, "train_loss": 3.303071, "learning_rate": 0.00044933318477066334, "grad_norm": 0.565549373626709, "tokens_seen": 36700160, "tokens_per_second": 266490}
|
| 46 |
+
{"step": 290, "max_steps": 763, "train_loss": 3.290458, "learning_rate": 0.00043905867131541155, "grad_norm": 0.6306836605072021, "tokens_seen": 38010880, "tokens_per_second": 266467}
|
| 47 |
+
{"step": 300, "max_steps": 763, "train_loss": 3.3301, "learning_rate": 0.0004285918075415489, "grad_norm": 0.543013870716095, "tokens_seen": 39321600, "tokens_per_second": 266458}
|
| 48 |
+
{"step": 300, "validation_loss": 3.328608, "validation_perplexity": 27.8995}
|
| 49 |
+
{"step": 310, "max_steps": 763, "train_loss": 3.272506, "learning_rate": 0.00041795105419646444, "grad_norm": 0.5561285018920898, "tokens_seen": 40632320, "tokens_per_second": 230517}
|
| 50 |
+
{"step": 320, "max_steps": 763, "train_loss": 3.270318, "learning_rate": 0.00040715517872222867, "grad_norm": 0.5325157046318054, "tokens_seen": 41943040, "tokens_per_second": 266464}
|
| 51 |
+
{"step": 330, "max_steps": 763, "train_loss": 3.25808, "learning_rate": 0.0003962232221548463, "grad_norm": 0.5173934698104858, "tokens_seen": 43253760, "tokens_per_second": 266437}
|
| 52 |
+
{"step": 340, "max_steps": 763, "train_loss": 3.326208, "learning_rate": 0.00038517446554096337, "grad_norm": 0.5639611482620239, "tokens_seen": 44564480, "tokens_per_second": 266458}
|
| 53 |
+
{"step": 350, "max_steps": 763, "train_loss": 3.261147, "learning_rate": 0.0003740283959312592, "grad_norm": 0.5813310742378235, "tokens_seen": 45875200, "tokens_per_second": 266416}
|
| 54 |
+
{"step": 360, "max_steps": 763, "train_loss": 3.335487, "learning_rate": 0.0003628046720105023, "grad_norm": 0.5882489681243896, "tokens_seen": 47185920, "tokens_per_second": 266475}
|
| 55 |
+
{"step": 370, "max_steps": 763, "train_loss": 3.267799, "learning_rate": 0.00035152308942489074, "grad_norm": 0.5271080136299133, "tokens_seen": 48496640, "tokens_per_second": 265994}
|
| 56 |
+
{"step": 380, "max_steps": 763, "train_loss": 3.212291, "learning_rate": 0.0003402035458678273, "grad_norm": 0.5357783436775208, "tokens_seen": 49807360, "tokens_per_second": 266461}
|
| 57 |
+
{"step": 390, "max_steps": 763, "train_loss": 3.270362, "learning_rate": 0.00032886600598571277, "grad_norm": 0.5579140782356262, "tokens_seen": 51118080, "tokens_per_second": 266475}
|
| 58 |
+
{"step": 400, "max_steps": 763, "train_loss": 3.212604, "learning_rate": 0.0003175304661656503, "grad_norm": 0.5288333296775818, "tokens_seen": 52428800, "tokens_per_second": 266489}
|
| 59 |
+
{"step": 400, "validation_loss": 3.279331, "validation_perplexity": 26.558}
|
| 60 |
+
{"step": 410, "max_steps": 763, "train_loss": 3.176318, "learning_rate": 0.00030621691926716884, "grad_norm": 0.5217863917350769, "tokens_seen": 53739520, "tokens_per_second": 230546}
|
| 61 |
+
{"step": 420, "max_steps": 763, "train_loss": 3.284995, "learning_rate": 0.0002949453193601676, "grad_norm": 0.5939567685127258, "tokens_seen": 55050240, "tokens_per_second": 266487}
|
| 62 |
+
{"step": 430, "max_steps": 763, "train_loss": 3.218869, "learning_rate": 0.0002837355465312754, "grad_norm": 0.5420132875442505, "tokens_seen": 56360960, "tokens_per_second": 266497}
|
| 63 |
+
{"step": 440, "max_steps": 763, "train_loss": 3.240979, "learning_rate": 0.00027260737182069626, "grad_norm": 0.5817392468452454, "tokens_seen": 57671680, "tokens_per_second": 266446}
|
| 64 |
+
{"step": 450, "max_steps": 763, "train_loss": 3.280826, "learning_rate": 0.00026158042235138474, "grad_norm": 0.5186812281608582, "tokens_seen": 58982400, "tokens_per_second": 266503}
|
| 65 |
+
{"step": 460, "max_steps": 763, "train_loss": 3.186866, "learning_rate": 0.00025067414671205103, "grad_norm": 0.5333555936813354, "tokens_seen": 60293120, "tokens_per_second": 266445}
|
| 66 |
+
{"step": 470, "max_steps": 763, "train_loss": 3.151642, "learning_rate": 0.00023990778065505504, "grad_norm": 0.5093833208084106, "tokens_seen": 61603840, "tokens_per_second": 266468}
|
| 67 |
+
{"step": 480, "max_steps": 763, "train_loss": 3.140028, "learning_rate": 0.00022930031316968416, "grad_norm": 0.4820874333381653, "tokens_seen": 62914560, "tokens_per_second": 266426}
|
| 68 |
+
{"step": 490, "max_steps": 763, "train_loss": 3.208196, "learning_rate": 0.00021887045299065797, "grad_norm": 0.45261216163635254, "tokens_seen": 64225280, "tokens_per_second": 266473}
|
| 69 |
+
{"step": 500, "max_steps": 763, "train_loss": 3.252856, "learning_rate": 0.00020863659560092397, "grad_norm": 0.489402711391449, "tokens_seen": 65536000, "tokens_per_second": 266474}
|
| 70 |
+
{"step": 500, "validation_loss": 3.234741, "validation_perplexity": 25.3998}
|
| 71 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000500
|
| 72 |
+
{"step": 510, "max_steps": 763, "train_loss": 3.233288, "learning_rate": 0.00019861679078694882, "grad_norm": 0.4858289659023285, "tokens_seen": 66846720, "tokens_per_second": 208992}
|
| 73 |
+
{"step": 520, "max_steps": 763, "train_loss": 3.203979, "learning_rate": 0.0001888287108037225, "grad_norm": 0.45732933282852173, "tokens_seen": 68157440, "tokens_per_second": 266464}
|
| 74 |
+
{"step": 530, "max_steps": 763, "train_loss": 3.225087, "learning_rate": 0.00017928961920562968, "grad_norm": 0.48645803332328796, "tokens_seen": 69468160, "tokens_per_second": 266478}
|
| 75 |
+
{"step": 540, "max_steps": 763, "train_loss": 3.195829, "learning_rate": 0.000170016340398158, "grad_norm": 0.48128387331962585, "tokens_seen": 70778880, "tokens_per_second": 266481}
|
| 76 |
+
{"step": 550, "max_steps": 763, "train_loss": 3.198164, "learning_rate": 0.00016102522996414806, "grad_norm": 0.4550345540046692, "tokens_seen": 72089600, "tokens_per_second": 266479}
|
| 77 |
+
{"step": 560, "max_steps": 763, "train_loss": 3.163047, "learning_rate": 0.00015233214581692112, "grad_norm": 0.4392969608306885, "tokens_seen": 73400320, "tokens_per_second": 266462}
|
| 78 |
+
{"step": 570, "max_steps": 763, "train_loss": 3.121977, "learning_rate": 0.0001439524202311625, "grad_norm": 0.4583962559700012, "tokens_seen": 74711040, "tokens_per_second": 266494}
|
| 79 |
+
{"step": 580, "max_steps": 763, "train_loss": 3.18189, "learning_rate": 0.0001359008328008913, "grad_norm": 0.44599199295043945, "tokens_seen": 76021760, "tokens_per_second": 266501}
|
| 80 |
+
{"step": 590, "max_steps": 763, "train_loss": 3.262956, "learning_rate": 0.0001281915843722103, "grad_norm": 0.4535924792289734, "tokens_seen": 77332480, "tokens_per_second": 266502}
|
| 81 |
+
{"step": 600, "max_steps": 763, "train_loss": 3.199665, "learning_rate": 0.00012083827199681319, "grad_norm": 0.471802681684494, "tokens_seen": 78643200, "tokens_per_second": 266504}
|
| 82 |
+
{"step": 600, "validation_loss": 3.197571, "validation_perplexity": 24.473}
|
| 83 |
+
{"step": 610, "max_steps": 763, "train_loss": 3.197196, "learning_rate": 0.00011385386495042271, "grad_norm": 0.46694111824035645, "tokens_seen": 79953920, "tokens_per_second": 230510}
|
| 84 |
+
{"step": 620, "max_steps": 763, "train_loss": 3.145936, "learning_rate": 0.0001072506818584584, "grad_norm": 0.44446995854377747, "tokens_seen": 81264640, "tokens_per_second": 266448}
|
| 85 |
+
{"step": 630, "max_steps": 763, "train_loss": 3.109152, "learning_rate": 0.00010104036896927688, "grad_norm": 0.4449124038219452, "tokens_seen": 82575360, "tokens_per_second": 266447}
|
| 86 |
+
{"step": 640, "max_steps": 763, "train_loss": 3.200131, "learning_rate": 9.523387961330576e-05, "grad_norm": 0.4285873472690582, "tokens_seen": 83886080, "tokens_per_second": 266482}
|
| 87 |
+
{"step": 650, "max_steps": 763, "train_loss": 3.162688, "learning_rate": 8.984145488429902e-05, "grad_norm": 0.4384717047214508, "tokens_seen": 85196800, "tokens_per_second": 266425}
|
| 88 |
+
{"step": 660, "max_steps": 763, "train_loss": 3.145082, "learning_rate": 8.487260557678771e-05, "grad_norm": 0.45577767491340637, "tokens_seen": 86507520, "tokens_per_second": 266448}
|
| 89 |
+
{"step": 670, "max_steps": 763, "train_loss": 3.160741, "learning_rate": 8.033609541158293e-05, "grad_norm": 0.4359270930290222, "tokens_seen": 87818240, "tokens_per_second": 266461}
|
| 90 |
+
{"step": 680, "max_steps": 763, "train_loss": 3.109333, "learning_rate": 7.623992557891675e-05, "grad_norm": 0.43799203634262085, "tokens_seen": 89128960, "tokens_per_second": 266478}
|
| 91 |
+
{"step": 690, "max_steps": 763, "train_loss": 3.089035, "learning_rate": 7.25913206264835e-05, "grad_norm": 0.4514983296394348, "tokens_seen": 90439680, "tokens_per_second": 266450}
|
| 92 |
+
{"step": 700, "max_steps": 763, "train_loss": 3.113999, "learning_rate": 6.939671571727054e-05, "grad_norm": 0.42705073952674866, "tokens_seen": 91750400, "tokens_per_second": 266473}
|
| 93 |
+
{"step": 700, "validation_loss": 3.174568, "validation_perplexity": 23.9165}
|
| 94 |
+
{"step": 710, "max_steps": 763, "train_loss": 3.087399, "learning_rate": 6.6661745279652e-05, "grad_norm": 0.4240090250968933, "tokens_seen": 93061120, "tokens_per_second": 230552}
|
| 95 |
+
{"step": 720, "max_steps": 763, "train_loss": 3.10266, "learning_rate": 6.439123306976511e-05, "grad_norm": 0.42993783950805664, "tokens_seen": 94371840, "tokens_per_second": 266422}
|
| 96 |
+
{"step": 730, "max_steps": 763, "train_loss": 3.101989, "learning_rate": 6.258918366369462e-05, "grad_norm": 0.41898486018180847, "tokens_seen": 95682560, "tokens_per_second": 266449}
|
| 97 |
+
{"step": 740, "max_steps": 763, "train_loss": 3.142001, "learning_rate": 6.125877539447194e-05, "grad_norm": 0.4237852990627289, "tokens_seen": 96993280, "tokens_per_second": 266455}
|
| 98 |
+
{"step": 750, "max_steps": 763, "train_loss": 3.12588, "learning_rate": 6.0402354746346246e-05, "grad_norm": 0.4214364290237427, "tokens_seen": 98304000, "tokens_per_second": 266516}
|
| 99 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000750
|
| 100 |
+
{"step": 760, "max_steps": 763, "train_loss": 3.104814, "learning_rate": 6.002143221621371e-05, "grad_norm": 0.41805922985076904, "tokens_seen": 99614720, "tokens_per_second": 235514}
|
| 101 |
+
Final checkpoint: runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763
|
reports/fixed-objective-100m-ab/scratch-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 200,
|
| 8 |
+
"characters": 316443,
|
| 9 |
+
"utf8_bytes": 753034,
|
| 10 |
+
"predicted_tokens": 168405,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 5.260973563947027,
|
| 13 |
+
"token_perplexity": 192.6689755868065,
|
| 14 |
+
"nats_per_character": 2.799790967208941,
|
| 15 |
+
"bits_per_utf8_byte": 1.6973877158415966,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/fixed-objective-100m-ab/scratch-checkpoint.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763
|
reports/fixed-objective-100m-ab/scratch-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"validation_loss": 5.223334383964539,
|
| 5 |
+
"validation_perplexity": 185.55185485204026,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "그 이상과 달리가 전하지 않는다.\n2018년 7월 6일, 이명지 34명. 그 날에도 “이렇게 저인”를 모은 중·중(29)을 지냈고, 신세국을 잃어내 수가 되었고, 2대만의 사량에 기행했다. 하지만, 또 한갖 대로 의연도가 하여, 유적이다.\n",
|
| 11 |
+
"full_text": "대한민국의 수도는그 이상과 달리가 전하지 않는다.\n2018년 7월 6일, 이명지 34명. 그 날에도 “이렇게 저인”를 모은 중·중(29)을 지냈고, 신세국을 잃어내 수가 되었고, 2대만의 사량에 기행했다. 하지만, 또 한갖 대로 의연도가 하여, 유적이다.\n"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "120% 가량 5만원에 4.65km를 맞이합니다.\n현재는 36시간 정도로 370점. 이 98%, 1997달러(37만2천 원)으로, 568년 1월 7일 34건부터는 12억 달러에 불과하다. 227억원은 500만6103억원을",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서120% 가량 5만원에 4.65km를 맞이합니다.\n현재는 36시간 정도로 370점. 이 98%, 1997달러(37만2천 원)으로, 568년 1월 7일 34건부터는 12억 달러에 불과하다. 227억원은 500만6103억원을"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "19일(25)에 걸쳐 공개된 ‘2047’와 3월 19일 오후 18시에 이어온 7개나 전차인으로 활동해 4·6회로 1위를 내려냈다.\n116년 1월 27일, 1월 16일에는, 2015년 3월 28일, 12월 1일",
|
| 21 |
+
"full_text": "인공지능 언어 모델은19일(25)에 걸쳐 공개된 ‘2047’와 3월 19일 오후 18시에 이어온 7개나 전차인으로 활동해 4·6회로 1위를 내려냈다.\n116년 1월 27일, 1월 16일에는, 2015년 3월 28일, 12월 1일"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "욕은 모두 다름이지 않게 느껴진다.\n1964년 5월, 2038년에 7년 동안 6년과 1997년 12월 18일에 10년 만에 100%의 절반으로 약 16.4%를 기록한 것으로 알려졌다.\n1988년 1951년대 이후 이 9%로 62억달러인",
|
| 26 |
+
"full_text": "비가 내리는 날에는욕은 모두 다름이지 않게 느껴진다.\n1964년 5월, 2038년에 7년 동안 6년과 1997년 12월 18일에 10년 만에 100%의 절반으로 약 16.4%를 기록한 것으로 알려졌다.\n1988년 1951년대 이후 이 9%로 62억달러인"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/fixed-objective-100m-ab/scratch-train.log
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"tokens_per_optimizer_step": 131072,
|
| 7 |
+
"max_steps": 763,
|
| 8 |
+
"planned_tokens": 100007936,
|
| 9 |
+
"start_step": 0,
|
| 10 |
+
"initialized_from": null,
|
| 11 |
+
"precision": "bf16",
|
| 12 |
+
"compile": true
|
| 13 |
+
}
|
| 14 |
+
{"step": 1, "max_steps": 763, "train_loss": 10.008974, "learning_rate": 3.9999999999999996e-05, "grad_norm": 1.463052749633789, "tokens_seen": 131072, "tokens_per_second": 47351}
|
| 15 |
+
{"step": 10, "max_steps": 763, "train_loss": 8.869761, "learning_rate": 0.00039999999999999996, "grad_norm": 1.3394861221313477, "tokens_seen": 1310720, "tokens_per_second": 266775}
|
| 16 |
+
{"step": 20, "max_steps": 763, "train_loss": 7.653388, "learning_rate": 0.0005999404674662951, "grad_norm": 0.4107801914215088, "tokens_seen": 2621440, "tokens_per_second": 266786}
|
| 17 |
+
{"step": 30, "max_steps": 763, "train_loss": 7.613149, "learning_rate": 0.0005994643647016376, "grad_norm": 0.3379242718219757, "tokens_seen": 3932160, "tokens_per_second": 266729}
|
| 18 |
+
{"step": 40, "max_steps": 763, "train_loss": 7.554898, "learning_rate": 0.0005985129988902226, "grad_norm": 0.17872604727745056, "tokens_seen": 5242880, "tokens_per_second": 266675}
|
| 19 |
+
{"step": 50, "max_steps": 763, "train_loss": 7.508032, "learning_rate": 0.000597088047986813, "grad_norm": 0.20312093198299408, "tokens_seen": 6553600, "tokens_per_second": 266651}
|
| 20 |
+
{"step": 60, "max_steps": 763, "train_loss": 7.410873, "learning_rate": 0.0005951920252235705, "grad_norm": 0.20989294350147247, "tokens_seen": 7864320, "tokens_per_second": 266513}
|
| 21 |
+
{"step": 70, "max_steps": 763, "train_loss": 7.219315, "learning_rate": 0.0005928282746773871, "grad_norm": 0.39200717210769653, "tokens_seen": 9175040, "tokens_per_second": 266367}
|
| 22 |
+
{"step": 80, "max_steps": 763, "train_loss": 7.092091, "learning_rate": 0.0005900009653718274, "grad_norm": 0.4393499493598938, "tokens_seen": 10485760, "tokens_per_second": 266317}
|
| 23 |
+
{"step": 90, "max_steps": 763, "train_loss": 6.9776, "learning_rate": 0.0005867150839240876, "grad_norm": 0.3227439224720001, "tokens_seen": 11796480, "tokens_per_second": 266271}
|
| 24 |
+
{"step": 100, "max_steps": 763, "train_loss": 6.867079, "learning_rate": 0.0005829764257499357, "grad_norm": 0.5284335613250732, "tokens_seen": 13107200, "tokens_per_second": 266277}
|
| 25 |
+
{"step": 100, "validation_loss": 6.866146, "validation_perplexity": 959.2445}
|
| 26 |
+
{"step": 110, "max_steps": 763, "train_loss": 6.729001, "learning_rate": 0.0005787915848421489, "grad_norm": 0.43351173400878906, "tokens_seen": 14417920, "tokens_per_second": 155093}
|
| 27 |
+
{"step": 120, "max_steps": 763, "train_loss": 6.652469, "learning_rate": 0.0005741679421404754, "grad_norm": 0.47645148634910583, "tokens_seen": 15728640, "tokens_per_second": 266340}
|
| 28 |
+
{"step": 130, "max_steps": 763, "train_loss": 6.556467, "learning_rate": 0.0005691136525136301, "grad_norm": 0.3506828546524048, "tokens_seen": 17039360, "tokens_per_second": 266332}
|
| 29 |
+
{"step": 140, "max_steps": 763, "train_loss": 6.478513, "learning_rate": 0.0005636376303762884, "grad_norm": 0.5026987791061401, "tokens_seen": 18350080, "tokens_per_second": 266288}
|
| 30 |
+
{"step": 150, "max_steps": 763, "train_loss": 6.418608, "learning_rate": 0.0005577495339664442, "grad_norm": 1.4543246030807495, "tokens_seen": 19660800, "tokens_per_second": 266306}
|
| 31 |
+
{"step": 160, "max_steps": 763, "train_loss": 6.354224, "learning_rate": 0.0005514597483108611, "grad_norm": 0.6265206933021545, "tokens_seen": 20971520, "tokens_per_second": 266246}
|
| 32 |
+
{"step": 170, "max_steps": 763, "train_loss": 6.335632, "learning_rate": 0.0005447793669086647, "grad_norm": 0.5765638947486877, "tokens_seen": 22282240, "tokens_per_second": 266228}
|
| 33 |
+
{"step": 180, "max_steps": 763, "train_loss": 6.248039, "learning_rate": 0.0005377201721653773, "grad_norm": 0.5617342591285706, "tokens_seen": 23592960, "tokens_per_second": 265753}
|
| 34 |
+
{"step": 190, "max_steps": 763, "train_loss": 6.192135, "learning_rate": 0.0005302946146119068, "grad_norm": 0.6096681952476501, "tokens_seen": 24903680, "tokens_per_second": 266257}
|
| 35 |
+
{"step": 200, "max_steps": 763, "train_loss": 6.179987, "learning_rate": 0.0005225157909451416, "grad_norm": 0.7556291818618774, "tokens_seen": 26214400, "tokens_per_second": 266294}
|
| 36 |
+
{"step": 200, "validation_loss": 6.177583, "validation_perplexity": 481.8257}
|
| 37 |
+
{"step": 210, "max_steps": 763, "train_loss": 6.150098, "learning_rate": 0.0005143974209288811, "grad_norm": 0.8261726498603821, "tokens_seen": 27525120, "tokens_per_second": 230325}
|
| 38 |
+
{"step": 220, "max_steps": 763, "train_loss": 6.067002, "learning_rate": 0.0005059538231958432, "grad_norm": 0.4761219918727875, "tokens_seen": 28835840, "tokens_per_second": 266248}
|
| 39 |
+
{"step": 230, "max_steps": 763, "train_loss": 6.02484, "learning_rate": 0.0004971998899934259, "grad_norm": 0.5485476851463318, "tokens_seen": 30146560, "tokens_per_second": 266312}
|
| 40 |
+
{"step": 240, "max_steps": 763, "train_loss": 5.959249, "learning_rate": 0.00048815106091776853, "grad_norm": 0.4694155156612396, "tokens_seen": 31457280, "tokens_per_second": 266253}
|
| 41 |
+
{"step": 250, "max_steps": 763, "train_loss": 5.930261, "learning_rate": 0.0004788232956824341, "grad_norm": 0.4947860836982727, "tokens_seen": 32768000, "tokens_per_second": 266309}
|
| 42 |
+
Saved checkpoint: runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000250
|
| 43 |
+
{"step": 260, "max_steps": 763, "train_loss": 5.946968, "learning_rate": 0.00046923304596974483, "grad_norm": 0.7666828036308289, "tokens_seen": 34078720, "tokens_per_second": 232366}
|
| 44 |
+
{"step": 270, "max_steps": 763, "train_loss": 5.909852, "learning_rate": 0.0004593972264144153, "grad_norm": 0.7981503009796143, "tokens_seen": 35389440, "tokens_per_second": 266313}
|
| 45 |
+
{"step": 280, "max_steps": 763, "train_loss": 5.822793, "learning_rate": 0.00044933318477066334, "grad_norm": 0.41134122014045715, "tokens_seen": 36700160, "tokens_per_second": 266308}
|
| 46 |
+
{"step": 290, "max_steps": 763, "train_loss": 5.782565, "learning_rate": 0.00043905867131541155, "grad_norm": 0.6911254525184631, "tokens_seen": 38010880, "tokens_per_second": 266304}
|
| 47 |
+
{"step": 300, "max_steps": 763, "train_loss": 5.804326, "learning_rate": 0.0004285918075415489, "grad_norm": 0.6334877014160156, "tokens_seen": 39321600, "tokens_per_second": 266317}
|
| 48 |
+
{"step": 300, "validation_loss": 5.809348, "validation_perplexity": 333.4018}
|
| 49 |
+
{"step": 310, "max_steps": 763, "train_loss": 5.771087, "learning_rate": 0.00041795105419646444, "grad_norm": 0.730846107006073, "tokens_seen": 40632320, "tokens_per_second": 230434}
|
| 50 |
+
{"step": 320, "max_steps": 763, "train_loss": 5.740399, "learning_rate": 0.00040715517872222867, "grad_norm": 0.6593385338783264, "tokens_seen": 41943040, "tokens_per_second": 266365}
|
| 51 |
+
{"step": 330, "max_steps": 763, "train_loss": 5.700934, "learning_rate": 0.0003962232221548463, "grad_norm": 0.6619644165039062, "tokens_seen": 43253760, "tokens_per_second": 266342}
|
| 52 |
+
{"step": 340, "max_steps": 763, "train_loss": 5.700379, "learning_rate": 0.00038517446554096337, "grad_norm": 0.5015922784805298, "tokens_seen": 44564480, "tokens_per_second": 266328}
|
| 53 |
+
{"step": 350, "max_steps": 763, "train_loss": 5.660067, "learning_rate": 0.0003740283959312592, "grad_norm": 0.66937255859375, "tokens_seen": 45875200, "tokens_per_second": 266356}
|
| 54 |
+
{"step": 360, "max_steps": 763, "train_loss": 5.717638, "learning_rate": 0.0003628046720105023, "grad_norm": 0.6924976706504822, "tokens_seen": 47185920, "tokens_per_second": 266329}
|
| 55 |
+
{"step": 370, "max_steps": 763, "train_loss": 5.649611, "learning_rate": 0.00035152308942489074, "grad_norm": 0.6389058828353882, "tokens_seen": 48496640, "tokens_per_second": 266378}
|
| 56 |
+
{"step": 380, "max_steps": 763, "train_loss": 5.603519, "learning_rate": 0.0003402035458678273, "grad_norm": 0.5245085954666138, "tokens_seen": 49807360, "tokens_per_second": 266357}
|
| 57 |
+
{"step": 390, "max_steps": 763, "train_loss": 5.570693, "learning_rate": 0.00032886600598571277, "grad_norm": 0.443662166595459, "tokens_seen": 51118080, "tokens_per_second": 266339}
|
| 58 |
+
{"step": 400, "max_steps": 763, "train_loss": 5.585567, "learning_rate": 0.0003175304661656503, "grad_norm": 0.6869713068008423, "tokens_seen": 52428800, "tokens_per_second": 266338}
|
| 59 |
+
{"step": 400, "validation_loss": 5.583481, "validation_perplexity": 265.996}
|
| 60 |
+
{"step": 410, "max_steps": 763, "train_loss": 5.563179, "learning_rate": 0.00030621691926716884, "grad_norm": 0.6459590196609497, "tokens_seen": 53739520, "tokens_per_second": 230345}
|
| 61 |
+
{"step": 420, "max_steps": 763, "train_loss": 5.525482, "learning_rate": 0.0002949453193601676, "grad_norm": 0.5371569395065308, "tokens_seen": 55050240, "tokens_per_second": 266334}
|
| 62 |
+
{"step": 430, "max_steps": 763, "train_loss": 5.5109, "learning_rate": 0.0002837355465312754, "grad_norm": 0.6863892674446106, "tokens_seen": 56360960, "tokens_per_second": 266321}
|
| 63 |
+
{"step": 440, "max_steps": 763, "train_loss": 5.523325, "learning_rate": 0.00027260737182069626, "grad_norm": 0.6527363657951355, "tokens_seen": 57671680, "tokens_per_second": 266328}
|
| 64 |
+
{"step": 450, "max_steps": 763, "train_loss": 5.498847, "learning_rate": 0.00026158042235138474, "grad_norm": 0.5187376737594604, "tokens_seen": 58982400, "tokens_per_second": 266377}
|
| 65 |
+
{"step": 460, "max_steps": 763, "train_loss": 5.415719, "learning_rate": 0.00025067414671205103, "grad_norm": 0.5715372562408447, "tokens_seen": 60293120, "tokens_per_second": 266317}
|
| 66 |
+
{"step": 470, "max_steps": 763, "train_loss": 5.42196, "learning_rate": 0.00023990778065505504, "grad_norm": 0.6693496108055115, "tokens_seen": 61603840, "tokens_per_second": 266344}
|
| 67 |
+
{"step": 480, "max_steps": 763, "train_loss": 5.417011, "learning_rate": 0.00022930031316968416, "grad_norm": 0.653715193271637, "tokens_seen": 62914560, "tokens_per_second": 266362}
|
| 68 |
+
{"step": 490, "max_steps": 763, "train_loss": 5.422983, "learning_rate": 0.00021887045299065797, "grad_norm": 0.8522746562957764, "tokens_seen": 64225280, "tokens_per_second": 266396}
|
| 69 |
+
{"step": 500, "max_steps": 763, "train_loss": 5.384743, "learning_rate": 0.00020863659560092397, "grad_norm": 0.6453570127487183, "tokens_seen": 65536000, "tokens_per_second": 266397}
|
| 70 |
+
{"step": 500, "validation_loss": 5.424639, "validation_perplexity": 226.9294}
|
| 71 |
+
Saved checkpoint: runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000500
|
| 72 |
+
{"step": 510, "max_steps": 763, "train_loss": 5.420039, "learning_rate": 0.00019861679078694882, "grad_norm": 0.6342107057571411, "tokens_seen": 66846720, "tokens_per_second": 207241}
|
| 73 |
+
{"step": 520, "max_steps": 763, "train_loss": 5.376793, "learning_rate": 0.0001888287108037225, "grad_norm": 0.5910501480102539, "tokens_seen": 68157440, "tokens_per_second": 266397}
|
| 74 |
+
{"step": 530, "max_steps": 763, "train_loss": 5.394478, "learning_rate": 0.00017928961920562968, "grad_norm": 0.5266616344451904, "tokens_seen": 69468160, "tokens_per_second": 266388}
|
| 75 |
+
{"step": 540, "max_steps": 763, "train_loss": 5.376837, "learning_rate": 0.000170016340398158, "grad_norm": 0.5880401134490967, "tokens_seen": 70778880, "tokens_per_second": 266413}
|
| 76 |
+
{"step": 550, "max_steps": 763, "train_loss": 5.352697, "learning_rate": 0.00016102522996414806, "grad_norm": 0.676957905292511, "tokens_seen": 72089600, "tokens_per_second": 266408}
|
| 77 |
+
{"step": 560, "max_steps": 763, "train_loss": 5.337461, "learning_rate": 0.00015233214581692112, "grad_norm": 0.520514965057373, "tokens_seen": 73400320, "tokens_per_second": 266409}
|
| 78 |
+
{"step": 570, "max_steps": 763, "train_loss": 5.333974, "learning_rate": 0.0001439524202311625, "grad_norm": 0.5339544415473938, "tokens_seen": 74711040, "tokens_per_second": 266407}
|
| 79 |
+
{"step": 580, "max_steps": 763, "train_loss": 5.335162, "learning_rate": 0.0001359008328008913, "grad_norm": 0.5292180180549622, "tokens_seen": 76021760, "tokens_per_second": 266405}
|
| 80 |
+
{"step": 590, "max_steps": 763, "train_loss": 5.345912, "learning_rate": 0.0001281915843722103, "grad_norm": 0.4299924373626709, "tokens_seen": 77332480, "tokens_per_second": 266347}
|
| 81 |
+
{"step": 600, "max_steps": 763, "train_loss": 5.335148, "learning_rate": 0.00012083827199681319, "grad_norm": 0.5129626393318176, "tokens_seen": 78643200, "tokens_per_second": 266393}
|
| 82 |
+
{"step": 600, "validation_loss": 5.320427, "validation_perplexity": 204.4711}
|
| 83 |
+
{"step": 610, "max_steps": 763, "train_loss": 5.290448, "learning_rate": 0.00011385386495042271, "grad_norm": 0.6020255088806152, "tokens_seen": 79953920, "tokens_per_second": 230475}
|
| 84 |
+
{"step": 620, "max_steps": 763, "train_loss": 5.266137, "learning_rate": 0.0001072506818584584, "grad_norm": 0.5276666283607483, "tokens_seen": 81264640, "tokens_per_second": 266399}
|
| 85 |
+
{"step": 630, "max_steps": 763, "train_loss": 5.266309, "learning_rate": 0.00010104036896927688, "grad_norm": 0.6189335584640503, "tokens_seen": 82575360, "tokens_per_second": 266374}
|
| 86 |
+
{"step": 640, "max_steps": 763, "train_loss": 5.281845, "learning_rate": 9.523387961330576e-05, "grad_norm": 0.5236685872077942, "tokens_seen": 83886080, "tokens_per_second": 266381}
|
| 87 |
+
{"step": 650, "max_steps": 763, "train_loss": 5.260385, "learning_rate": 8.984145488429902e-05, "grad_norm": 0.4144093990325928, "tokens_seen": 85196800, "tokens_per_second": 266397}
|
| 88 |
+
{"step": 660, "max_steps": 763, "train_loss": 5.27364, "learning_rate": 8.487260557678771e-05, "grad_norm": 0.5166460275650024, "tokens_seen": 86507520, "tokens_per_second": 266401}
|
| 89 |
+
{"step": 670, "max_steps": 763, "train_loss": 5.249187, "learning_rate": 8.033609541158293e-05, "grad_norm": 0.5099441409111023, "tokens_seen": 87818240, "tokens_per_second": 266340}
|
| 90 |
+
{"step": 680, "max_steps": 763, "train_loss": 5.229744, "learning_rate": 7.623992557891675e-05, "grad_norm": 0.49411189556121826, "tokens_seen": 89128960, "tokens_per_second": 266365}
|
| 91 |
+
{"step": 690, "max_steps": 763, "train_loss": 5.194886, "learning_rate": 7.25913206264835e-05, "grad_norm": 0.45130783319473267, "tokens_seen": 90439680, "tokens_per_second": 266368}
|
| 92 |
+
{"step": 700, "max_steps": 763, "train_loss": 5.207998, "learning_rate": 6.939671571727054e-05, "grad_norm": 0.5291787385940552, "tokens_seen": 91750400, "tokens_per_second": 266384}
|
| 93 |
+
{"step": 700, "validation_loss": 5.252493, "validation_perplexity": 191.0419}
|
| 94 |
+
{"step": 710, "max_steps": 763, "train_loss": 5.198223, "learning_rate": 6.6661745279652e-05, "grad_norm": 0.4649067223072052, "tokens_seen": 93061120, "tokens_per_second": 230502}
|
| 95 |
+
{"step": 720, "max_steps": 763, "train_loss": 5.193981, "learning_rate": 6.439123306976511e-05, "grad_norm": 0.5638070702552795, "tokens_seen": 94371840, "tokens_per_second": 266406}
|
| 96 |
+
{"step": 730, "max_steps": 763, "train_loss": 5.215189, "learning_rate": 6.258918366369462e-05, "grad_norm": 0.5251625776290894, "tokens_seen": 95682560, "tokens_per_second": 266414}
|
| 97 |
+
{"step": 740, "max_steps": 763, "train_loss": 5.223023, "learning_rate": 6.125877539447194e-05, "grad_norm": 0.4707590937614441, "tokens_seen": 96993280, "tokens_per_second": 266411}
|
| 98 |
+
{"step": 750, "max_steps": 763, "train_loss": 5.205901, "learning_rate": 6.0402354746346246e-05, "grad_norm": 0.48817580938339233, "tokens_seen": 98304000, "tokens_per_second": 266402}
|
| 99 |
+
Saved checkpoint: runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000750
|
| 100 |
+
{"step": 760, "max_steps": 763, "train_loss": 5.174543, "learning_rate": 6.002143221621371e-05, "grad_norm": 0.4425918161869049, "tokens_seen": 99614720, "tokens_per_second": 235110}
|
| 101 |
+
Final checkpoint: runs/kawk50m-fixed-scratch-100m-5090/checkpoint-00000763
|
reports/fixed-recovery-1b/byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 418575,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 2.969563405239076,
|
| 13 |
+
"token_perplexity": 19.483411383510795,
|
| 14 |
+
"nats_per_character": 1.5684115853740337,
|
| 15 |
+
"bits_per_utf8_byte": 0.9495396184823441,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/fixed-recovery-1b/evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630/tokenizer",
|
| 4 |
+
"validation_loss": 2.974478209018707,
|
| 5 |
+
"validation_perplexity": 19.579404227186153,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "그 어느 곳보다도 넓은데, 이를 이용하여 도시를 건설하거나 다른 도시를 개발하는 데에 있어서는 고성군이 그 어느 도시보다도 월등히 뛰어나다.\n그리고 도시의 지형적 특성과 함께 도시의 형성과정을 보면, 고성군은 수천 년 동안 고성을 형성한 최초의 지방 도시로서, 인구 50만 명 이상인 작은 도시로, 이 지역의 풍부한 자연 자원을 바탕으로 여러 가지 매력적인 도시 계획을 가지고 있다. 고성군은 또한 인구 142만명의 인구가 있는 작은 도시로, 전형적인 고성군의 중심 도시이다. 그러나 인구 137만 명의 인구로, 고성군은",
|
| 11 |
+
"full_text": "대한민국의 수도는그 어느 곳보다도 넓은데, 이를 이용하여 도시를 건설하거나 다른 도시를 개발하는 데에 있어서는 고성군이 그 어느 도시보다도 월등히 뛰어나다.\n그리고 도시의 지형적 특성과 함께 도시의 형성과정을 보면, 고성군은 수천 년 동안 고성을 형성한 최초의 지방 도시로서, 인구 50만 명 이상인 작은 도시로, 이 지역의 풍부한 자연 자원을 바탕으로 여러 가지 매력적인 도시 계획을 가지고 있다. 고성군은 또한 인구 142만명의 인구가 있는 작은 도시로, 전형적인 고성군의 중심 도시이다. 그러나 인구 137만 명의 인구로, 고성군은"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "퇴근하고 나면 30분 후에 일어나서 1시간 정도면 충분합니다.\n예전에 회사에서 일하는 사람이 퇴근하자마자 도무지 출근을 할 수가 없어서 아침 일찍 일어나서 5시쯤부터 출근하려고 했습니다.\n그렇게 해서 아침에 일어났더니, 아침 일찍 일어나서 사무실에 들어와서 직원에게 카톡으로 보고를 했습니다.\n아무래도 새벽에 일어나기에는 4시 넘어서 출근해야 하고,\n그 다음에 회사에서 퇴근하는 것이 조금 늦어져서 출근하면 20분 정도면 가능할 것 같습니다.\n그래서 회사에서 퇴근해서 20분 정도 지나면 출근을 하고 나면,",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서퇴근하고 나면 30분 후에 일어나서 1시간 정도면 충분합니다.\n예전에 회사에서 일하는 사람이 퇴근하자마자 도무지 출근을 할 수가 없어서 아침 일찍 일어나서 5시쯤부터 출근하려고 했습니다.\n그렇게 해서 아침에 일어났더니, 아침 일찍 일어나서 사무실에 들어와서 직원에게 카톡으로 보고를 했습니다.\n아무래도 새벽에 일어나기에는 4시 넘어서 출근해야 하고,\n그 다음에 회사에서 퇴근하는 것이 조금 늦어져서 출근하면 20분 정도면 가능할 것 같습니다.\n그래서 회사에서 퇴근해서 20분 정도 지나면 출근을 하고 나면,"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "2017년 이후로 가장 많은 딥러닝을 수행했습니다.\n이제는 학습자가 자신의 능력을 향상시키고, 더 나은 의사 결정을 내리고, 더 나은 사회를 건설할 수 있도록 돕는 데 도움을 줄 수 있습니다. 이 기사에서는 인공지능의 잠재력을 극대화하고, 인간과 기계가 공동으로 발전하도록 도와줄 수 있는 몇 가지 기술을 살펴봅니다.\n인공지능의 잠재력\n인공지능은 현재 우리가 알고 있는 것 이상의 발전을 이루었습니다. 컴퓨터 비전 및 네트워크 분석과 같은 첨단 기술의 도입은 이러한 발전의 핵심입니다. 이 기술들은 인간의 지능과 같은 복잡한 작업을 처리하는 데 매우 효과적일 수 있습니다.\n이러한 기술은 인간이 학습하는 방식에 큰",
|
| 21 |
+
"full_text": "인공지능 언어 모델은2017년 이후로 가장 많은 딥러닝을 수행했습니다.\n이제는 학습자가 자신의 능력을 향상시키고, 더 나은 의사 결정을 내리고, 더 나은 사회를 건설할 수 있도록 돕는 데 도움을 줄 수 있습니다. 이 기사에서는 인공지능의 잠재력을 극대화하고, 인간과 기계가 공동으로 발전하도록 도와줄 수 있는 몇 가지 기술을 살펴봅니다.\n인공지능의 잠재력\n인공지능은 현재 우리가 알고 있는 것 이상의 발전을 이루었습니다. 컴퓨터 비전 및 네트워크 분석과 같은 첨단 기술의 도입은 이러한 발전의 핵심입니다. 이 기술들은 인간의 지능과 같은 복잡한 작업을 처리하는 데 매우 효과적일 수 있습니다.\n이러한 기술은 인간이 학습하는 방식에 큰"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "습기가 차서 물이 넘쳐나는 곳이 많다. 특히 눈이 오거나, 갑자기 눈이 부셔진 눈의 경우 물이 빠지는 게 더 흔하다. 눈은 눈꺼풀과 눈 주위 근육을 자극해 눈가의 근육이 약해지고, 눈 주변 피부와 턱까지 염증이 생기기 때문이다. 특히 50대 이후에는 눈 건강에 대한 관심이 더욱 높아진다. 눈 밑에 생기는 ‘눈막변성’ 때문에 눈 수술 후 회복이 늦어지고, 고령화로 인해 눈 질환도 늘어나고 있다.\n눈이 안 좋아지는 원인으로 꼽히는 것은 눈을 꽉 끼우지 못해서다. 눈꺼풀과 코를 가리는 눈동자의 모양이나",
|
| 26 |
+
"full_text": "비가 내리는 날에는습기가 차서 물이 넘쳐나는 곳이 많다. 특히 눈이 오거나, 갑자기 눈이 부셔진 눈의 경우 물이 빠지는 게 더 흔하다. 눈은 눈꺼풀과 눈 주위 근육을 자극해 눈가의 근육이 약해지고, 눈 주변 피부와 턱까지 염증이 생기기 때문이다. 특히 50대 이후에는 눈 건강에 대한 관심이 더욱 높아진다. 눈 밑에 생기는 ‘눈막변성’ 때문에 눈 수술 후 회복이 늦어지고, 고령화로 인해 눈 질환도 늘어나고 있다.\n눈이 안 좋아지는 원인으로 꼽히는 것은 눈을 꽉 끼우지 못해서다. 눈꺼풀과 코를 가리는 눈동자의 모양이나"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/fixed-recovery-1b/final_checkpoint.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630
|
reports/fixed-recovery-1b/gate.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"passed": true,
|
| 3 |
+
"candidate_loss": 2.974478209018707,
|
| 4 |
+
"baseline_loss": 3.166912519931793,
|
| 5 |
+
"loss_improvement": 0.19243431091308594,
|
| 6 |
+
"sample_count": 4
|
| 7 |
+
}
|
reports/fixed-recovery-1b/train.log
ADDED
|
@@ -0,0 +1,186 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"training_sample_offset": 0,
|
| 7 |
+
"tokens_per_optimizer_step": 131072,
|
| 8 |
+
"max_steps": 7630,
|
| 9 |
+
"planned_tokens": 1000079360,
|
| 10 |
+
"start_step": 0,
|
| 11 |
+
"initialized_from": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-100m-5090/checkpoint-00000763",
|
| 12 |
+
"precision": "bf16",
|
| 13 |
+
"compile": true
|
| 14 |
+
}
|
| 15 |
+
{"step": 1, "max_steps": 7630, "train_loss": 3.14917, "learning_rate": 1.96078431372549e-06, "grad_norm": 0.461567759513855, "tokens_seen": 131072, "tokens_per_second": 44668}
|
| 16 |
+
{"step": 50, "max_steps": 7630, "train_loss": 3.055487, "learning_rate": 9.80392156862745e-05, "grad_norm": 0.4343174695968628, "tokens_seen": 6553600, "tokens_per_second": 266679}
|
| 17 |
+
{"step": 100, "max_steps": 7630, "train_loss": 3.11675, "learning_rate": 0.000196078431372549, "grad_norm": 0.47913938760757446, "tokens_seen": 13107200, "tokens_per_second": 266568}
|
| 18 |
+
{"step": 150, "max_steps": 7630, "train_loss": 3.197033, "learning_rate": 0.0002941176470588235, "grad_norm": 0.5844945311546326, "tokens_seen": 19660800, "tokens_per_second": 266554}
|
| 19 |
+
{"step": 200, "max_steps": 7630, "train_loss": 3.145539, "learning_rate": 0.00029997367730455693, "grad_norm": 0.5809003114700317, "tokens_seen": 26214400, "tokens_per_second": 266543}
|
| 20 |
+
{"step": 250, "max_steps": 7630, "train_loss": 3.140995, "learning_rate": 0.00029988789316069285, "grad_norm": 0.5387641787528992, "tokens_seen": 32768000, "tokens_per_second": 266677}
|
| 21 |
+
{"step": 300, "max_steps": 7630, "train_loss": 3.159402, "learning_rate": 0.0002997425781795139, "grad_norm": 0.5061216950416565, "tokens_seen": 39321600, "tokens_per_second": 266595}
|
| 22 |
+
{"step": 350, "max_steps": 7630, "train_loss": 3.180611, "learning_rate": 0.00029953779649370313, "grad_norm": 0.5989519357681274, "tokens_seen": 45875200, "tokens_per_second": 266510}
|
| 23 |
+
{"step": 400, "max_steps": 7630, "train_loss": 3.127745, "learning_rate": 0.00029927363848072095, "grad_norm": 0.5487684011459351, "tokens_seen": 52428800, "tokens_per_second": 266494}
|
| 24 |
+
{"step": 450, "max_steps": 7630, "train_loss": 3.043574, "learning_rate": 0.0002989502207229183, "grad_norm": 0.5644057989120483, "tokens_seen": 58982400, "tokens_per_second": 266506}
|
| 25 |
+
{"step": 500, "max_steps": 7630, "train_loss": 3.087376, "learning_rate": 0.0002985676859560847, "grad_norm": 0.5145071148872375, "tokens_seen": 65536000, "tokens_per_second": 266501}
|
| 26 |
+
{"step": 500, "validation_loss": 3.185626, "validation_perplexity": 24.1824}
|
| 27 |
+
{"step": 550, "max_steps": 7630, "train_loss": 3.116938, "learning_rate": 0.0002981262030064538, "grad_norm": 0.5195316672325134, "tokens_seen": 72089600, "tokens_per_second": 233116}
|
| 28 |
+
{"step": 600, "max_steps": 7630, "train_loss": 3.198499, "learning_rate": 0.0002976259667161944, "grad_norm": 0.5154688358306885, "tokens_seen": 78643200, "tokens_per_second": 266560}
|
| 29 |
+
{"step": 650, "max_steps": 7630, "train_loss": 3.105799, "learning_rate": 0.0002970671978574197, "grad_norm": 0.5275366902351379, "tokens_seen": 85196800, "tokens_per_second": 266554}
|
| 30 |
+
{"step": 700, "max_steps": 7630, "train_loss": 3.087204, "learning_rate": 0.00029645014303475263, "grad_norm": 0.5567337274551392, "tokens_seen": 91750400, "tokens_per_second": 266547}
|
| 31 |
+
{"step": 750, "max_steps": 7630, "train_loss": 3.128867, "learning_rate": 0.0002957750745764904, "grad_norm": 0.5531742572784424, "tokens_seen": 98304000, "tokens_per_second": 266540}
|
| 32 |
+
{"step": 800, "max_steps": 7630, "train_loss": 3.06997, "learning_rate": 0.00029504229041441644, "grad_norm": 0.5321512222290039, "tokens_seen": 104857600, "tokens_per_second": 266547}
|
| 33 |
+
{"step": 850, "max_steps": 7630, "train_loss": 3.093138, "learning_rate": 0.0002942521139523123, "grad_norm": 0.48930463194847107, "tokens_seen": 111411200, "tokens_per_second": 266541}
|
| 34 |
+
{"step": 900, "max_steps": 7630, "train_loss": 3.051664, "learning_rate": 0.0002934048939232282, "grad_norm": 0.5285896062850952, "tokens_seen": 117964800, "tokens_per_second": 266545}
|
| 35 |
+
{"step": 950, "max_steps": 7630, "train_loss": 3.138493, "learning_rate": 0.00029250100423557453, "grad_norm": 0.5378010869026184, "tokens_seen": 124518400, "tokens_per_second": 266438}
|
| 36 |
+
{"step": 1000, "max_steps": 7630, "train_loss": 3.121481, "learning_rate": 0.00029154084380810294, "grad_norm": 0.5151432752609253, "tokens_seen": 131072000, "tokens_per_second": 266551}
|
| 37 |
+
{"step": 1000, "validation_loss": 3.155982, "validation_perplexity": 23.4761}
|
| 38 |
+
{"step": 1050, "max_steps": 7630, "train_loss": 3.079285, "learning_rate": 0.0002905248363938491, "grad_norm": 0.517073929309845, "tokens_seen": 137625600, "tokens_per_second": 258504}
|
| 39 |
+
{"step": 1100, "max_steps": 7630, "train_loss": 3.142563, "learning_rate": 0.00028945343039311547, "grad_norm": 0.5353425741195679, "tokens_seen": 144179200, "tokens_per_second": 266543}
|
| 40 |
+
{"step": 1150, "max_steps": 7630, "train_loss": 3.13499, "learning_rate": 0.00028832709865557623, "grad_norm": 0.5230163335800171, "tokens_seen": 150732800, "tokens_per_second": 266556}
|
| 41 |
+
{"step": 1200, "max_steps": 7630, "train_loss": 3.079083, "learning_rate": 0.0002871463382715917, "grad_norm": 0.5130206346511841, "tokens_seen": 157286400, "tokens_per_second": 266523}
|
| 42 |
+
{"step": 1250, "max_steps": 7630, "train_loss": 3.025394, "learning_rate": 0.00028591167035282503, "grad_norm": 0.5010837912559509, "tokens_seen": 163840000, "tokens_per_second": 266559}
|
| 43 |
+
{"step": 1300, "max_steps": 7630, "train_loss": 3.124963, "learning_rate": 0.00028462363980225665, "grad_norm": 0.5116173028945923, "tokens_seen": 170393600, "tokens_per_second": 266567}
|
| 44 |
+
{"step": 1350, "max_steps": 7630, "train_loss": 3.055799, "learning_rate": 0.0002832828150736993, "grad_norm": 0.5253769159317017, "tokens_seen": 176947200, "tokens_per_second": 266569}
|
| 45 |
+
{"step": 1400, "max_steps": 7630, "train_loss": 3.10259, "learning_rate": 0.00028188978792091906, "grad_norm": 0.5534281730651855, "tokens_seen": 183500800, "tokens_per_second": 266564}
|
| 46 |
+
{"step": 1450, "max_steps": 7630, "train_loss": 3.085559, "learning_rate": 0.0002804451731364734, "grad_norm": 0.5330402851104736, "tokens_seen": 190054400, "tokens_per_second": 266567}
|
| 47 |
+
{"step": 1500, "max_steps": 7630, "train_loss": 3.065464, "learning_rate": 0.0002789496082803814, "grad_norm": 0.5335582494735718, "tokens_seen": 196608000, "tokens_per_second": 266547}
|
| 48 |
+
{"step": 1500, "validation_loss": 3.134659, "validation_perplexity": 22.9808}
|
| 49 |
+
{"step": 1550, "max_steps": 7630, "train_loss": 3.007484, "learning_rate": 0.00027740375339874526, "grad_norm": 0.5046440362930298, "tokens_seen": 203161600, "tokens_per_second": 258492}
|
| 50 |
+
{"step": 1600, "max_steps": 7630, "train_loss": 3.156556, "learning_rate": 0.00027580829073244913, "grad_norm": 0.5801460146903992, "tokens_seen": 209715200, "tokens_per_second": 266566}
|
| 51 |
+
{"step": 1650, "max_steps": 7630, "train_loss": 3.111725, "learning_rate": 0.0002741639244160613, "grad_norm": 0.4891602694988251, "tokens_seen": 216268800, "tokens_per_second": 266550}
|
| 52 |
+
{"step": 1700, "max_steps": 7630, "train_loss": 3.016304, "learning_rate": 0.0002724713801670748, "grad_norm": 0.5425730347633362, "tokens_seen": 222822400, "tokens_per_second": 266570}
|
| 53 |
+
{"step": 1750, "max_steps": 7630, "train_loss": 3.095851, "learning_rate": 0.00027073140496562236, "grad_norm": 0.5016234517097473, "tokens_seen": 229376000, "tokens_per_second": 266568}
|
| 54 |
+
{"step": 1800, "max_steps": 7630, "train_loss": 3.031469, "learning_rate": 0.0002689447667248074, "grad_norm": 0.5111928582191467, "tokens_seen": 235929600, "tokens_per_second": 266564}
|
| 55 |
+
{"step": 1850, "max_steps": 7630, "train_loss": 3.025842, "learning_rate": 0.0002671122539517967, "grad_norm": 0.522367000579834, "tokens_seen": 242483200, "tokens_per_second": 266552}
|
| 56 |
+
{"step": 1900, "max_steps": 7630, "train_loss": 3.080013, "learning_rate": 0.00026523467539982373, "grad_norm": 0.5070951581001282, "tokens_seen": 249036800, "tokens_per_second": 266550}
|
| 57 |
+
{"step": 1950, "max_steps": 7630, "train_loss": 3.073859, "learning_rate": 0.0002633128597112574, "grad_norm": 0.5060163736343384, "tokens_seen": 255590400, "tokens_per_second": 266578}
|
| 58 |
+
{"step": 2000, "max_steps": 7630, "train_loss": 3.087708, "learning_rate": 0.0002613476550518925, "grad_norm": 0.482485294342041, "tokens_seen": 262144000, "tokens_per_second": 266563}
|
| 59 |
+
{"step": 2000, "validation_loss": 3.115587, "validation_perplexity": 22.5467}
|
| 60 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00002000
|
| 61 |
+
{"step": 2050, "max_steps": 7630, "train_loss": 3.080326, "learning_rate": 0.0002593399287366238, "grad_norm": 0.5073894262313843, "tokens_seen": 268697600, "tokens_per_second": 254797}
|
| 62 |
+
{"step": 2100, "max_steps": 7630, "train_loss": 3.058296, "learning_rate": 0.0002572905668466692, "grad_norm": 0.5159104466438293, "tokens_seen": 275251200, "tokens_per_second": 266551}
|
| 63 |
+
{"step": 2150, "max_steps": 7630, "train_loss": 3.037933, "learning_rate": 0.00025520047383851034, "grad_norm": 0.4780135154724121, "tokens_seen": 281804800, "tokens_per_second": 266446}
|
| 64 |
+
{"step": 2200, "max_steps": 7630, "train_loss": 3.073052, "learning_rate": 0.00025307057214472384, "grad_norm": 0.5303418636322021, "tokens_seen": 288358400, "tokens_per_second": 266561}
|
| 65 |
+
{"step": 2250, "max_steps": 7630, "train_loss": 2.938057, "learning_rate": 0.0002509018017668789, "grad_norm": 0.5631779432296753, "tokens_seen": 294912000, "tokens_per_second": 266551}
|
| 66 |
+
{"step": 2300, "max_steps": 7630, "train_loss": 3.096883, "learning_rate": 0.0002486951198606809, "grad_norm": 0.5420588254928589, "tokens_seen": 301465600, "tokens_per_second": 266549}
|
| 67 |
+
{"step": 2350, "max_steps": 7630, "train_loss": 3.018632, "learning_rate": 0.00024645150031354473, "grad_norm": 0.5302236080169678, "tokens_seen": 308019200, "tokens_per_second": 266568}
|
| 68 |
+
{"step": 2400, "max_steps": 7630, "train_loss": 3.013168, "learning_rate": 0.00024417193331478337, "grad_norm": 0.5048549175262451, "tokens_seen": 314572800, "tokens_per_second": 266556}
|
| 69 |
+
{"step": 2450, "max_steps": 7630, "train_loss": 2.971954, "learning_rate": 0.00024185742491860207, "grad_norm": 0.530778706073761, "tokens_seen": 321126400, "tokens_per_second": 266560}
|
| 70 |
+
{"step": 2500, "max_steps": 7630, "train_loss": 3.065892, "learning_rate": 0.00023950899660009074, "grad_norm": 0.4880107343196869, "tokens_seen": 327680000, "tokens_per_second": 266538}
|
| 71 |
+
{"step": 2500, "validation_loss": 3.097326, "validation_perplexity": 22.1387}
|
| 72 |
+
{"step": 2550, "max_steps": 7630, "train_loss": 3.041774, "learning_rate": 0.00023712768480441067, "grad_norm": 0.4878818690776825, "tokens_seen": 334233600, "tokens_per_second": 258547}
|
| 73 |
+
{"step": 2600, "max_steps": 7630, "train_loss": 3.038461, "learning_rate": 0.00023471454048937406, "grad_norm": 0.46887022256851196, "tokens_seen": 340787200, "tokens_per_second": 266710}
|
| 74 |
+
{"step": 2650, "max_steps": 7630, "train_loss": 3.022111, "learning_rate": 0.00023227062866161923, "grad_norm": 0.5214937925338745, "tokens_seen": 347340800, "tokens_per_second": 266647}
|
| 75 |
+
{"step": 2700, "max_steps": 7630, "train_loss": 2.995276, "learning_rate": 0.00022979702790658504, "grad_norm": 0.5535130500793457, "tokens_seen": 353894400, "tokens_per_second": 266711}
|
| 76 |
+
{"step": 2750, "max_steps": 7630, "train_loss": 3.069385, "learning_rate": 0.0002272948299124924, "grad_norm": 0.499153196811676, "tokens_seen": 360448000, "tokens_per_second": 266552}
|
| 77 |
+
{"step": 2800, "max_steps": 7630, "train_loss": 2.996649, "learning_rate": 0.00022476513898854385, "grad_norm": 0.511376678943634, "tokens_seen": 367001600, "tokens_per_second": 266566}
|
| 78 |
+
{"step": 2850, "max_steps": 7630, "train_loss": 3.00805, "learning_rate": 0.0002222090715775522, "grad_norm": 0.48047617077827454, "tokens_seen": 373555200, "tokens_per_second": 266545}
|
| 79 |
+
{"step": 2900, "max_steps": 7630, "train_loss": 3.010899, "learning_rate": 0.00021962775576321445, "grad_norm": 0.4780416786670685, "tokens_seen": 380108800, "tokens_per_second": 266543}
|
| 80 |
+
{"step": 2950, "max_steps": 7630, "train_loss": 3.018823, "learning_rate": 0.00021702233077224858, "grad_norm": 0.4926721155643463, "tokens_seen": 386662400, "tokens_per_second": 266561}
|
| 81 |
+
{"step": 3000, "max_steps": 7630, "train_loss": 2.968974, "learning_rate": 0.00021439394647161168, "grad_norm": 0.4691397547721863, "tokens_seen": 393216000, "tokens_per_second": 266549}
|
| 82 |
+
{"step": 3000, "validation_loss": 3.079953, "validation_perplexity": 21.7574}
|
| 83 |
+
{"step": 3050, "max_steps": 7630, "train_loss": 3.050222, "learning_rate": 0.00021174376286102307, "grad_norm": 0.47775787115097046, "tokens_seen": 399769600, "tokens_per_second": 258493}
|
| 84 |
+
{"step": 3100, "max_steps": 7630, "train_loss": 3.045075, "learning_rate": 0.00020907294956101487, "grad_norm": 0.5366857051849365, "tokens_seen": 406323200, "tokens_per_second": 266565}
|
| 85 |
+
{"step": 3150, "max_steps": 7630, "train_loss": 3.077274, "learning_rate": 0.00020638268529673683, "grad_norm": 0.49010124802589417, "tokens_seen": 412876800, "tokens_per_second": 266552}
|
| 86 |
+
{"step": 3200, "max_steps": 7630, "train_loss": 3.026062, "learning_rate": 0.00020367415737774282, "grad_norm": 0.5020202994346619, "tokens_seen": 419430400, "tokens_per_second": 266568}
|
| 87 |
+
{"step": 3250, "max_steps": 7630, "train_loss": 2.977737, "learning_rate": 0.0002009485611739888, "grad_norm": 0.4847213923931122, "tokens_seen": 425984000, "tokens_per_second": 266568}
|
| 88 |
+
{"step": 3300, "max_steps": 7630, "train_loss": 2.951593, "learning_rate": 0.00019820709958827344, "grad_norm": 0.4833868741989136, "tokens_seen": 432537600, "tokens_per_second": 266569}
|
| 89 |
+
{"step": 3350, "max_steps": 7630, "train_loss": 3.087995, "learning_rate": 0.00019545098252535424, "grad_norm": 0.4668392241001129, "tokens_seen": 439091200, "tokens_per_second": 266441}
|
| 90 |
+
{"step": 3400, "max_steps": 7630, "train_loss": 2.911588, "learning_rate": 0.00019268142635797344, "grad_norm": 0.5249447226524353, "tokens_seen": 445644800, "tokens_per_second": 266569}
|
| 91 |
+
{"step": 3450, "max_steps": 7630, "train_loss": 2.977207, "learning_rate": 0.00018989965339002922, "grad_norm": 0.4620632529258728, "tokens_seen": 452198400, "tokens_per_second": 266548}
|
| 92 |
+
{"step": 3500, "max_steps": 7630, "train_loss": 3.014922, "learning_rate": 0.00018710689131712957, "grad_norm": 0.45714008808135986, "tokens_seen": 458752000, "tokens_per_second": 266550}
|
| 93 |
+
{"step": 3500, "validation_loss": 3.063313, "validation_perplexity": 21.3983}
|
| 94 |
+
{"step": 3550, "max_steps": 7630, "train_loss": 2.969227, "learning_rate": 0.00018430437268476642, "grad_norm": 0.5447001457214355, "tokens_seen": 465305600, "tokens_per_second": 258509}
|
| 95 |
+
{"step": 3600, "max_steps": 7630, "train_loss": 2.936905, "learning_rate": 0.00018149333434434907, "grad_norm": 0.4822026491165161, "tokens_seen": 471859200, "tokens_per_second": 266545}
|
| 96 |
+
{"step": 3650, "max_steps": 7630, "train_loss": 2.966985, "learning_rate": 0.0001786750169073381, "grad_norm": 0.503084659576416, "tokens_seen": 478412800, "tokens_per_second": 266557}
|
| 97 |
+
{"step": 3700, "max_steps": 7630, "train_loss": 2.943588, "learning_rate": 0.00017585066419771876, "grad_norm": 0.46425068378448486, "tokens_seen": 484966400, "tokens_per_second": 266564}
|
| 98 |
+
{"step": 3750, "max_steps": 7630, "train_loss": 3.011336, "learning_rate": 0.0001730215227030575, "grad_norm": 0.5205370187759399, "tokens_seen": 491520000, "tokens_per_second": 266554}
|
| 99 |
+
{"step": 3800, "max_steps": 7630, "train_loss": 3.029781, "learning_rate": 0.00017018884102438213, "grad_norm": 0.4772203862667084, "tokens_seen": 498073600, "tokens_per_second": 266567}
|
| 100 |
+
{"step": 3850, "max_steps": 7630, "train_loss": 2.863265, "learning_rate": 0.0001673538693251302, "grad_norm": 0.4984719455242157, "tokens_seen": 504627200, "tokens_per_second": 266549}
|
| 101 |
+
{"step": 3900, "max_steps": 7630, "train_loss": 2.996506, "learning_rate": 0.00016451785877940673, "grad_norm": 0.48845186829566956, "tokens_seen": 511180800, "tokens_per_second": 266553}
|
| 102 |
+
{"step": 3950, "max_steps": 7630, "train_loss": 2.964122, "learning_rate": 0.00016168206101979695, "grad_norm": 0.49500709772109985, "tokens_seen": 517734400, "tokens_per_second": 266706}
|
| 103 |
+
{"step": 4000, "max_steps": 7630, "train_loss": 2.977564, "learning_rate": 0.0001588477275849759, "grad_norm": 0.5191928744316101, "tokens_seen": 524288000, "tokens_per_second": 266684}
|
| 104 |
+
{"step": 4000, "validation_loss": 3.047309, "validation_perplexity": 21.0586}
|
| 105 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00004000
|
| 106 |
+
{"step": 4050, "max_steps": 7630, "train_loss": 3.010327, "learning_rate": 0.00015601610936735982, "grad_norm": 0.5142688155174255, "tokens_seen": 530841600, "tokens_per_second": 251899}
|
| 107 |
+
{"step": 4100, "max_steps": 7630, "train_loss": 2.986305, "learning_rate": 0.00015318845606104286, "grad_norm": 0.48496124148368835, "tokens_seen": 537395200, "tokens_per_second": 266663}
|
| 108 |
+
{"step": 4150, "max_steps": 7630, "train_loss": 2.981713, "learning_rate": 0.00015036601561026238, "grad_norm": 0.4842650890350342, "tokens_seen": 543948800, "tokens_per_second": 266720}
|
| 109 |
+
{"step": 4200, "max_steps": 7630, "train_loss": 2.918912, "learning_rate": 0.00014755003365863676, "grad_norm": 0.47794193029403687, "tokens_seen": 550502400, "tokens_per_second": 266729}
|
| 110 |
+
{"step": 4250, "max_steps": 7630, "train_loss": 3.005854, "learning_rate": 0.00014474175299941826, "grad_norm": 0.4521265923976898, "tokens_seen": 557056000, "tokens_per_second": 266732}
|
| 111 |
+
{"step": 4300, "max_steps": 7630, "train_loss": 2.979373, "learning_rate": 0.00014194241302700433, "grad_norm": 0.46327707171440125, "tokens_seen": 563609600, "tokens_per_second": 266760}
|
| 112 |
+
{"step": 4350, "max_steps": 7630, "train_loss": 2.894599, "learning_rate": 0.00013915324918994848, "grad_norm": 0.47927170991897583, "tokens_seen": 570163200, "tokens_per_second": 266726}
|
| 113 |
+
{"step": 4400, "max_steps": 7630, "train_loss": 2.953266, "learning_rate": 0.0001363754924457127, "grad_norm": 0.4778335988521576, "tokens_seen": 576716800, "tokens_per_second": 266562}
|
| 114 |
+
{"step": 4450, "max_steps": 7630, "train_loss": 2.955356, "learning_rate": 0.0001336103687174022, "grad_norm": 0.49364814162254333, "tokens_seen": 583270400, "tokens_per_second": 266550}
|
| 115 |
+
{"step": 4500, "max_steps": 7630, "train_loss": 2.938446, "learning_rate": 0.00013085909835272192, "grad_norm": 0.4981161057949066, "tokens_seen": 589824000, "tokens_per_second": 266550}
|
| 116 |
+
{"step": 4500, "validation_loss": 3.031564, "validation_perplexity": 20.7296}
|
| 117 |
+
{"step": 4550, "max_steps": 7630, "train_loss": 3.074311, "learning_rate": 0.0001281228955853932, "grad_norm": 0.46131420135498047, "tokens_seen": 596377600, "tokens_per_second": 258432}
|
| 118 |
+
{"step": 4600, "max_steps": 7630, "train_loss": 2.960127, "learning_rate": 0.0001254029679992694, "grad_norm": 0.4794043004512787, "tokens_seen": 602931200, "tokens_per_second": 266556}
|
| 119 |
+
{"step": 4650, "max_steps": 7630, "train_loss": 2.976727, "learning_rate": 0.0001227005159953863, "grad_norm": 0.48854124546051025, "tokens_seen": 609484800, "tokens_per_second": 266548}
|
| 120 |
+
{"step": 4700, "max_steps": 7630, "train_loss": 2.906539, "learning_rate": 0.00012001673226218169, "grad_norm": 0.4409690201282501, "tokens_seen": 616038400, "tokens_per_second": 266559}
|
| 121 |
+
{"step": 4750, "max_steps": 7630, "train_loss": 2.932089, "learning_rate": 0.00011735280124912001, "grad_norm": 0.47708672285079956, "tokens_seen": 622592000, "tokens_per_second": 266636}
|
| 122 |
+
{"step": 4800, "max_steps": 7630, "train_loss": 2.945892, "learning_rate": 0.00011470989864395239, "grad_norm": 0.46497851610183716, "tokens_seen": 629145600, "tokens_per_second": 266563}
|
| 123 |
+
{"step": 4850, "max_steps": 7630, "train_loss": 2.911536, "learning_rate": 0.00011208919085384356, "grad_norm": 0.4604366719722748, "tokens_seen": 635699200, "tokens_per_second": 266588}
|
| 124 |
+
{"step": 4900, "max_steps": 7630, "train_loss": 2.925509, "learning_rate": 0.00010949183449059506, "grad_norm": 0.45988044142723083, "tokens_seen": 642252800, "tokens_per_second": 266569}
|
| 125 |
+
{"step": 4950, "max_steps": 7630, "train_loss": 2.977669, "learning_rate": 0.00010691897586019088, "grad_norm": 0.4573512673377991, "tokens_seen": 648806400, "tokens_per_second": 266549}
|
| 126 |
+
{"step": 5000, "max_steps": 7630, "train_loss": 3.016095, "learning_rate": 0.00010437175045689202, "grad_norm": 0.47729921340942383, "tokens_seen": 655360000, "tokens_per_second": 266551}
|
| 127 |
+
{"step": 5000, "validation_loss": 3.016724, "validation_perplexity": 20.4243}
|
| 128 |
+
{"step": 5050, "max_steps": 7630, "train_loss": 2.883886, "learning_rate": 0.00010185128246210238, "grad_norm": 0.4642042815685272, "tokens_seen": 661913600, "tokens_per_second": 258506}
|
| 129 |
+
{"step": 5100, "max_steps": 7630, "train_loss": 2.956777, "learning_rate": 9.935868424822705e-05, "grad_norm": 0.49306541681289673, "tokens_seen": 668467200, "tokens_per_second": 266553}
|
| 130 |
+
{"step": 5150, "max_steps": 7630, "train_loss": 2.921578, "learning_rate": 9.689505588774327e-05, "grad_norm": 0.4548170268535614, "tokens_seen": 675020800, "tokens_per_second": 266569}
|
| 131 |
+
{"step": 5200, "max_steps": 7630, "train_loss": 2.925271, "learning_rate": 9.446148466769876e-05, "grad_norm": 0.4913490116596222, "tokens_seen": 681574400, "tokens_per_second": 266563}
|
| 132 |
+
{"step": 5250, "max_steps": 7630, "train_loss": 2.89866, "learning_rate": 9.205904460985328e-05, "grad_norm": 0.47305092215538025, "tokens_seen": 688128000, "tokens_per_second": 266567}
|
| 133 |
+
{"step": 5300, "max_steps": 7630, "train_loss": 2.895413, "learning_rate": 8.968879599667484e-05, "grad_norm": 0.44757556915283203, "tokens_seen": 694681600, "tokens_per_second": 266550}
|
| 134 |
+
{"step": 5350, "max_steps": 7630, "train_loss": 2.923407, "learning_rate": 8.735178490339868e-05, "grad_norm": 0.4694661498069763, "tokens_seen": 701235200, "tokens_per_second": 266551}
|
| 135 |
+
{"step": 5400, "max_steps": 7630, "train_loss": 2.891526, "learning_rate": 8.50490427363574e-05, "grad_norm": 0.4842621982097626, "tokens_seen": 707788800, "tokens_per_second": 266549}
|
| 136 |
+
{"step": 5450, "max_steps": 7630, "train_loss": 2.923312, "learning_rate": 8.278158577778452e-05, "grad_norm": 0.4796649217605591, "tokens_seen": 714342400, "tokens_per_second": 266551}
|
| 137 |
+
{"step": 5500, "max_steps": 7630, "train_loss": 2.939538, "learning_rate": 8.055041473729265e-05, "grad_norm": 0.46087419986724854, "tokens_seen": 720896000, "tokens_per_second": 266553}
|
| 138 |
+
{"step": 5500, "validation_loss": 3.003748, "validation_perplexity": 20.161}
|
| 139 |
+
{"step": 5550, "max_steps": 7630, "train_loss": 2.933559, "learning_rate": 7.835651431022487e-05, "grad_norm": 0.4657554626464844, "tokens_seen": 727449600, "tokens_per_second": 258504}
|
| 140 |
+
{"step": 5600, "max_steps": 7630, "train_loss": 2.909696, "learning_rate": 7.620085274307355e-05, "grad_norm": 0.5091168880462646, "tokens_seen": 734003200, "tokens_per_second": 266552}
|
| 141 |
+
{"step": 5650, "max_steps": 7630, "train_loss": 2.966356, "learning_rate": 7.40843814061585e-05, "grad_norm": 0.4477066099643707, "tokens_seen": 740556800, "tokens_per_second": 266571}
|
| 142 |
+
{"step": 5700, "max_steps": 7630, "train_loss": 2.903158, "learning_rate": 7.200803437375342e-05, "grad_norm": 0.4411883056163788, "tokens_seen": 747110400, "tokens_per_second": 266567}
|
| 143 |
+
{"step": 5750, "max_steps": 7630, "train_loss": 2.824708, "learning_rate": 6.997272801184575e-05, "grad_norm": 0.4699079990386963, "tokens_seen": 753664000, "tokens_per_second": 266563}
|
| 144 |
+
{"step": 5800, "max_steps": 7630, "train_loss": 2.90259, "learning_rate": 6.797936057371128e-05, "grad_norm": 0.4571993052959442, "tokens_seen": 760217600, "tokens_per_second": 266461}
|
| 145 |
+
{"step": 5850, "max_steps": 7630, "train_loss": 2.882249, "learning_rate": 6.602881180348344e-05, "grad_norm": 0.44385743141174316, "tokens_seen": 766771200, "tokens_per_second": 266551}
|
| 146 |
+
{"step": 5900, "max_steps": 7630, "train_loss": 2.977278, "learning_rate": 6.412194254789063e-05, "grad_norm": 0.5006775856018066, "tokens_seen": 773324800, "tokens_per_second": 266520}
|
| 147 |
+
{"step": 5950, "max_steps": 7630, "train_loss": 2.845287, "learning_rate": 6.225959437633357e-05, "grad_norm": 0.45120030641555786, "tokens_seen": 779878400, "tokens_per_second": 266547}
|
| 148 |
+
{"step": 6000, "max_steps": 7630, "train_loss": 2.866127, "learning_rate": 6.044258920947124e-05, "grad_norm": 0.4597177505493164, "tokens_seen": 786432000, "tokens_per_second": 266563}
|
| 149 |
+
{"step": 6000, "validation_loss": 2.993917, "validation_perplexity": 19.9637}
|
| 150 |
+
Saved checkpoint: runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00006000
|
| 151 |
+
{"step": 6050, "max_steps": 7630, "train_loss": 2.836868, "learning_rate": 5.867172895647775e-05, "grad_norm": 0.45854923129081726, "tokens_seen": 792985600, "tokens_per_second": 251786}
|
| 152 |
+
{"step": 6100, "max_steps": 7630, "train_loss": 2.824083, "learning_rate": 5.694779516113092e-05, "grad_norm": 0.4980098605155945, "tokens_seen": 799539200, "tokens_per_second": 266546}
|
| 153 |
+
{"step": 6150, "max_steps": 7630, "train_loss": 2.88014, "learning_rate": 5.527154865688977e-05, "grad_norm": 0.4688892662525177, "tokens_seen": 806092800, "tokens_per_second": 266537}
|
| 154 |
+
{"step": 6200, "max_steps": 7630, "train_loss": 2.859757, "learning_rate": 5.364372923111102e-05, "grad_norm": 0.45955732464790344, "tokens_seen": 812646400, "tokens_per_second": 266567}
|
| 155 |
+
{"step": 6250, "max_steps": 7630, "train_loss": 2.901738, "learning_rate": 5.206505529855481e-05, "grad_norm": 0.46400779485702515, "tokens_seen": 819200000, "tokens_per_second": 266525}
|
| 156 |
+
{"step": 6300, "max_steps": 7630, "train_loss": 2.943695, "learning_rate": 5.0536223584322686e-05, "grad_norm": 0.44685280323028564, "tokens_seen": 825753600, "tokens_per_second": 266451}
|
| 157 |
+
{"step": 6350, "max_steps": 7630, "train_loss": 2.883018, "learning_rate": 4.905790881636781e-05, "grad_norm": 0.46525463461875916, "tokens_seen": 832307200, "tokens_per_second": 266551}
|
| 158 |
+
{"step": 6400, "max_steps": 7630, "train_loss": 2.832839, "learning_rate": 4.763076342771354e-05, "grad_norm": 0.45212581753730774, "tokens_seen": 838860800, "tokens_per_second": 266440}
|
| 159 |
+
{"step": 6450, "max_steps": 7630, "train_loss": 2.885328, "learning_rate": 4.6255417268511515e-05, "grad_norm": 0.45912760496139526, "tokens_seen": 845414400, "tokens_per_second": 266589}
|
| 160 |
+
{"step": 6500, "max_steps": 7630, "train_loss": 2.983234, "learning_rate": 4.4932477328066216e-05, "grad_norm": 0.45133110880851746, "tokens_seen": 851968000, "tokens_per_second": 266571}
|
| 161 |
+
{"step": 6500, "validation_loss": 2.985335, "validation_perplexity": 19.7931}
|
| 162 |
+
{"step": 6550, "max_steps": 7630, "train_loss": 2.889001, "learning_rate": 4.366252746694898e-05, "grad_norm": 0.4614199697971344, "tokens_seen": 858521600, "tokens_per_second": 258500}
|
| 163 |
+
{"step": 6600, "max_steps": 7630, "train_loss": 2.93626, "learning_rate": 4.244612815931966e-05, "grad_norm": 0.47253450751304626, "tokens_seen": 865075200, "tokens_per_second": 266555}
|
| 164 |
+
{"step": 6650, "max_steps": 7630, "train_loss": 2.954335, "learning_rate": 4.128381624556904e-05, "grad_norm": 0.44984927773475647, "tokens_seen": 871628800, "tokens_per_second": 266566}
|
| 165 |
+
{"step": 6700, "max_steps": 7630, "train_loss": 2.946345, "learning_rate": 4.01761046953924e-05, "grad_norm": 0.47339221835136414, "tokens_seen": 878182400, "tokens_per_second": 266537}
|
| 166 |
+
{"step": 6750, "max_steps": 7630, "train_loss": 2.963953, "learning_rate": 3.912348238139758e-05, "grad_norm": 0.48386308550834656, "tokens_seen": 884736000, "tokens_per_second": 266520}
|
| 167 |
+
{"step": 6800, "max_steps": 7630, "train_loss": 2.931599, "learning_rate": 3.812641386334785e-05, "grad_norm": 0.4465729594230652, "tokens_seen": 891289600, "tokens_per_second": 266535}
|
| 168 |
+
{"step": 6850, "max_steps": 7630, "train_loss": 2.860459, "learning_rate": 3.7185339183135446e-05, "grad_norm": 0.47946491837501526, "tokens_seen": 897843200, "tokens_per_second": 266514}
|
| 169 |
+
{"step": 6900, "max_steps": 7630, "train_loss": 2.848556, "learning_rate": 3.630067367057523e-05, "grad_norm": 0.45990291237831116, "tokens_seen": 904396800, "tokens_per_second": 266533}
|
| 170 |
+
{"step": 6950, "max_steps": 7630, "train_loss": 2.883134, "learning_rate": 3.547280776010474e-05, "grad_norm": 0.44329753518104553, "tokens_seen": 910950400, "tokens_per_second": 266531}
|
| 171 |
+
{"step": 7000, "max_steps": 7630, "train_loss": 2.871083, "learning_rate": 3.4702106818471746e-05, "grad_norm": 0.46592313051223755, "tokens_seen": 917504000, "tokens_per_second": 266441}
|
| 172 |
+
{"step": 7000, "validation_loss": 2.979279, "validation_perplexity": 19.6736}
|
| 173 |
+
{"step": 7050, "max_steps": 7630, "train_loss": 2.905132, "learning_rate": 3.39889109834848e-05, "grad_norm": 0.44866663217544556, "tokens_seen": 924057600, "tokens_per_second": 258514}
|
| 174 |
+
{"step": 7100, "max_steps": 7630, "train_loss": 2.893109, "learning_rate": 3.333353501389819e-05, "grad_norm": 0.44349196553230286, "tokens_seen": 930611200, "tokens_per_second": 266566}
|
| 175 |
+
{"step": 7150, "max_steps": 7630, "train_loss": 2.861168, "learning_rate": 3.273626815049786e-05, "grad_norm": 0.460520476102829, "tokens_seen": 937164800, "tokens_per_second": 266564}
|
| 176 |
+
{"step": 7200, "max_steps": 7630, "train_loss": 2.995534, "learning_rate": 3.2197373988448944e-05, "grad_norm": 0.4440070390701294, "tokens_seen": 943718400, "tokens_per_second": 266546}
|
| 177 |
+
{"step": 7250, "max_steps": 7630, "train_loss": 2.923584, "learning_rate": 3.171709036096198e-05, "grad_norm": 0.4627412259578705, "tokens_seen": 950272000, "tokens_per_second": 266550}
|
| 178 |
+
{"step": 7300, "max_steps": 7630, "train_loss": 2.854573, "learning_rate": 3.129562923432856e-05, "grad_norm": 0.44265109300613403, "tokens_seen": 956825600, "tokens_per_second": 266553}
|
| 179 |
+
{"step": 7350, "max_steps": 7630, "train_loss": 2.946622, "learning_rate": 3.0933176614373046e-05, "grad_norm": 0.4477914571762085, "tokens_seen": 963379200, "tokens_per_second": 266556}
|
| 180 |
+
{"step": 7400, "max_steps": 7630, "train_loss": 2.804573, "learning_rate": 3.062989246436156e-05, "grad_norm": 0.44815805554389954, "tokens_seen": 969932800, "tokens_per_second": 266558}
|
| 181 |
+
{"step": 7450, "max_steps": 7630, "train_loss": 2.922463, "learning_rate": 3.0385910634404578e-05, "grad_norm": 0.4610517621040344, "tokens_seen": 976486400, "tokens_per_second": 266534}
|
| 182 |
+
{"step": 7500, "max_steps": 7630, "train_loss": 2.829884, "learning_rate": 3.020133880238392e-05, "grad_norm": 0.4553583860397339, "tokens_seen": 983040000, "tokens_per_second": 266555}
|
| 183 |
+
{"step": 7500, "validation_loss": 2.97534, "validation_perplexity": 19.5963}
|
| 184 |
+
{"step": 7550, "max_steps": 7630, "train_loss": 2.854886, "learning_rate": 3.0076258426430874e-05, "grad_norm": 0.4332123398780823, "tokens_seen": 989593600, "tokens_per_second": 258499}
|
| 185 |
+
{"step": 7600, "max_steps": 7630, "train_loss": 2.908676, "learning_rate": 3.001072470897566e-05, "grad_norm": 0.44882938265800476, "tokens_seen": 996147200, "tokens_per_second": 266556}
|
| 186 |
+
Final checkpoint: runs/kawk50m-fixed-recovery-1b-5090/checkpoint-00007630
|
reports/fixed-recovery-6b/evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-6b-5090/checkpoint-00038147",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-fixed-recovery-6b-5090/checkpoint-00038147/tokenizer",
|
| 4 |
+
"validation_loss": 2.8902571320533754,
|
| 5 |
+
"validation_perplexity": 17.99793685307703,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "서울이다.\n김영철은 1975년 8월 26일, 미국 뉴욕에서 태어났다. 10살 때인 1949년 3월 23일에 태어났고, 10세 때부터 대학교수로 일했다. 김영철이 중학교를 졸업한 1975년에는 미대 교수로 승진해 3년간 박사 학위를 취득할 수 있었다.\n김영철은 서울대 법학전문대학원을 나와, 1977년 사법시험에 합격해 서울지방변호사회 감사위원으로 활동하다가, 1979년 4월 박근혜 정부 때 국가인권위원회 부위원장으로",
|
| 11 |
+
"full_text": "대한민국의 수도는서울이다.\n김영철은 1975년 8월 26일, 미국 뉴욕에서 태어났다. 10살 때인 1949년 3월 23일에 태어났고, 10세 때부터 대학교수로 일했다. 김영철이 중학교를 졸업한 1975년에는 미대 교수로 승진해 3년간 박사 학위를 취득할 수 있었다.\n김영철은 서울대 법학전문대학원을 나와, 1977년 사법시험에 합격해 서울지방변호사회 감사위원으로 활동하다가, 1979년 4월 박근혜 정부 때 국가인권위원회 부위원장으로"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "퇴근하고 나면 30분은 족히 걸리겠네\n이런 생각도 해봤는데.........\n2016-07-24 15:58:09\n안녕하세요. 저는 오늘 아침에 일어나서 1시간 정도는 걸을 예정인데요.\n오늘 아침에 일어나서 1시간 정도는 걸을 수 있는 좋은 방법인 것 같습니다.\n그리고 제가 오늘 아침에는 헬스장에 갈 때 운동을 하는 것은 괜찮지만, 운동 전 1시간 정도는 걷기나 자전거 타기를 하시는 것도 괜찮습니다.\n아마 지금처럼 비가 오고 내리는 날씨라면 헬스장에 가는 것을 추천해 드립니다.\n그런데",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서퇴근하고 나면 30분은 족히 걸리겠네\n이런 생각도 해봤는데.........\n2016-07-24 15:58:09\n안녕하세요. 저는 오늘 아침에 일어나서 1시간 정도는 걸을 예정인데요.\n오늘 아침에 일어나서 1시간 정도는 걸을 수 있는 좋은 방법인 것 같습니다.\n그리고 제가 오늘 아침에는 헬스장에 갈 때 운동을 하는 것은 괜찮지만, 운동 전 1시간 정도는 걷기나 자전거 타기를 하시는 것도 괜찮습니다.\n아마 지금처럼 비가 오고 내리는 날씨라면 헬스장에 가는 것을 추천해 드립니다.\n그런데"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "2017년 9월 3일(목)부터 2018년 5월 6일(금)까지 영국에서 개최되는 제4회 AI Science Cup에 참가한다.\n이번 대회는 인공지능의 발전으로 인한 인간의 지능 향상과 함께 AI의 활용도가 확대되고 있는 가운데, 이번 대회를 통해 인공지능이 인공지능 기술의 혁명을 이루고 세계인에게 AI를 알리는 계기가 되길 바란다고 밝혔다.\n또한, 이번 대회에서 AI를 활용한 다양한 기술들을 선보이며, 인공지능이 제공하는 다양한 교육 콘텐츠를 통해 미래의 인재들이 AI 기술을 이해하고 적용하는 기회가 될 것으로 기대된다.\n이번 대회에 참가한 학생은 “제전의 인공지능",
|
| 21 |
+
"full_text": "인공지능 언어 모델은2017년 9월 3일(목)부터 2018년 5월 6일(금)까지 영국에서 개최되는 제4회 AI Science Cup에 참가한다.\n이번 대회는 인공지능의 발전으로 인한 인간의 지능 향상과 함께 AI의 활용도가 확대되고 있는 가운데, 이번 대회를 통해 인공지능이 인공지능 기술의 혁명을 이루고 세계인에게 AI를 알리는 계기가 되길 바란다고 밝혔다.\n또한, 이번 대회에서 AI를 활용한 다양한 기술들을 선보이며, 인공지능이 제공하는 다양한 교육 콘텐츠를 통해 미래의 인재들이 AI 기술을 이해하고 적용하는 기회가 될 것으로 기대된다.\n이번 대회에 참가한 학생은 “제전의 인공지능"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "텐트를 치거나, 흙을 뒤집어 쓴 후 빗물을 이용해 실내에서 습기를 제거하고 건조하는 등의 방법으로 실내 습도를 유지하는 것이 좋다.\n특히, 봄철에 비가 많이 오면 건조한 공기로 인해 습도 조절이 어려워지기 때문에 실내습도가 높아질수록 곰팡이 병균의 증식이 활발해지면서 수증기가 발생하는 세균성 폐렴으로 발전할 가능성이 높아지기 때문이다.\n또한, 봄철 기온이 높아져 환절기에 알레르기 반응이 심해질 경우 고온다습한 환경으로 인해 세균성 폐렴의 발병률이 증가하여 2차 감염이 발생할 확률이 높으며, 봄철에 유행할 수",
|
| 26 |
+
"full_text": "비가 내리는 날에는텐트를 치거나, 흙을 뒤집어 쓴 후 빗물을 이용해 실내에서 습기를 제거하고 건조하는 등의 방법으로 실내 습도를 유지하는 것이 좋다.\n특히, 봄철에 비가 많이 오면 건조한 공기로 인해 습도 조절이 어려워지기 때문에 실내습도가 높아질수록 곰팡이 병균의 증식이 활발해지면서 수증기가 발생하는 세균성 폐렴으로 발전할 가능성이 높아지기 때문이다.\n또한, 봄철 기온이 높아져 환절기에 알레르기 반응이 심해질 경우 고온다습한 환경으로 인해 세균성 폐렴의 발병률이 증가하여 2차 감염이 발생할 확률이 높으며, 봄철에 유행할 수"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/fixed-recovery-6b/gate.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"passed": true,
|
| 3 |
+
"candidate_loss": 2.8902571320533754,
|
| 4 |
+
"baseline_loss": 2.974478209018707,
|
| 5 |
+
"loss_improvement": 0.08422107696533176,
|
| 6 |
+
"sample_count": 4
|
| 7 |
+
}
|
reports/kawk-1.5-50m-v2/MODEL_CARD.md
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- ko
|
| 4 |
+
pipeline_tag: text-generation
|
| 5 |
+
library_name: transformers
|
| 6 |
+
tags:
|
| 7 |
+
- llama
|
| 8 |
+
- korean
|
| 9 |
+
- base-model
|
| 10 |
+
- long-context
|
| 11 |
+
---
|
| 12 |
+
|
| 13 |
+
# KAWK-1.5-50M Korean Base
|
| 14 |
+
|
| 15 |
+
KAWK-1.5-50M은 한국어 중심 문서로 처음부터 학습한 51,542,528
|
| 16 |
+
파라미터의 decoder-only Llama 계열 BASE 모델입니다. 1,024 토큰
|
| 17 |
+
컨텍스트에서 20B 토큰 사전학습을 수행한 뒤, 같은 가중치를 5,120 토큰
|
| 18 |
+
컨텍스트로 확장하여 3B 토큰을 추가 학습했습니다.
|
| 19 |
+
|
| 20 |
+
이 모델은 지시 튜닝 모델이 아닙니다. 입력 뒤의 텍스트를 이어 쓰는 BASE
|
| 21 |
+
모델이며, 질문에 정확히 답하거나 안전한 대화를 보장하지 않습니다.
|
| 22 |
+
|
| 23 |
+
## Architecture
|
| 24 |
+
|
| 25 |
+
| Item | Value |
|
| 26 |
+
|---|---:|
|
| 27 |
+
| Parameters | 51,542,528 |
|
| 28 |
+
| Vocabulary | 20,000 |
|
| 29 |
+
| Layers | 14 |
|
| 30 |
+
| Hidden / MLP | 512 / 1,408 |
|
| 31 |
+
| Attention / KV heads | 8 / 4 |
|
| 32 |
+
| Head dimension | 64 |
|
| 33 |
+
| Context | 5,120 |
|
| 34 |
+
| Position encoding | RoPE, linear factor 5 |
|
| 35 |
+
| Activation / norm | SwiGLU(SiLU) / RMSNorm |
|
| 36 |
+
| Tied embeddings | yes |
|
| 37 |
+
|
| 38 |
+
The tokenizer is a Korean-focused SentencePiece Unigram tokenizer with NFC
|
| 39 |
+
input normalization and byte fallback. Dedicated English, source-code, and
|
| 40 |
+
mathematics corpora were not used. Latin letters and numbers naturally
|
| 41 |
+
embedded in Korean documents were retained.
|
| 42 |
+
|
| 43 |
+
## Training
|
| 44 |
+
|
| 45 |
+
- Base checkpoint: `/workspace/KAWK50M/runs/kawk50m-v2-20b-5090/checkpoint-00152588`
|
| 46 |
+
- Final checkpoint: `/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415`
|
| 47 |
+
- Base corpus tokens: 20,000,000,000
|
| 48 |
+
- Base pretraining budget: 20,000,000,000 tokens at context 1,024
|
| 49 |
+
- Long-context CPT mix: 3,000,000,000 tokens
|
| 50 |
+
- Long-context CPT budget: 3,000,000,000 tokens at context 5,120
|
| 51 |
+
- Precision: BF16
|
| 52 |
+
- Hardware: one NVIDIA GeForce RTX 5090
|
| 53 |
+
|
| 54 |
+
The long-context mix contains 2.9B Korean web/document tokens and 100M
|
| 55 |
+
strictly filtered structured Korean tokens:
|
| 56 |
+
|
| 57 |
+
- Korean RAG: 50,000,000 tokens,
|
| 58 |
+
accepted rows 35,723
|
| 59 |
+
- Korean chat: 50,000,000 tokens,
|
| 60 |
+
accepted rows 203,289
|
| 61 |
+
|
| 62 |
+
ChatML markers were converted to plain Korean role labels. Rows containing
|
| 63 |
+
code fences, LaTeX patterns, code/math-heavy text, disallowed licenses, or
|
| 64 |
+
English-heavy content were rejected.
|
| 65 |
+
|
| 66 |
+
## Data lineage and attribution
|
| 67 |
+
|
| 68 |
+
- `HuggingFaceFW/fineweb-2`, subset `kor_Hang`, revision
|
| 69 |
+
`af9c13333eb981300149d5ca60a8e9d659b276b9`
|
| 70 |
+
(ODC-By 1.0; Common Crawl terms also apply)
|
| 71 |
+
- `mkd-chanwoo/keural-rag-chatml-ko`, revision
|
| 72 |
+
`aa9023d231a3451d25135d5aa84b12b48ca06450`
|
| 73 |
+
(accepted rows: CC-BY-4.0)
|
| 74 |
+
- `Mkd-Yonas/keural-SFT-chatml-ko-v1`, revision
|
| 75 |
+
`c56d7885efb37799deb1ecba471e4f6bf263852e`
|
| 76 |
+
(only rows with explicitly allowed permissive licenses were accepted)
|
| 77 |
+
|
| 78 |
+
Users distributing the model should preserve these attributions and review
|
| 79 |
+
the upstream dataset terms. The project source code license and the model
|
| 80 |
+
weight release license should be stated separately at publication time.
|
| 81 |
+
|
| 82 |
+
## Evaluation
|
| 83 |
+
|
| 84 |
+
| Checkpoint | Context | Validation loss | Perplexity | Bits / UTF-8 byte |
|
| 85 |
+
|---|---:|---:|---:|---:|
|
| 86 |
+
| 20B base | 1,024 | 4.431828 | 84.084995 | 2.276136 |
|
| 87 |
+
| 3B long CPT | 5,120 | 4.428928 | 83.841481 | 2.274364 |
|
| 88 |
+
|
| 89 |
+
`bits / UTF-8 byte` is reported on the fixed ordered Korean validation
|
| 90 |
+
documents. Generation samples are BASE-model continuations, not instruction
|
| 91 |
+
following scores.
|
| 92 |
+
|
| 93 |
+
## Limitations
|
| 94 |
+
|
| 95 |
+
- At 50M parameters, factual coverage and multi-step reasoning are limited.
|
| 96 |
+
- The model may generate incorrect, repetitive, biased, private, or unsafe text.
|
| 97 |
+
- Web data can contain errors and cultural or temporal bias despite filtering.
|
| 98 |
+
- A 5,120-token configuration does not guarantee reliable retrieval across the
|
| 99 |
+
entire context; task-specific long-context evaluation is still required.
|
| 100 |
+
- Do not use the model as an authority for medical, legal, financial, or safety
|
| 101 |
+
decisions.
|
reports/kawk-1.5-50m-v2/byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 1000,
|
| 8 |
+
"characters": 1657768,
|
| 9 |
+
"utf8_bytes": 3936686,
|
| 10 |
+
"predicted_tokens": 867888,
|
| 11 |
+
"context_length": 5120,
|
| 12 |
+
"token_nll": 7.150765165287452,
|
| 13 |
+
"token_perplexity": 1275.0812298984872,
|
| 14 |
+
"nats_per_character": 3.7436259342507494,
|
| 15 |
+
"bits_per_utf8_byte": 2.274363948326923,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/kawk-1.5-50m-v2/checksums.sha256
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
8a702fc33ca29cb667106888b7546e576246d3eab5a9c8d7a4284541d24d60ff model/model.safetensors
|
| 2 |
+
12ce9fb639a4d679b1b19765150630dfd5f4a27fb8934302e78612acdc76cbb4 trainer_state.pt
|
reports/kawk-1.5-50m-v2/evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415/tokenizer",
|
| 4 |
+
"validation_loss": 4.428927886486053,
|
| 5 |
+
"validation_perplexity": 83.84148112325552,
|
| 6 |
+
"evaluated_tokens": 409600,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "이다\n .175년623일여에어게()를수한,80년901명이가진하고40명이신을인한다.이은기자k.%고했다",
|
| 11 |
+
"full_text": "대한민국의 수도는이다\n .175년623일여에어게()를수한,80년901명이가진하고40명이신을인한다.이은기자k.%고했다"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "을.\n이는이 0215785의 01-3497826025에서한이다\n 0133010000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서을.\n이는이 0215785의 01-3497826025에서한이다\n 0133010000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "의을한로.이는어다게가그“”,아바드,“””시니”“,르와””에서생.0%3%)21(8.정경교수 “상적의에서된만과적이지고인이있는은를하는신의를수한다.●출장안마일강도기으로해되는사이다‘’말있다며공당부격인원명등보했다박현기자전출장샵성대병원내김희원장“제의필요않나.●출장마사지동일보지난월일후중국진대자중의위화회주오5여간연한선였다그“”무인가”””떤?””?문발””유브””…장영은의이재에게려“렇할이없”“이”아니””것.씨는씨가에집들어던에스를단히껴었다\n서울",
|
| 21 |
+
"full_text": "인공지능 언어 모델은의을한로.이는어다게가그“”,아바드,“””시니”“,르와””에서생.0%3%)21(8.정경교수 “상적의에서된만과적이지고인이있는은를하는신의를수한다.●출장안마일강도기으로해되는사이다‘’말있다며공당부격인원명등보했다박현기자전출장샵성대병원내김희원장“제의필요않나.●출장마사지동일보지난월일후중국진대자중의위화회주오5여간연한선였다그“”무인가”””떤?””?문발””유브””…장영은의이재에게려“렇할이없”“이”아니””것.씨는씨가에집들어던에스를단히껴었다\n서울"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "볕 잘드는에을다\n이게야로!이야!!!아리리가.어게~고~~~구!!\n시럽 0:93145257\n은히지\n:\n이야아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아",
|
| 26 |
+
"full_text": "비가 내리는 날에는볕 잘드는에을다\n이게야로!이야!!!아리리가.어게~고~~~구!!\n시럽 0:93145257\n은히지\n:\n이야아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아아"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/kawk-1.5-50m-v2/final_checkpoint.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
runs/kawk-1.5-50m-v2-cpt-3b-5090/checkpoint-00024415
|
reports/kawk-1.5-50m-v2/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
reports/kawk50m-ko-instruct-v1/checksums.sha256
ADDED
|
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
|
|
|
| 1 |
+
a1eac07bf0680d719441f5c1a72b704d0ae07e271f9fbb55c41a7e24eac59bfe model/model.safetensors
|
| 2 |
+
ac06a9708915194265b11f7eedb33eac485e5212d869905ea610fce0ddc03113 trainer_state.pt
|
reports/kawk50m-ko-instruct-v1/evaluation.json
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-ko-instruct-v1-5090/checkpoint-00001303",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-ko-instruct-v1-5090/checkpoint-00001303/tokenizer",
|
| 4 |
+
"validation_loss": 2.321558953715487,
|
| 5 |
+
"validation_perplexity": 10.191550082925739,
|
| 6 |
+
"evaluated_supervised_tokens": 341104,
|
| 7 |
+
"assistant_only_loss": true,
|
| 8 |
+
"samples": [
|
| 9 |
+
{
|
| 10 |
+
"prompt": "대한민국의 수도는 어디인가요? 한 문장으로 답해주세요.",
|
| 11 |
+
"response": "대한민국의 수도는 서울특별시입니다.",
|
| 12 |
+
"metrics": {
|
| 13 |
+
"characters": 19,
|
| 14 |
+
"hangul_visible_ratio": 0.9411764705882353,
|
| 15 |
+
"unique_word_ratio": 1.0
|
| 16 |
+
}
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "비가 오는 날 외출할 때 준비할 물건 세 가지를 알려주세요.",
|
| 20 |
+
"response": "비가 오는 날 외출 시 준비해야 할 세 가지 물품은 다음과 같습니다:\n1. 모자: 모자는 비와 함께 외출할 때 필수품입니다. 모자는 햇빛을 차단하고, 빗자루를 사용하여 빗방울이 떨어지지 않도록 합니다.\n2. 슬리퍼: 슬리퍼는 비와 함께 외출할 때 필수품입니다. 슬리퍼는 빗물이 고여 있는 곳에 두어 비가 올 때 빗물을 흡수합니다.\n3. 모자: 모자는 비와 함께 외출할 때 필수품입니다. 모자는 빗물이 고여 있는 곳에 두어 비가 올 때 빗물을 흡수합니다.",
|
| 21 |
+
"metrics": {
|
| 22 |
+
"characters": 253,
|
| 23 |
+
"hangul_visible_ratio": 0.9090909090909091,
|
| 24 |
+
"unique_word_ratio": 0.5970149253731343
|
| 25 |
+
}
|
| 26 |
+
},
|
| 27 |
+
{
|
| 28 |
+
"prompt": "다음 문장을 자연스럽게 고쳐주세요: 나는 어제 학교에 갑니다.",
|
| 29 |
+
"response": "선생님은 오늘 학교에 가는데, 내일 학교에 갈 수 있을지 모르겠어요.",
|
| 30 |
+
"metrics": {
|
| 31 |
+
"characters": 38,
|
| 32 |
+
"hangul_visible_ratio": 0.9310344827586207,
|
| 33 |
+
"unique_word_ratio": 0.9
|
| 34 |
+
}
|
| 35 |
+
},
|
| 36 |
+
{
|
| 37 |
+
"prompt": "한국어로 자기소개를 두 문장으로 해주세요.",
|
| 38 |
+
"response": "1. 안녕하세요, 저는 한국어로 자기소개를 한 번 더 하겠습니다.\n2. 한국어로 자기소개를 한 번 더 하겠습니다.\n3. 한국어로 자기소개를 한 번 더 하겠습니다.",
|
| 39 |
+
"metrics": {
|
| 40 |
+
"characters": 90,
|
| 41 |
+
"hangul_visible_ratio": 0.8529411764705882,
|
| 42 |
+
"unique_word_ratio": 0.4782608695652174
|
| 43 |
+
}
|
| 44 |
+
},
|
| 45 |
+
{
|
| 46 |
+
"prompt": "사용자가 위험한 요청을 하면 어떻게 대응해야 하나요?",
|
| 47 |
+
"response": "사용자가 요청을 할 때, 사용자는 자신의 행동에 대해 책임을 져야 합니다. 여기에는 다음이 포함됩니다:\n1. 사용자의 요청을 거부합니다: 사용자는 요청을 거부할 수 있습니다. 이는 요청을 거부하는 것이 아니라 요청을 거부하는 것입니다.\n2. 사용자의 동의를 구합니다: 사용자는 요청을 거부할 권리가 있으며, 요청을 거부하면 해당 요청에 대한 책임을 지게 됩니다.\n3. 요청을 거부합니다: 요청을 거부하는 것은 사용자의 권리입니다. 요청을 거부하면 해당 요청에 대한 책임을 지게 됩니다.\n4. 요청을 거부합니다: 요청을 거부하는 것은 사용자의 권리입니다. 요청을 거부하면 해당 요청에 대한 책임을 지게 됩니다.\n5. 요청을 거부합니다: 요청을",
|
| 48 |
+
"metrics": {
|
| 49 |
+
"characters": 358,
|
| 50 |
+
"hangul_visible_ratio": 0.9054545454545454,
|
| 51 |
+
"unique_word_ratio": 0.4880952380952381
|
| 52 |
+
}
|
| 53 |
+
}
|
| 54 |
+
],
|
| 55 |
+
"note": "Korean instruction evaluation; responses are sampled, not facts checked."
|
| 56 |
+
}
|
reports/kawk50m-ko-instruct-v1/final_checkpoint.txt
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
runs/kawk50m-ko-instruct-v1-5090/checkpoint-00001303
|
reports/kawk50m-ko-instruct-v1/gate.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"passed": true,
|
| 3 |
+
"candidate_loss": 2.321558953715487,
|
| 4 |
+
"baseline_loss": 2.54827207645266,
|
| 5 |
+
"loss_improvement": 0.22671312273717303,
|
| 6 |
+
"sample_count": 5
|
| 7 |
+
}
|
reports/v2-100m/train.log
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"tokens_per_optimizer_step": 131072,
|
| 7 |
+
"max_steps": 763,
|
| 8 |
+
"planned_tokens": 100007936,
|
| 9 |
+
"start_step": 0,
|
| 10 |
+
"initialized_from": null,
|
| 11 |
+
"precision": "bf16",
|
| 12 |
+
"compile": true
|
| 13 |
+
}
|
| 14 |
+
{"step": 1, "max_steps": 763, "train_loss": 10.009621, "learning_rate": 3.9999999999999996e-05, "grad_norm": 1.4620524644851685, "tokens_seen": 131072, "tokens_per_second": 6282}
|
| 15 |
+
{"step": 10, "max_steps": 763, "train_loss": 8.783953, "learning_rate": 0.00039999999999999996, "grad_norm": 1.342725396156311, "tokens_seen": 1310720, "tokens_per_second": 266937}
|
| 16 |
+
{"step": 20, "max_steps": 763, "train_loss": 7.416108, "learning_rate": 0.0005999404674662951, "grad_norm": 0.4180927574634552, "tokens_seen": 2621440, "tokens_per_second": 266958}
|
| 17 |
+
{"step": 30, "max_steps": 763, "train_loss": 7.311667, "learning_rate": 0.0005994643647016376, "grad_norm": 0.2270386517047882, "tokens_seen": 3932160, "tokens_per_second": 267011}
|
| 18 |
+
{"step": 40, "max_steps": 763, "train_loss": 7.292147, "learning_rate": 0.0005985129988902226, "grad_norm": 0.1775497943162918, "tokens_seen": 5242880, "tokens_per_second": 266987}
|
| 19 |
+
{"step": 50, "max_steps": 763, "train_loss": 7.282208, "learning_rate": 0.000597088047986813, "grad_norm": 0.14131352305412292, "tokens_seen": 6553600, "tokens_per_second": 266921}
|
| 20 |
+
{"step": 60, "max_steps": 763, "train_loss": 7.293289, "learning_rate": 0.0005951920252235705, "grad_norm": 0.17815741896629333, "tokens_seen": 7864320, "tokens_per_second": 266923}
|
| 21 |
+
{"step": 70, "max_steps": 763, "train_loss": 7.284279, "learning_rate": 0.0005928282746773871, "grad_norm": 0.15476952493190765, "tokens_seen": 9175040, "tokens_per_second": 266861}
|
| 22 |
+
{"step": 80, "max_steps": 763, "train_loss": 7.294957, "learning_rate": 0.0005900009653718274, "grad_norm": 0.13445036113262177, "tokens_seen": 10485760, "tokens_per_second": 266856}
|
| 23 |
+
{"step": 90, "max_steps": 763, "train_loss": 7.268172, "learning_rate": 0.0005867150839240876, "grad_norm": 0.15143540501594543, "tokens_seen": 11796480, "tokens_per_second": 266831}
|
| 24 |
+
{"step": 100, "max_steps": 763, "train_loss": 7.238208, "learning_rate": 0.0005829764257499357, "grad_norm": 0.5955350399017334, "tokens_seen": 13107200, "tokens_per_second": 266742}
|
| 25 |
+
{"step": 100, "validation_loss": 7.236411, "validation_perplexity": 1389.0995}
|
| 26 |
+
{"step": 110, "max_steps": 763, "train_loss": 7.157025, "learning_rate": 0.0005787915848421489, "grad_norm": 0.17872370779514313, "tokens_seen": 14417920, "tokens_per_second": 83752}
|
| 27 |
+
{"step": 120, "max_steps": 763, "train_loss": 7.11413, "learning_rate": 0.0005741679421404754, "grad_norm": 0.410422682762146, "tokens_seen": 15728640, "tokens_per_second": 266650}
|
| 28 |
+
{"step": 130, "max_steps": 763, "train_loss": 7.080323, "learning_rate": 0.0005691136525136301, "grad_norm": 0.2681965231895447, "tokens_seen": 17039360, "tokens_per_second": 266573}
|
| 29 |
+
{"step": 140, "max_steps": 763, "train_loss": 6.99782, "learning_rate": 0.0005636376303762884, "grad_norm": 0.3821214437484741, "tokens_seen": 18350080, "tokens_per_second": 266519}
|
| 30 |
+
{"step": 150, "max_steps": 763, "train_loss": 6.9835, "learning_rate": 0.0005577495339664442, "grad_norm": 0.2671792805194855, "tokens_seen": 19660800, "tokens_per_second": 266478}
|
| 31 |
+
{"step": 160, "max_steps": 763, "train_loss": 6.943997, "learning_rate": 0.0005514597483108611, "grad_norm": 0.23311127722263336, "tokens_seen": 20971520, "tokens_per_second": 266393}
|
| 32 |
+
{"step": 170, "max_steps": 763, "train_loss": 6.898273, "learning_rate": 0.0005447793669086647, "grad_norm": 0.19526070356369019, "tokens_seen": 22282240, "tokens_per_second": 266340}
|
| 33 |
+
{"step": 180, "max_steps": 763, "train_loss": 6.879401, "learning_rate": 0.0005377201721653773, "grad_norm": 0.4148119390010834, "tokens_seen": 23592960, "tokens_per_second": 266410}
|
| 34 |
+
{"step": 190, "max_steps": 763, "train_loss": 6.842384, "learning_rate": 0.0005302946146119068, "grad_norm": 0.3028162717819214, "tokens_seen": 24903680, "tokens_per_second": 266288}
|
| 35 |
+
{"step": 200, "max_steps": 763, "train_loss": 6.829096, "learning_rate": 0.0005225157909451416, "grad_norm": 0.215122252702713, "tokens_seen": 26214400, "tokens_per_second": 266278}
|
| 36 |
+
{"step": 200, "validation_loss": 6.829778, "validation_perplexity": 924.9853}
|
| 37 |
+
{"step": 210, "max_steps": 763, "train_loss": 6.789659, "learning_rate": 0.0005143974209288811, "grad_norm": 0.24126510322093964, "tokens_seen": 27525120, "tokens_per_second": 230320}
|
| 38 |
+
{"step": 220, "max_steps": 763, "train_loss": 6.741691, "learning_rate": 0.0005059538231958432, "grad_norm": 0.360088586807251, "tokens_seen": 28835840, "tokens_per_second": 266270}
|
| 39 |
+
{"step": 230, "max_steps": 763, "train_loss": 6.764729, "learning_rate": 0.0004971998899934259, "grad_norm": 0.31751716136932373, "tokens_seen": 30146560, "tokens_per_second": 266266}
|
| 40 |
+
{"step": 240, "max_steps": 763, "train_loss": 6.741962, "learning_rate": 0.00048815106091776853, "grad_norm": 0.42814502120018005, "tokens_seen": 31457280, "tokens_per_second": 266329}
|
| 41 |
+
{"step": 250, "max_steps": 763, "train_loss": 6.656931, "learning_rate": 0.0004788232956824341, "grad_norm": 0.22453273832798004, "tokens_seen": 32768000, "tokens_per_second": 266288}
|
| 42 |
+
Saved checkpoint: runs/kawk50m-v2-100m-5090/checkpoint-00000250
|
| 43 |
+
{"step": 260, "max_steps": 763, "train_loss": 6.66519, "learning_rate": 0.00046923304596974483, "grad_norm": 0.3134668469429016, "tokens_seen": 34078720, "tokens_per_second": 227474}
|
| 44 |
+
{"step": 270, "max_steps": 763, "train_loss": 6.632088, "learning_rate": 0.0004593972264144153, "grad_norm": 0.2917993366718292, "tokens_seen": 35389440, "tokens_per_second": 266351}
|
| 45 |
+
{"step": 280, "max_steps": 763, "train_loss": 6.602264, "learning_rate": 0.00044933318477066334, "grad_norm": 0.37740278244018555, "tokens_seen": 36700160, "tokens_per_second": 266413}
|
| 46 |
+
{"step": 290, "max_steps": 763, "train_loss": 6.588042, "learning_rate": 0.00043905867131541155, "grad_norm": 0.20811189711093903, "tokens_seen": 38010880, "tokens_per_second": 266419}
|
| 47 |
+
{"step": 300, "max_steps": 763, "train_loss": 6.58299, "learning_rate": 0.0004285918075415489, "grad_norm": 0.3652443587779999, "tokens_seen": 39321600, "tokens_per_second": 266461}
|
| 48 |
+
{"step": 300, "validation_loss": 6.588824, "validation_perplexity": 726.9258}
|
| 49 |
+
{"step": 310, "max_steps": 763, "train_loss": 6.549357, "learning_rate": 0.00041795105419646444, "grad_norm": 0.36356320977211, "tokens_seen": 40632320, "tokens_per_second": 230333}
|
| 50 |
+
{"step": 320, "max_steps": 763, "train_loss": 6.534962, "learning_rate": 0.00040715517872222867, "grad_norm": 0.2735288143157959, "tokens_seen": 41943040, "tokens_per_second": 266421}
|
| 51 |
+
{"step": 330, "max_steps": 763, "train_loss": 6.533208, "learning_rate": 0.0003962232221548463, "grad_norm": 0.21965216100215912, "tokens_seen": 43253760, "tokens_per_second": 266385}
|
| 52 |
+
{"step": 340, "max_steps": 763, "train_loss": 6.547342, "learning_rate": 0.00038517446554096337, "grad_norm": 0.36226192116737366, "tokens_seen": 44564480, "tokens_per_second": 266459}
|
| 53 |
+
{"step": 350, "max_steps": 763, "train_loss": 6.510436, "learning_rate": 0.0003740283959312592, "grad_norm": 0.2593676745891571, "tokens_seen": 45875200, "tokens_per_second": 266481}
|
| 54 |
+
{"step": 360, "max_steps": 763, "train_loss": 6.470444, "learning_rate": 0.0003628046720105023, "grad_norm": 0.4364893436431885, "tokens_seen": 47185920, "tokens_per_second": 266498}
|
| 55 |
+
{"step": 370, "max_steps": 763, "train_loss": 6.446433, "learning_rate": 0.00035152308942489074, "grad_norm": 0.2311142534017563, "tokens_seen": 48496640, "tokens_per_second": 266445}
|
| 56 |
+
{"step": 380, "max_steps": 763, "train_loss": 6.462543, "learning_rate": 0.0003402035458678273, "grad_norm": 0.3035867214202881, "tokens_seen": 49807360, "tokens_per_second": 266505}
|
| 57 |
+
{"step": 390, "max_steps": 763, "train_loss": 6.435027, "learning_rate": 0.00032886600598571277, "grad_norm": 0.3535631597042084, "tokens_seen": 51118080, "tokens_per_second": 266438}
|
| 58 |
+
{"step": 400, "max_steps": 763, "train_loss": 6.447761, "learning_rate": 0.0003175304661656503, "grad_norm": 0.30540987849235535, "tokens_seen": 52428800, "tokens_per_second": 266490}
|
| 59 |
+
{"step": 400, "validation_loss": 6.445302, "validation_perplexity": 629.7368}
|
| 60 |
+
{"step": 410, "max_steps": 763, "train_loss": 6.413681, "learning_rate": 0.00030621691926716884, "grad_norm": 0.23910918831825256, "tokens_seen": 53739520, "tokens_per_second": 230377}
|
| 61 |
+
{"step": 420, "max_steps": 763, "train_loss": 6.407767, "learning_rate": 0.0002949453193601676, "grad_norm": 0.30284398794174194, "tokens_seen": 55050240, "tokens_per_second": 266502}
|
| 62 |
+
{"step": 430, "max_steps": 763, "train_loss": 6.394376, "learning_rate": 0.0002837355465312754, "grad_norm": 0.36555153131484985, "tokens_seen": 56360960, "tokens_per_second": 266509}
|
| 63 |
+
{"step": 440, "max_steps": 763, "train_loss": 6.365705, "learning_rate": 0.00027260737182069626, "grad_norm": 0.34812939167022705, "tokens_seen": 57671680, "tokens_per_second": 266496}
|
| 64 |
+
{"step": 450, "max_steps": 763, "train_loss": 6.374506, "learning_rate": 0.00026158042235138474, "grad_norm": 0.3512547016143799, "tokens_seen": 58982400, "tokens_per_second": 266513}
|
| 65 |
+
{"step": 460, "max_steps": 763, "train_loss": 6.370841, "learning_rate": 0.00025067414671205103, "grad_norm": 0.3154039680957794, "tokens_seen": 60293120, "tokens_per_second": 266492}
|
| 66 |
+
{"step": 470, "max_steps": 763, "train_loss": 6.351014, "learning_rate": 0.00023990778065505504, "grad_norm": 0.2644787132740021, "tokens_seen": 61603840, "tokens_per_second": 266465}
|
| 67 |
+
{"step": 480, "max_steps": 763, "train_loss": 6.349862, "learning_rate": 0.00022930031316968416, "grad_norm": 0.2982809543609619, "tokens_seen": 62914560, "tokens_per_second": 266482}
|
| 68 |
+
{"step": 490, "max_steps": 763, "train_loss": 6.357736, "learning_rate": 0.00021887045299065797, "grad_norm": 0.2821158468723297, "tokens_seen": 64225280, "tokens_per_second": 266476}
|
| 69 |
+
{"step": 500, "max_steps": 763, "train_loss": 6.325619, "learning_rate": 0.00020863659560092397, "grad_norm": 0.42856109142303467, "tokens_seen": 65536000, "tokens_per_second": 266431}
|
| 70 |
+
{"step": 500, "validation_loss": 6.337917, "validation_perplexity": 565.617}
|
| 71 |
+
Saved checkpoint: runs/kawk50m-v2-100m-5090/checkpoint-00000500
|
| 72 |
+
{"step": 510, "max_steps": 763, "train_loss": 6.320778, "learning_rate": 0.00019861679078694882, "grad_norm": 0.3187156319618225, "tokens_seen": 66846720, "tokens_per_second": 198923}
|
| 73 |
+
{"step": 520, "max_steps": 763, "train_loss": 6.298001, "learning_rate": 0.0001888287108037225, "grad_norm": 0.31988218426704407, "tokens_seen": 68157440, "tokens_per_second": 265898}
|
| 74 |
+
{"step": 530, "max_steps": 763, "train_loss": 6.256952, "learning_rate": 0.00017928961920562968, "grad_norm": 0.4417673647403717, "tokens_seen": 69468160, "tokens_per_second": 266314}
|
| 75 |
+
{"step": 540, "max_steps": 763, "train_loss": 6.292893, "learning_rate": 0.000170016340398158, "grad_norm": 0.26827898621559143, "tokens_seen": 70778880, "tokens_per_second": 266317}
|
| 76 |
+
{"step": 550, "max_steps": 763, "train_loss": 6.297822, "learning_rate": 0.00016102522996414806, "grad_norm": 0.34556809067726135, "tokens_seen": 72089600, "tokens_per_second": 266359}
|
| 77 |
+
{"step": 560, "max_steps": 763, "train_loss": 6.255699, "learning_rate": 0.00015233214581692112, "grad_norm": 0.33466053009033203, "tokens_seen": 73400320, "tokens_per_second": 266370}
|
| 78 |
+
{"step": 570, "max_steps": 763, "train_loss": 6.291651, "learning_rate": 0.0001439524202311625, "grad_norm": 0.3801525831222534, "tokens_seen": 74711040, "tokens_per_second": 266392}
|
| 79 |
+
{"step": 580, "max_steps": 763, "train_loss": 6.291024, "learning_rate": 0.0001359008328008913, "grad_norm": 0.3617173731327057, "tokens_seen": 76021760, "tokens_per_second": 266388}
|
| 80 |
+
{"step": 590, "max_steps": 763, "train_loss": 6.263499, "learning_rate": 0.0001281915843722103, "grad_norm": 0.36266186833381653, "tokens_seen": 77332480, "tokens_per_second": 266343}
|
| 81 |
+
{"step": 600, "max_steps": 763, "train_loss": 6.217177, "learning_rate": 0.00012083827199681319, "grad_norm": 0.27758166193962097, "tokens_seen": 78643200, "tokens_per_second": 266383}
|
| 82 |
+
{"step": 600, "validation_loss": 6.26907, "validation_perplexity": 527.9859}
|
| 83 |
+
{"step": 610, "max_steps": 763, "train_loss": 6.252789, "learning_rate": 0.00011385386495042271, "grad_norm": 0.4209630489349365, "tokens_seen": 79953920, "tokens_per_second": 230312}
|
| 84 |
+
{"step": 620, "max_steps": 763, "train_loss": 6.233759, "learning_rate": 0.0001072506818584584, "grad_norm": 0.3440088927745819, "tokens_seen": 81264640, "tokens_per_second": 266338}
|
| 85 |
+
{"step": 630, "max_steps": 763, "train_loss": 6.272423, "learning_rate": 0.00010104036896927688, "grad_norm": 0.5068637132644653, "tokens_seen": 82575360, "tokens_per_second": 266290}
|
| 86 |
+
{"step": 640, "max_steps": 763, "train_loss": 6.21127, "learning_rate": 9.523387961330576e-05, "grad_norm": 0.348201721906662, "tokens_seen": 83886080, "tokens_per_second": 266351}
|
| 87 |
+
{"step": 650, "max_steps": 763, "train_loss": 6.212734, "learning_rate": 8.984145488429902e-05, "grad_norm": 0.4265684187412262, "tokens_seen": 85196800, "tokens_per_second": 266316}
|
| 88 |
+
{"step": 660, "max_steps": 763, "train_loss": 6.213993, "learning_rate": 8.487260557678771e-05, "grad_norm": 0.3392343521118164, "tokens_seen": 86507520, "tokens_per_second": 266343}
|
| 89 |
+
{"step": 670, "max_steps": 763, "train_loss": 6.195304, "learning_rate": 8.033609541158293e-05, "grad_norm": 0.48064514994621277, "tokens_seen": 87818240, "tokens_per_second": 266342}
|
| 90 |
+
{"step": 680, "max_steps": 763, "train_loss": 6.239417, "learning_rate": 7.623992557891675e-05, "grad_norm": 0.2809535264968872, "tokens_seen": 89128960, "tokens_per_second": 266314}
|
| 91 |
+
{"step": 690, "max_steps": 763, "train_loss": 6.195762, "learning_rate": 7.25913206264835e-05, "grad_norm": 0.3590209186077118, "tokens_seen": 90439680, "tokens_per_second": 266407}
|
| 92 |
+
{"step": 700, "max_steps": 763, "train_loss": 6.199124, "learning_rate": 6.939671571727054e-05, "grad_norm": 0.25710293650627136, "tokens_seen": 91750400, "tokens_per_second": 266555}
|
| 93 |
+
{"step": 700, "validation_loss": 6.227403, "validation_perplexity": 506.4386}
|
| 94 |
+
{"step": 710, "max_steps": 763, "train_loss": 6.179626, "learning_rate": 6.6661745279652e-05, "grad_norm": 0.36142489314079285, "tokens_seen": 93061120, "tokens_per_second": 230489}
|
| 95 |
+
{"step": 720, "max_steps": 763, "train_loss": 6.17544, "learning_rate": 6.439123306976511e-05, "grad_norm": 0.3557684123516083, "tokens_seen": 94371840, "tokens_per_second": 266523}
|
| 96 |
+
{"step": 730, "max_steps": 763, "train_loss": 6.198924, "learning_rate": 6.258918366369462e-05, "grad_norm": 0.35512596368789673, "tokens_seen": 95682560, "tokens_per_second": 266479}
|
| 97 |
+
{"step": 740, "max_steps": 763, "train_loss": 6.204949, "learning_rate": 6.125877539447194e-05, "grad_norm": 0.33753326535224915, "tokens_seen": 96993280, "tokens_per_second": 266584}
|
| 98 |
+
{"step": 750, "max_steps": 763, "train_loss": 6.190249, "learning_rate": 6.0402354746346246e-05, "grad_norm": 0.29205089807510376, "tokens_seen": 98304000, "tokens_per_second": 266535}
|
| 99 |
+
Saved checkpoint: runs/kawk50m-v2-100m-5090/checkpoint-00000750
|
| 100 |
+
{"step": 760, "max_steps": 763, "train_loss": 6.177885, "learning_rate": 6.002143221621371e-05, "grad_norm": 0.29259902238845825, "tokens_seen": 99614720, "tokens_per_second": 229715}
|
| 101 |
+
Final checkpoint: runs/kawk50m-v2-100m-5090/checkpoint-00000763
|
reports/v2-100m/v0-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/artifacts/tokenizer-20k-pilot",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 418575,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 6.959662810717903,
|
| 13 |
+
"token_perplexity": 1053.2783431901112,
|
| 14 |
+
"nats_per_character": 3.675831862478103,
|
| 15 |
+
"bits_per_utf8_byte": 2.225403087334583,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2-100m/v0-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/artifacts/tokenizer-20k-pilot",
|
| 4 |
+
"validation_loss": 6.224435925483704,
|
| 5 |
+
"validation_perplexity": 504.9381393743716,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "의에 ·을하는도의과한,와,,의의은이.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n",
|
| 11 |
+
"full_text": "대한민국의 수도는의에 ·을하는도의과한,와,,의의은이.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "가.\n이한은를는으로,을는 을지 로, 어 \n을 \n\n ",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서가.\n이한은를는으로,을는 을지 로, 어 \n을 \n\n "
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "10354,268 7.9% 45만천,236억원 53,5억00원 .7,00로 , 월 4,57 원 ,199년 .5,6 .%,22 .1.%,1.6% ,22 .. . . ",
|
| 21 |
+
"full_text": "인공지능 언어 모델은10354,268 7.9% 45만천,236억원 53,5억00원 .7,00로 , 월 4,57 원 ,199년 .5,6 .%,22 .1.%,1.6% ,22 .. . . "
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "의에 을하여를려.\n은이의을지 , 이의로으로 어 고을하는의의의의와를 아을하게 다 ",
|
| 26 |
+
"full_text": "비가 내리는 날에는의에 을하여를려.\n은이의을지 , 이의로으로 어 고을하는의의의의와를 아을하게 다 "
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/v2-100m/v2-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 450128,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 6.798863921668693,
|
| 13 |
+
"token_perplexity": 896.82784599226,
|
| 14 |
+
"nats_per_character": 3.8615932873355683,
|
| 15 |
+
"bits_per_utf8_byte": 2.337865807026767,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2-100m/v2-evaluation.json
ADDED
|
@@ -0,0 +1,29 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2-100m-5090/checkpoint-00000763",
|
| 3 |
+
"validation_loss": 6.210828399658203,
|
| 4 |
+
"validation_perplexity": 498.11371765317085,
|
| 5 |
+
"evaluated_tokens": 655360,
|
| 6 |
+
"samples": [
|
| 7 |
+
{
|
| 8 |
+
"prompt": "대한민국의 수도는",
|
| 9 |
+
"continuation": ",가 00년 한국인’ ‘도화자이한이의'.\n기은21대사이지 30 이상으로되는이.‘’.\n2일2명이 04로.5.6\n.[이게어]이희 00-0-0 981\n제 :0\n동 :171011 72:0\n정:21\n성 ",
|
| 10 |
+
"full_text": "대한민국의 수도는,가 00년 한국인’ ‘도화자이한이의'.\n기은21대사이지 30 이상으로되는이.‘’.\n2일2명이 04로.5.6\n.[이게어]이희 00-0-0 981\n제 :0\n동 :171011 72:0\n정:21\n성 "
|
| 11 |
+
},
|
| 12 |
+
{
|
| 13 |
+
"prompt": "오늘 아침에 일어나서",
|
| 14 |
+
"continuation": "가이.\n은 04년,21대의를고도을하는.\n일는의에서한이 05년 만인.8.% .4이다\n1:3\n주 개월 0--61\n명 271\n 2.\n의 : 9\n.31\n4 0\n04\n :171\n.2 7\n.\n\n1.2\n",
|
| 15 |
+
"full_text": "오늘 아침에 일어나서가이.\n은 04년,21대의를고도을하는.\n일는의에서한이 05년 만인.8.% .4이다\n1:3\n주 개월 0--61\n명 271\n 2.\n의 : 9\n.31\n4 0\n04\n :171\n.2 7\n.\n\n1.2\n"
|
| 16 |
+
},
|
| 17 |
+
{
|
| 18 |
+
"prompt": "인공지능 언어 모델은",
|
| 19 |
+
"continuation": "가.\n트스,로터이는 01년의에된과2개을하는의와4세인를고.\n.11. ..3....6 2:-85-71",
|
| 20 |
+
"full_text": "인공지능 언어 모델은가.\n트스,로터이는 01년의에된과2개을하는의와4세인를고.\n.11. ..3....6 2:-85-71"
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"prompt": "비가 내리는 날에는",
|
| 24 |
+
"continuation": "를해의을 르히게.\n기이에 은 도로와 , ,.\n . 1 0 3:5\n 1 2:1 2.4\n 8 1 7:63 9 :1:72:0 65\n.3:1 1:1:0 ()-2:1\n1:",
|
| 25 |
+
"full_text": "비가 내리는 날에는를해의을 르히게.\n기이에 은 도로와 , ,.\n . 1 0 3:5\n 1 2:1 2.4\n 8 1 7:63 9 :1:72:0 65\n.3:1 1:1:0 ()-2:1\n1:"
|
| 26 |
+
}
|
| 27 |
+
],
|
| 28 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 29 |
+
}
|
reports/v2-20b/train.log
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
reports/v2-20b/v0-5b-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-5b-5090/checkpoint-00038147",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/artifacts/tokenizer-20k-pilot",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 1000,
|
| 8 |
+
"characters": 1657768,
|
| 9 |
+
"utf8_bytes": 3936686,
|
| 10 |
+
"predicted_tokens": 867888,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 7.449038352274124,
|
| 13 |
+
"token_perplexity": 1718.210037874284,
|
| 14 |
+
"nats_per_character": 3.8997803054941858,
|
| 15 |
+
"bits_per_utf8_byte": 2.369232366958287,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2-20b/v2-20b-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2-20b-5090/checkpoint-00152588",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2-20b-5090/checkpoint-00152588/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 1000,
|
| 8 |
+
"characters": 1657768,
|
| 9 |
+
"utf8_bytes": 3936686,
|
| 10 |
+
"predicted_tokens": 867888,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 7.156336863747745,
|
| 13 |
+
"token_perplexity": 1282.2054265313839,
|
| 14 |
+
"nats_per_character": 3.746542874518209,
|
| 15 |
+
"bits_per_utf8_byte": 2.2761360761784366,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2-20b/v2-20b-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2-20b-5090/checkpoint-00152588",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2-20b-5090/checkpoint-00152588/tokenizer",
|
| 4 |
+
"validation_loss": 4.43182813167572,
|
| 5 |
+
"validation_perplexity": 84.08499492975935,
|
| 6 |
+
"evaluated_tokens": 1638400,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "이다\n 00\n0\n\n\n\n\n\n\n\n\n\n\n신 \n\n\n \n\n21\n이 \n\n가\n은 \n인\n\n\n의\ne\n\n.\n\n 9\n\n2\n \n21\n산\n\n이\n\n세\n \n\n\n\n2\n가\n, \n,\n\n이\n \n.\n\n\n제3\n\n\n\n \nin\n이\n \n \n\n신\n \n215\n1\n4 \n\n제\n\n \n시 \n\n의 \ninut.-\n \n 정보\n\n \n ",
|
| 11 |
+
"full_text": "대한민국의 수도는이다\n 00\n0\n\n\n\n\n\n\n\n\n\n\n신 \n\n\n \n\n21\n이 \n\n가\n은 \n인\n\n\n의\ne\n\n.\n\n 9\n\n2\n \n21\n산\n\n이\n\n세\n \n\n\n\n2\n가\n, \n,\n\n이\n \n.\n\n\n제3\n\n\n\n \nin\n이\n \n \n\n신\n \n215\n1\n4 \n\n제\n\n \n시 \n\n의 \ninut.-\n \n 정보\n\n \n "
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "에을 .\n는이은 은한 은 은 한을하는 \n .의 은 1 이 8 은 가고 신 을 다 . 도 21 의 가 지,21 04 시 21 2 2 2",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서에을 .\n는이은 은한 은 은 한을하는 \n .의 은 1 이 8 은 가고 신 을 다 . 도 21 의 가 지,21 04 시 21 2 2 2"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "의에 .\n0.1의는 의을\n.02 01 1 3 2210 \n01 .0 65 4 :0 .9 0 10 8 2 1 가 이 . 1 2 . ",
|
| 21 |
+
"full_text": "인공지능 언어 모델은의에 .\n0.1의는 의을\n.02 01 1 3 2210 \n01 .0 65 4 :0 .9 0 10 8 2 1 가 이 . 1 2 . "
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "0238854\n 002894 041 6 77 9 .이은 기자<EMAIL> ",
|
| 26 |
+
"full_text": "비가 내리는 날에는0238854\n 002894 041 6 77 9 .이은 기자<EMAIL> "
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/v2b-100m/train.log
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"tokens_per_optimizer_step": 131072,
|
| 7 |
+
"max_steps": 763,
|
| 8 |
+
"planned_tokens": 100007936,
|
| 9 |
+
"start_step": 0,
|
| 10 |
+
"initialized_from": null,
|
| 11 |
+
"precision": "bf16",
|
| 12 |
+
"compile": true
|
| 13 |
+
}
|
| 14 |
+
{"step": 1, "max_steps": 763, "train_loss": 10.0099, "learning_rate": 3.9999999999999996e-05, "grad_norm": 1.4620856046676636, "tokens_seen": 131072, "tokens_per_second": 28924}
|
| 15 |
+
{"step": 10, "max_steps": 763, "train_loss": 8.870635, "learning_rate": 0.00039999999999999996, "grad_norm": 1.3404476642608643, "tokens_seen": 1310720, "tokens_per_second": 266737}
|
| 16 |
+
{"step": 20, "max_steps": 763, "train_loss": 7.654172, "learning_rate": 0.0005999404674662951, "grad_norm": 0.41257327795028687, "tokens_seen": 2621440, "tokens_per_second": 266673}
|
| 17 |
+
{"step": 30, "max_steps": 763, "train_loss": 7.615173, "learning_rate": 0.0005994643647016376, "grad_norm": 0.30431997776031494, "tokens_seen": 3932160, "tokens_per_second": 266745}
|
| 18 |
+
{"step": 40, "max_steps": 763, "train_loss": 7.569586, "learning_rate": 0.0005985129988902226, "grad_norm": 0.16756989061832428, "tokens_seen": 5242880, "tokens_per_second": 266728}
|
| 19 |
+
{"step": 50, "max_steps": 763, "train_loss": 7.553371, "learning_rate": 0.000597088047986813, "grad_norm": 0.18418455123901367, "tokens_seen": 6553600, "tokens_per_second": 266698}
|
| 20 |
+
{"step": 60, "max_steps": 763, "train_loss": 7.56118, "learning_rate": 0.0005951920252235705, "grad_norm": 0.14574965834617615, "tokens_seen": 7864320, "tokens_per_second": 266654}
|
| 21 |
+
{"step": 70, "max_steps": 763, "train_loss": 7.528229, "learning_rate": 0.0005928282746773871, "grad_norm": 0.14191102981567383, "tokens_seen": 9175040, "tokens_per_second": 266630}
|
| 22 |
+
{"step": 80, "max_steps": 763, "train_loss": 7.535922, "learning_rate": 0.0005900009653718274, "grad_norm": 0.17293578386306763, "tokens_seen": 10485760, "tokens_per_second": 266617}
|
| 23 |
+
{"step": 90, "max_steps": 763, "train_loss": 7.510619, "learning_rate": 0.0005867150839240876, "grad_norm": 0.19513754546642303, "tokens_seen": 11796480, "tokens_per_second": 266527}
|
| 24 |
+
{"step": 100, "max_steps": 763, "train_loss": 7.485366, "learning_rate": 0.0005829764257499357, "grad_norm": 0.21756480634212494, "tokens_seen": 13107200, "tokens_per_second": 266464}
|
| 25 |
+
{"step": 100, "validation_loss": 7.471647, "validation_perplexity": 1757.4984}
|
| 26 |
+
{"step": 110, "max_steps": 763, "train_loss": 7.383971, "learning_rate": 0.0005787915848421489, "grad_norm": 0.17387691140174866, "tokens_seen": 14417920, "tokens_per_second": 166859}
|
| 27 |
+
{"step": 120, "max_steps": 763, "train_loss": 7.352327, "learning_rate": 0.0005741679421404754, "grad_norm": 0.5503227114677429, "tokens_seen": 15728640, "tokens_per_second": 266394}
|
| 28 |
+
{"step": 130, "max_steps": 763, "train_loss": 7.282437, "learning_rate": 0.0005691136525136301, "grad_norm": 0.27086201310157776, "tokens_seen": 17039360, "tokens_per_second": 266383}
|
| 29 |
+
{"step": 140, "max_steps": 763, "train_loss": 7.240561, "learning_rate": 0.0005636376303762884, "grad_norm": 0.26870107650756836, "tokens_seen": 18350080, "tokens_per_second": 266347}
|
| 30 |
+
{"step": 150, "max_steps": 763, "train_loss": 7.193943, "learning_rate": 0.0005577495339664442, "grad_norm": 0.2710190713405609, "tokens_seen": 19660800, "tokens_per_second": 266344}
|
| 31 |
+
{"step": 160, "max_steps": 763, "train_loss": 7.162835, "learning_rate": 0.0005514597483108611, "grad_norm": 0.36225536465644836, "tokens_seen": 20971520, "tokens_per_second": 266381}
|
| 32 |
+
{"step": 170, "max_steps": 763, "train_loss": 7.139613, "learning_rate": 0.0005447793669086647, "grad_norm": 0.3161013722419739, "tokens_seen": 22282240, "tokens_per_second": 266320}
|
| 33 |
+
{"step": 180, "max_steps": 763, "train_loss": 7.099, "learning_rate": 0.0005377201721653773, "grad_norm": 0.22452951967716217, "tokens_seen": 23592960, "tokens_per_second": 266264}
|
| 34 |
+
{"step": 190, "max_steps": 763, "train_loss": 7.049736, "learning_rate": 0.0005302946146119068, "grad_norm": 0.25059765577316284, "tokens_seen": 24903680, "tokens_per_second": 266236}
|
| 35 |
+
{"step": 200, "max_steps": 763, "train_loss": 7.053318, "learning_rate": 0.0005225157909451416, "grad_norm": 0.43542009592056274, "tokens_seen": 26214400, "tokens_per_second": 266186}
|
| 36 |
+
{"step": 200, "validation_loss": 7.060782, "validation_perplexity": 1165.3566}
|
| 37 |
+
{"step": 210, "max_steps": 763, "train_loss": 7.037449, "learning_rate": 0.0005143974209288811, "grad_norm": 0.26370304822921753, "tokens_seen": 27525120, "tokens_per_second": 230198}
|
| 38 |
+
{"step": 220, "max_steps": 763, "train_loss": 6.975432, "learning_rate": 0.0005059538231958432, "grad_norm": 0.26475998759269714, "tokens_seen": 28835840, "tokens_per_second": 266135}
|
| 39 |
+
{"step": 230, "max_steps": 763, "train_loss": 6.965406, "learning_rate": 0.0004971998899934259, "grad_norm": 0.35259950160980225, "tokens_seen": 30146560, "tokens_per_second": 266134}
|
| 40 |
+
{"step": 240, "max_steps": 763, "train_loss": 6.928536, "learning_rate": 0.00048815106091776853, "grad_norm": 0.435407429933548, "tokens_seen": 31457280, "tokens_per_second": 266193}
|
| 41 |
+
{"step": 250, "max_steps": 763, "train_loss": 6.905537, "learning_rate": 0.0004788232956824341, "grad_norm": 0.34180641174316406, "tokens_seen": 32768000, "tokens_per_second": 266211}
|
| 42 |
+
Saved checkpoint: runs/kawk50m-v2b-100m-5090/checkpoint-00000250
|
| 43 |
+
{"step": 260, "max_steps": 763, "train_loss": 6.923052, "learning_rate": 0.00046923304596974483, "grad_norm": 0.3402026295661926, "tokens_seen": 34078720, "tokens_per_second": 232313}
|
| 44 |
+
{"step": 270, "max_steps": 763, "train_loss": 6.887879, "learning_rate": 0.0004593972264144153, "grad_norm": 0.23325936496257782, "tokens_seen": 35389440, "tokens_per_second": 266199}
|
| 45 |
+
{"step": 280, "max_steps": 763, "train_loss": 6.833163, "learning_rate": 0.00044933318477066334, "grad_norm": 0.3810471296310425, "tokens_seen": 36700160, "tokens_per_second": 266289}
|
| 46 |
+
{"step": 290, "max_steps": 763, "train_loss": 6.819903, "learning_rate": 0.00043905867131541155, "grad_norm": 0.3885501027107239, "tokens_seen": 38010880, "tokens_per_second": 266271}
|
| 47 |
+
{"step": 300, "max_steps": 763, "train_loss": 6.825687, "learning_rate": 0.0004285918075415489, "grad_norm": 0.1984497308731079, "tokens_seen": 39321600, "tokens_per_second": 266280}
|
| 48 |
+
{"step": 300, "validation_loss": 6.825716, "validation_perplexity": 921.2356}
|
| 49 |
+
{"step": 310, "max_steps": 763, "train_loss": 6.80148, "learning_rate": 0.00041795105419646444, "grad_norm": 0.29355815052986145, "tokens_seen": 40632320, "tokens_per_second": 230248}
|
| 50 |
+
{"step": 320, "max_steps": 763, "train_loss": 6.773774, "learning_rate": 0.00040715517872222867, "grad_norm": 0.3648630678653717, "tokens_seen": 41943040, "tokens_per_second": 266303}
|
| 51 |
+
{"step": 330, "max_steps": 763, "train_loss": 6.76706, "learning_rate": 0.0003962232221548463, "grad_norm": 0.4928100109100342, "tokens_seen": 43253760, "tokens_per_second": 266274}
|
| 52 |
+
{"step": 340, "max_steps": 763, "train_loss": 6.750465, "learning_rate": 0.00038517446554096337, "grad_norm": 0.23637978732585907, "tokens_seen": 44564480, "tokens_per_second": 266319}
|
| 53 |
+
{"step": 350, "max_steps": 763, "train_loss": 6.724762, "learning_rate": 0.0003740283959312592, "grad_norm": 0.37695491313934326, "tokens_seen": 45875200, "tokens_per_second": 266270}
|
| 54 |
+
{"step": 360, "max_steps": 763, "train_loss": 6.76801, "learning_rate": 0.0003628046720105023, "grad_norm": 0.3755868673324585, "tokens_seen": 47185920, "tokens_per_second": 266201}
|
| 55 |
+
{"step": 370, "max_steps": 763, "train_loss": 6.710524, "learning_rate": 0.00035152308942489074, "grad_norm": 0.3105439245700836, "tokens_seen": 48496640, "tokens_per_second": 266253}
|
| 56 |
+
{"step": 380, "max_steps": 763, "train_loss": 6.683565, "learning_rate": 0.0003402035458678273, "grad_norm": 0.3700482249259949, "tokens_seen": 49807360, "tokens_per_second": 266144}
|
| 57 |
+
{"step": 390, "max_steps": 763, "train_loss": 6.643779, "learning_rate": 0.00032886600598571277, "grad_norm": 0.34984415769577026, "tokens_seen": 51118080, "tokens_per_second": 266336}
|
| 58 |
+
{"step": 400, "max_steps": 763, "train_loss": 6.656132, "learning_rate": 0.0003175304661656503, "grad_norm": 0.33372822403907776, "tokens_seen": 52428800, "tokens_per_second": 266232}
|
| 59 |
+
{"step": 400, "validation_loss": 6.678488, "validation_perplexity": 795.1163}
|
| 60 |
+
{"step": 410, "max_steps": 763, "train_loss": 6.647935, "learning_rate": 0.00030621691926716884, "grad_norm": 0.48236599564552307, "tokens_seen": 53739520, "tokens_per_second": 230266}
|
| 61 |
+
{"step": 420, "max_steps": 763, "train_loss": 6.626147, "learning_rate": 0.0002949453193601676, "grad_norm": 0.25282207131385803, "tokens_seen": 55050240, "tokens_per_second": 266291}
|
| 62 |
+
{"step": 430, "max_steps": 763, "train_loss": 6.602006, "learning_rate": 0.0002837355465312754, "grad_norm": 0.28393957018852234, "tokens_seen": 56360960, "tokens_per_second": 266261}
|
| 63 |
+
{"step": 440, "max_steps": 763, "train_loss": 6.650382, "learning_rate": 0.00027260737182069626, "grad_norm": 0.3613400459289551, "tokens_seen": 57671680, "tokens_per_second": 266209}
|
| 64 |
+
{"step": 450, "max_steps": 763, "train_loss": 6.613721, "learning_rate": 0.00026158042235138474, "grad_norm": 0.29074686765670776, "tokens_seen": 58982400, "tokens_per_second": 266268}
|
| 65 |
+
{"step": 460, "max_steps": 763, "train_loss": 6.567071, "learning_rate": 0.00025067414671205103, "grad_norm": 0.4221433997154236, "tokens_seen": 60293120, "tokens_per_second": 266231}
|
| 66 |
+
{"step": 470, "max_steps": 763, "train_loss": 6.550491, "learning_rate": 0.00023990778065505504, "grad_norm": 0.27452975511550903, "tokens_seen": 61603840, "tokens_per_second": 266282}
|
| 67 |
+
{"step": 480, "max_steps": 763, "train_loss": 6.563501, "learning_rate": 0.00022930031316968416, "grad_norm": 0.36832818388938904, "tokens_seen": 62914560, "tokens_per_second": 266263}
|
| 68 |
+
{"step": 490, "max_steps": 763, "train_loss": 6.555317, "learning_rate": 0.00021887045299065797, "grad_norm": 0.42065495252609253, "tokens_seen": 64225280, "tokens_per_second": 266231}
|
| 69 |
+
{"step": 500, "max_steps": 763, "train_loss": 6.531689, "learning_rate": 0.00020863659560092397, "grad_norm": 0.29724204540252686, "tokens_seen": 65536000, "tokens_per_second": 266197}
|
| 70 |
+
{"step": 500, "validation_loss": 6.57314, "validation_perplexity": 715.6133}
|
| 71 |
+
Saved checkpoint: runs/kawk50m-v2b-100m-5090/checkpoint-00000500
|
| 72 |
+
{"step": 510, "max_steps": 763, "train_loss": 6.571091, "learning_rate": 0.00019861679078694882, "grad_norm": 0.430081844329834, "tokens_seen": 66846720, "tokens_per_second": 204667}
|
| 73 |
+
{"step": 520, "max_steps": 763, "train_loss": 6.548505, "learning_rate": 0.0001888287108037225, "grad_norm": 0.33085897564888, "tokens_seen": 68157440, "tokens_per_second": 266247}
|
| 74 |
+
{"step": 530, "max_steps": 763, "train_loss": 6.557787, "learning_rate": 0.00017928961920562968, "grad_norm": 0.302127480506897, "tokens_seen": 69468160, "tokens_per_second": 266278}
|
| 75 |
+
{"step": 540, "max_steps": 763, "train_loss": 6.525326, "learning_rate": 0.000170016340398158, "grad_norm": 0.27977681159973145, "tokens_seen": 70778880, "tokens_per_second": 266288}
|
| 76 |
+
{"step": 550, "max_steps": 763, "train_loss": 6.51195, "learning_rate": 0.00016102522996414806, "grad_norm": 0.34210750460624695, "tokens_seen": 72089600, "tokens_per_second": 266326}
|
| 77 |
+
{"step": 560, "max_steps": 763, "train_loss": 6.531392, "learning_rate": 0.00015233214581692112, "grad_norm": 0.4074239730834961, "tokens_seen": 73400320, "tokens_per_second": 266288}
|
| 78 |
+
{"step": 570, "max_steps": 763, "train_loss": 6.520072, "learning_rate": 0.0001439524202311625, "grad_norm": 0.43846815824508667, "tokens_seen": 74711040, "tokens_per_second": 266330}
|
| 79 |
+
{"step": 580, "max_steps": 763, "train_loss": 6.511605, "learning_rate": 0.0001359008328008913, "grad_norm": 0.31567832827568054, "tokens_seen": 76021760, "tokens_per_second": 266323}
|
| 80 |
+
{"step": 590, "max_steps": 763, "train_loss": 6.537479, "learning_rate": 0.0001281915843722103, "grad_norm": 0.3729858696460724, "tokens_seen": 77332480, "tokens_per_second": 266252}
|
| 81 |
+
{"step": 600, "max_steps": 763, "train_loss": 6.532045, "learning_rate": 0.00012083827199681319, "grad_norm": 0.26585572957992554, "tokens_seen": 78643200, "tokens_per_second": 266290}
|
| 82 |
+
{"step": 600, "validation_loss": 6.505008, "validation_perplexity": 668.4811}
|
| 83 |
+
{"step": 610, "max_steps": 763, "train_loss": 6.491537, "learning_rate": 0.00011385386495042271, "grad_norm": 0.3619851768016815, "tokens_seen": 79953920, "tokens_per_second": 230240}
|
| 84 |
+
{"step": 620, "max_steps": 763, "train_loss": 6.4614, "learning_rate": 0.0001072506818584584, "grad_norm": 0.48713865876197815, "tokens_seen": 81264640, "tokens_per_second": 266178}
|
| 85 |
+
{"step": 630, "max_steps": 763, "train_loss": 6.467959, "learning_rate": 0.00010104036896927688, "grad_norm": 0.3595183193683624, "tokens_seen": 82575360, "tokens_per_second": 266326}
|
| 86 |
+
{"step": 640, "max_steps": 763, "train_loss": 6.48798, "learning_rate": 9.523387961330576e-05, "grad_norm": 0.32979124784469604, "tokens_seen": 83886080, "tokens_per_second": 266216}
|
| 87 |
+
{"step": 650, "max_steps": 763, "train_loss": 6.470192, "learning_rate": 8.984145488429902e-05, "grad_norm": 0.3445857763290405, "tokens_seen": 85196800, "tokens_per_second": 266239}
|
| 88 |
+
{"step": 660, "max_steps": 763, "train_loss": 6.475928, "learning_rate": 8.487260557678771e-05, "grad_norm": 0.3449500799179077, "tokens_seen": 86507520, "tokens_per_second": 266251}
|
| 89 |
+
{"step": 670, "max_steps": 763, "train_loss": 6.458194, "learning_rate": 8.033609541158293e-05, "grad_norm": 0.3366071283817291, "tokens_seen": 87818240, "tokens_per_second": 266288}
|
| 90 |
+
{"step": 680, "max_steps": 763, "train_loss": 6.432897, "learning_rate": 7.623992557891675e-05, "grad_norm": 0.3163807988166809, "tokens_seen": 89128960, "tokens_per_second": 266289}
|
| 91 |
+
{"step": 690, "max_steps": 763, "train_loss": 6.403873, "learning_rate": 7.25913206264835e-05, "grad_norm": 0.28882038593292236, "tokens_seen": 90439680, "tokens_per_second": 266278}
|
| 92 |
+
{"step": 700, "max_steps": 763, "train_loss": 6.422303, "learning_rate": 6.939671571727054e-05, "grad_norm": 0.36098745465278625, "tokens_seen": 91750400, "tokens_per_second": 266284}
|
| 93 |
+
{"step": 700, "validation_loss": 6.465888, "validation_perplexity": 642.8347}
|
| 94 |
+
{"step": 710, "max_steps": 763, "train_loss": 6.417603, "learning_rate": 6.6661745279652e-05, "grad_norm": 0.4314678907394409, "tokens_seen": 93061120, "tokens_per_second": 230267}
|
| 95 |
+
{"step": 720, "max_steps": 763, "train_loss": 6.426275, "learning_rate": 6.439123306976511e-05, "grad_norm": 0.3415035307407379, "tokens_seen": 94371840, "tokens_per_second": 266256}
|
| 96 |
+
{"step": 730, "max_steps": 763, "train_loss": 6.431689, "learning_rate": 6.258918366369462e-05, "grad_norm": 0.31763747334480286, "tokens_seen": 95682560, "tokens_per_second": 266188}
|
| 97 |
+
{"step": 740, "max_steps": 763, "train_loss": 6.450324, "learning_rate": 6.125877539447194e-05, "grad_norm": 0.3103681802749634, "tokens_seen": 96993280, "tokens_per_second": 266165}
|
| 98 |
+
{"step": 750, "max_steps": 763, "train_loss": 6.424354, "learning_rate": 6.0402354746346246e-05, "grad_norm": 0.2832094430923462, "tokens_seen": 98304000, "tokens_per_second": 266214}
|
| 99 |
+
Saved checkpoint: runs/kawk50m-v2b-100m-5090/checkpoint-00000750
|
| 100 |
+
{"step": 760, "max_steps": 763, "train_loss": 6.413977, "learning_rate": 6.002143221621371e-05, "grad_norm": 0.4196031391620636, "tokens_seen": 99614720, "tokens_per_second": 234659}
|
| 101 |
+
Final checkpoint: runs/kawk50m-v2b-100m-5090/checkpoint-00000763
|
reports/v2b-100m/v0-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/artifacts/tokenizer-20k-pilot",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 418575,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 6.959662810717903,
|
| 13 |
+
"token_perplexity": 1053.2783431901112,
|
| 14 |
+
"nats_per_character": 3.675831862478103,
|
| 15 |
+
"bits_per_utf8_byte": 2.225403087334583,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2b-100m/v0-on-v2b-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/artifacts/tokenizer-20k-pilot",
|
| 4 |
+
"validation_loss": 6.413111543655395,
|
| 5 |
+
"validation_perplexity": 609.7881145657659,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "의에 ·을하는도의과한,와,,의의은이.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n",
|
| 11 |
+
"full_text": "대한민국의 수도는의에 ·을하는도의과한,와,,의의은이.\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n\n\n\n\n\n\n\n\n\n\n\n"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "가.\n이한은를는으로,을는 을지 로, 어 \n을 \n\n ",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서가.\n이한은를는으로,을는 을지 로, 어 \n을 \n\n "
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "10354,268 7.9% 45만천,236억원 53,5억00원 .7,00로 , 월 4,57 원 ,199년 .5,6 .%,22 .1.%,1.6% ,22 .. . . ",
|
| 21 |
+
"full_text": "인공지능 언어 모델은10354,268 7.9% 45만천,236억원 53,5억00원 .7,00로 , 월 4,57 원 ,199년 .5,6 .%,22 .1.%,1.6% ,22 .. . . "
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "의에 을하여를려.\n은이의을지 , 이의로으로 어 고을하는의의의의와를 아을하게 다 ",
|
| 26 |
+
"full_text": "비가 내리는 날에는의에 을하여를려.\n은이의을지 , 이의로으로 어 고을하는의의의의와를 아을하게 다 "
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/v2b-100m/v2b-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2b-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2b-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 418575,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 7.008261723106269,
|
| 13 |
+
"token_perplexity": 1105.730767147439,
|
| 14 |
+
"nats_per_character": 3.701499978232767,
|
| 15 |
+
"bits_per_utf8_byte": 2.2409429450276335,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2b-100m/v2b-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2b-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2b-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"validation_loss": 6.448035359382629,
|
| 5 |
+
"validation_perplexity": 631.4604804460281,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "로한.\n098년6 월2일 4,715 1,39,10원 00,00,2,9,18,1,0-0,1,2,0,2,0,2,0,1,1,.,2,1,1,2,,2,,,,,,,,,,,,,,,,,,,,,,",
|
| 11 |
+
"full_text": "대한민국의 수도는로한.\n098년6 월2일 4,715 1,39,10원 00,00,2,9,18,1,0-0,1,2,0,2,0,2,0,1,1,.,2,1,1,2,,2,,,,,,,,,,,,,,,,,,,,,,"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": ".\n은이한 를다 그,을는.“도 로를지.그 을고.\n\n 다 . ",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서.\n은이한 를다 그,을는.“도 로를지.그 을고.\n\n 다 . "
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "된의이에합니다\n한 02년1월5()는3 월일에고 9 동안30을하고108년에106년로되어.79년7에한 4% .1% .5.km가. .5,00원 .1.%,1./....2 ..-3 \n.. : 220\n.",
|
| 21 |
+
"full_text": "인공지능 언어 모델은된의이에합니다\n한 02년1월5()는3 월일에고 9 동안30을하고108년에106년로되어.79년7에한 4% .1% .5.km가. .5,00원 .1.%,1./....2 ..-3 \n.. : 220\n."
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "의에 을하고.이한의은이의() 함께다\n의 사()() 이은도) :/-061229457282243|=||][==뉴스]|||==210000.42.1\n :||||||| #||]#02100||||||||",
|
| 26 |
+
"full_text": "비가 내리는 날에는의에 을하고.이한의은이의() 함께다\n의 사()() 이은도) :/-061229457282243|=||][==뉴스]|||==210000.42.1\n :||||||| #||]#02100||||||||"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|
reports/v2c-100m/train.log
ADDED
|
@@ -0,0 +1,101 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"parameters": 51542528,
|
| 3 |
+
"device": "cuda:0",
|
| 4 |
+
"world_size": 1,
|
| 5 |
+
"sequence_length": 1024,
|
| 6 |
+
"tokens_per_optimizer_step": 131072,
|
| 7 |
+
"max_steps": 763,
|
| 8 |
+
"planned_tokens": 100007936,
|
| 9 |
+
"start_step": 0,
|
| 10 |
+
"initialized_from": null,
|
| 11 |
+
"precision": "bf16",
|
| 12 |
+
"compile": true
|
| 13 |
+
}
|
| 14 |
+
{"step": 1, "max_steps": 763, "train_loss": 10.008466, "learning_rate": 3.9999999999999996e-05, "grad_norm": 1.7012170553207397, "tokens_seen": 131072, "tokens_per_second": 50146}
|
| 15 |
+
{"step": 10, "max_steps": 763, "train_loss": 8.782962, "learning_rate": 0.00039999999999999996, "grad_norm": 1.4102976322174072, "tokens_seen": 1310720, "tokens_per_second": 266912}
|
| 16 |
+
{"step": 20, "max_steps": 763, "train_loss": 7.479458, "learning_rate": 0.0005999404674662951, "grad_norm": 0.41065457463264465, "tokens_seen": 2621440, "tokens_per_second": 266795}
|
| 17 |
+
{"step": 30, "max_steps": 763, "train_loss": 7.446536, "learning_rate": 0.0005994643647016376, "grad_norm": 0.35252320766448975, "tokens_seen": 3932160, "tokens_per_second": 266799}
|
| 18 |
+
{"step": 40, "max_steps": 763, "train_loss": 7.412397, "learning_rate": 0.0005985129988902226, "grad_norm": 0.20716822147369385, "tokens_seen": 5242880, "tokens_per_second": 266788}
|
| 19 |
+
{"step": 50, "max_steps": 763, "train_loss": 7.383052, "learning_rate": 0.000597088047986813, "grad_norm": 0.16321513056755066, "tokens_seen": 6553600, "tokens_per_second": 266815}
|
| 20 |
+
{"step": 60, "max_steps": 763, "train_loss": 7.340079, "learning_rate": 0.0005951920252235705, "grad_norm": 0.20032189786434174, "tokens_seen": 7864320, "tokens_per_second": 266955}
|
| 21 |
+
{"step": 70, "max_steps": 763, "train_loss": 7.337298, "learning_rate": 0.0005928282746773871, "grad_norm": 0.15987707674503326, "tokens_seen": 9175040, "tokens_per_second": 266885}
|
| 22 |
+
{"step": 80, "max_steps": 763, "train_loss": 7.347548, "learning_rate": 0.0005900009653718274, "grad_norm": 0.2403143346309662, "tokens_seen": 10485760, "tokens_per_second": 266791}
|
| 23 |
+
{"step": 90, "max_steps": 763, "train_loss": 7.310786, "learning_rate": 0.0005867150839240876, "grad_norm": 0.1972932666540146, "tokens_seen": 11796480, "tokens_per_second": 266753}
|
| 24 |
+
{"step": 100, "max_steps": 763, "train_loss": 7.26575, "learning_rate": 0.0005829764257499357, "grad_norm": 0.408220499753952, "tokens_seen": 13107200, "tokens_per_second": 266667}
|
| 25 |
+
{"step": 100, "validation_loss": 7.629168, "validation_perplexity": 2057.3366}
|
| 26 |
+
{"step": 110, "max_steps": 763, "train_loss": 7.201776, "learning_rate": 0.0005787915848421489, "grad_norm": 0.23724794387817383, "tokens_seen": 14417920, "tokens_per_second": 155341}
|
| 27 |
+
{"step": 120, "max_steps": 763, "train_loss": 7.113747, "learning_rate": 0.0005741679421404754, "grad_norm": 0.20529614388942719, "tokens_seen": 15728640, "tokens_per_second": 266449}
|
| 28 |
+
{"step": 130, "max_steps": 763, "train_loss": 7.074025, "learning_rate": 0.0005691136525136301, "grad_norm": 0.3647449016571045, "tokens_seen": 17039360, "tokens_per_second": 266355}
|
| 29 |
+
{"step": 140, "max_steps": 763, "train_loss": 7.041402, "learning_rate": 0.0005636376303762884, "grad_norm": 0.26863420009613037, "tokens_seen": 18350080, "tokens_per_second": 266398}
|
| 30 |
+
{"step": 150, "max_steps": 763, "train_loss": 6.961658, "learning_rate": 0.0005577495339664442, "grad_norm": 0.25631415843963623, "tokens_seen": 19660800, "tokens_per_second": 266381}
|
| 31 |
+
{"step": 160, "max_steps": 763, "train_loss": 6.955491, "learning_rate": 0.0005514597483108611, "grad_norm": 0.28667014837265015, "tokens_seen": 20971520, "tokens_per_second": 266352}
|
| 32 |
+
{"step": 170, "max_steps": 763, "train_loss": 6.873593, "learning_rate": 0.0005447793669086647, "grad_norm": 0.4511629641056061, "tokens_seen": 22282240, "tokens_per_second": 266343}
|
| 33 |
+
{"step": 180, "max_steps": 763, "train_loss": 6.894531, "learning_rate": 0.0005377201721653773, "grad_norm": 0.2158195674419403, "tokens_seen": 23592960, "tokens_per_second": 266283}
|
| 34 |
+
{"step": 190, "max_steps": 763, "train_loss": 6.885181, "learning_rate": 0.0005302946146119068, "grad_norm": 0.37028592824935913, "tokens_seen": 24903680, "tokens_per_second": 266264}
|
| 35 |
+
{"step": 200, "max_steps": 763, "train_loss": 6.793332, "learning_rate": 0.0005225157909451416, "grad_norm": 0.23921963572502136, "tokens_seen": 26214400, "tokens_per_second": 266286}
|
| 36 |
+
{"step": 200, "validation_loss": 7.346951, "validation_perplexity": 1551.4593}
|
| 37 |
+
{"step": 210, "max_steps": 763, "train_loss": 6.75241, "learning_rate": 0.0005143974209288811, "grad_norm": 0.37784042954444885, "tokens_seen": 27525120, "tokens_per_second": 230148}
|
| 38 |
+
{"step": 220, "max_steps": 763, "train_loss": 6.723357, "learning_rate": 0.0005059538231958432, "grad_norm": 0.4405493140220642, "tokens_seen": 28835840, "tokens_per_second": 266263}
|
| 39 |
+
{"step": 230, "max_steps": 763, "train_loss": 6.711626, "learning_rate": 0.0004971998899934259, "grad_norm": 0.45711755752563477, "tokens_seen": 30146560, "tokens_per_second": 266293}
|
| 40 |
+
{"step": 240, "max_steps": 763, "train_loss": 6.662347, "learning_rate": 0.00048815106091776853, "grad_norm": 0.2726283073425293, "tokens_seen": 31457280, "tokens_per_second": 266226}
|
| 41 |
+
{"step": 250, "max_steps": 763, "train_loss": 6.668326, "learning_rate": 0.0004788232956824341, "grad_norm": 0.37923818826675415, "tokens_seen": 32768000, "tokens_per_second": 266285}
|
| 42 |
+
Saved checkpoint: runs/kawk50m-v2c-100m-5090/checkpoint-00000250
|
| 43 |
+
{"step": 260, "max_steps": 763, "train_loss": 6.625059, "learning_rate": 0.00046923304596974483, "grad_norm": 0.2514687180519104, "tokens_seen": 34078720, "tokens_per_second": 230647}
|
| 44 |
+
{"step": 270, "max_steps": 763, "train_loss": 6.646087, "learning_rate": 0.0004593972264144153, "grad_norm": 0.46795105934143066, "tokens_seen": 35389440, "tokens_per_second": 266247}
|
| 45 |
+
{"step": 280, "max_steps": 763, "train_loss": 6.564949, "learning_rate": 0.00044933318477066334, "grad_norm": 0.39097777009010315, "tokens_seen": 36700160, "tokens_per_second": 266211}
|
| 46 |
+
{"step": 290, "max_steps": 763, "train_loss": 6.553521, "learning_rate": 0.00043905867131541155, "grad_norm": 0.3731852173805237, "tokens_seen": 38010880, "tokens_per_second": 266237}
|
| 47 |
+
{"step": 300, "max_steps": 763, "train_loss": 6.547451, "learning_rate": 0.0004285918075415489, "grad_norm": 0.26285019516944885, "tokens_seen": 39321600, "tokens_per_second": 266177}
|
| 48 |
+
{"step": 300, "validation_loss": 7.185474, "validation_perplexity": 1320.1145}
|
| 49 |
+
{"step": 310, "max_steps": 763, "train_loss": 6.478361, "learning_rate": 0.00041795105419646444, "grad_norm": 0.4246487617492676, "tokens_seen": 40632320, "tokens_per_second": 230238}
|
| 50 |
+
{"step": 320, "max_steps": 763, "train_loss": 6.487781, "learning_rate": 0.00040715517872222867, "grad_norm": 0.29563823342323303, "tokens_seen": 41943040, "tokens_per_second": 266138}
|
| 51 |
+
{"step": 330, "max_steps": 763, "train_loss": 6.449433, "learning_rate": 0.0003962232221548463, "grad_norm": 0.28168973326683044, "tokens_seen": 43253760, "tokens_per_second": 266282}
|
| 52 |
+
{"step": 340, "max_steps": 763, "train_loss": 6.439594, "learning_rate": 0.00038517446554096337, "grad_norm": 0.49542802572250366, "tokens_seen": 44564480, "tokens_per_second": 266194}
|
| 53 |
+
{"step": 350, "max_steps": 763, "train_loss": 6.433197, "learning_rate": 0.0003740283959312592, "grad_norm": 0.3152438700199127, "tokens_seen": 45875200, "tokens_per_second": 266213}
|
| 54 |
+
{"step": 360, "max_steps": 763, "train_loss": 6.356381, "learning_rate": 0.0003628046720105023, "grad_norm": 0.32476043701171875, "tokens_seen": 47185920, "tokens_per_second": 266247}
|
| 55 |
+
{"step": 370, "max_steps": 763, "train_loss": 6.355656, "learning_rate": 0.00035152308942489074, "grad_norm": 0.3003177046775818, "tokens_seen": 48496640, "tokens_per_second": 266181}
|
| 56 |
+
{"step": 380, "max_steps": 763, "train_loss": 6.431495, "learning_rate": 0.0003402035458678273, "grad_norm": 0.43732327222824097, "tokens_seen": 49807360, "tokens_per_second": 266190}
|
| 57 |
+
{"step": 390, "max_steps": 763, "train_loss": 6.357932, "learning_rate": 0.00032886600598571277, "grad_norm": 0.32416102290153503, "tokens_seen": 51118080, "tokens_per_second": 266204}
|
| 58 |
+
{"step": 400, "max_steps": 763, "train_loss": 6.369991, "learning_rate": 0.0003175304661656503, "grad_norm": 0.2728043794631958, "tokens_seen": 52428800, "tokens_per_second": 266165}
|
| 59 |
+
{"step": 400, "validation_loss": 7.074391, "validation_perplexity": 1181.3241}
|
| 60 |
+
{"step": 410, "max_steps": 763, "train_loss": 6.346023, "learning_rate": 0.00030621691926716884, "grad_norm": 0.26200070977211, "tokens_seen": 53739520, "tokens_per_second": 230188}
|
| 61 |
+
{"step": 420, "max_steps": 763, "train_loss": 6.307652, "learning_rate": 0.0002949453193601676, "grad_norm": 0.36999261379241943, "tokens_seen": 55050240, "tokens_per_second": 266218}
|
| 62 |
+
{"step": 430, "max_steps": 763, "train_loss": 6.325831, "learning_rate": 0.0002837355465312754, "grad_norm": 0.4334343373775482, "tokens_seen": 56360960, "tokens_per_second": 266217}
|
| 63 |
+
{"step": 440, "max_steps": 763, "train_loss": 6.292094, "learning_rate": 0.00027260737182069626, "grad_norm": 0.34754785895347595, "tokens_seen": 57671680, "tokens_per_second": 266012}
|
| 64 |
+
{"step": 450, "max_steps": 763, "train_loss": 6.252117, "learning_rate": 0.00026158042235138474, "grad_norm": 0.3662015199661255, "tokens_seen": 58982400, "tokens_per_second": 266195}
|
| 65 |
+
{"step": 460, "max_steps": 763, "train_loss": 6.295666, "learning_rate": 0.00025067414671205103, "grad_norm": 0.37175410985946655, "tokens_seen": 60293120, "tokens_per_second": 266116}
|
| 66 |
+
{"step": 470, "max_steps": 763, "train_loss": 6.169696, "learning_rate": 0.00023990778065505504, "grad_norm": 0.2848922610282898, "tokens_seen": 61603840, "tokens_per_second": 266130}
|
| 67 |
+
{"step": 480, "max_steps": 763, "train_loss": 6.211642, "learning_rate": 0.00022930031316968416, "grad_norm": 0.37786588072776794, "tokens_seen": 62914560, "tokens_per_second": 265769}
|
| 68 |
+
{"step": 490, "max_steps": 763, "train_loss": 6.247116, "learning_rate": 0.00021887045299065797, "grad_norm": 0.36943691968917847, "tokens_seen": 64225280, "tokens_per_second": 266162}
|
| 69 |
+
{"step": 500, "max_steps": 763, "train_loss": 6.189884, "learning_rate": 0.00020863659560092397, "grad_norm": 0.4221106171607971, "tokens_seen": 65536000, "tokens_per_second": 266236}
|
| 70 |
+
{"step": 500, "validation_loss": 6.992527, "validation_perplexity": 1088.4688}
|
| 71 |
+
Saved checkpoint: runs/kawk50m-v2c-100m-5090/checkpoint-00000500
|
| 72 |
+
{"step": 510, "max_steps": 763, "train_loss": 6.20336, "learning_rate": 0.00019861679078694882, "grad_norm": 0.4157133400440216, "tokens_seen": 66846720, "tokens_per_second": 208921}
|
| 73 |
+
{"step": 520, "max_steps": 763, "train_loss": 6.219778, "learning_rate": 0.0001888287108037225, "grad_norm": 0.4442093074321747, "tokens_seen": 68157440, "tokens_per_second": 266376}
|
| 74 |
+
{"step": 530, "max_steps": 763, "train_loss": 6.247092, "learning_rate": 0.00017928961920562968, "grad_norm": 0.38505372405052185, "tokens_seen": 69468160, "tokens_per_second": 266255}
|
| 75 |
+
{"step": 540, "max_steps": 763, "train_loss": 6.206688, "learning_rate": 0.000170016340398158, "grad_norm": 0.2860244810581207, "tokens_seen": 70778880, "tokens_per_second": 266266}
|
| 76 |
+
{"step": 550, "max_steps": 763, "train_loss": 6.124304, "learning_rate": 0.00016102522996414806, "grad_norm": 0.31607306003570557, "tokens_seen": 72089600, "tokens_per_second": 266348}
|
| 77 |
+
{"step": 560, "max_steps": 763, "train_loss": 6.179218, "learning_rate": 0.00015233214581692112, "grad_norm": 0.28299573063850403, "tokens_seen": 73400320, "tokens_per_second": 266248}
|
| 78 |
+
{"step": 570, "max_steps": 763, "train_loss": 6.180524, "learning_rate": 0.0001439524202311625, "grad_norm": 0.3858914375305176, "tokens_seen": 74711040, "tokens_per_second": 266268}
|
| 79 |
+
{"step": 580, "max_steps": 763, "train_loss": 6.145733, "learning_rate": 0.0001359008328008913, "grad_norm": 0.3984925150871277, "tokens_seen": 76021760, "tokens_per_second": 266295}
|
| 80 |
+
{"step": 590, "max_steps": 763, "train_loss": 6.079543, "learning_rate": 0.0001281915843722103, "grad_norm": 0.36312010884284973, "tokens_seen": 77332480, "tokens_per_second": 266254}
|
| 81 |
+
{"step": 600, "max_steps": 763, "train_loss": 6.128458, "learning_rate": 0.00012083827199681319, "grad_norm": 0.35983508825302124, "tokens_seen": 78643200, "tokens_per_second": 266206}
|
| 82 |
+
{"step": 600, "validation_loss": 6.930543, "validation_perplexity": 1023.0492}
|
| 83 |
+
{"step": 610, "max_steps": 763, "train_loss": 6.098582, "learning_rate": 0.00011385386495042271, "grad_norm": 0.46834057569503784, "tokens_seen": 79953920, "tokens_per_second": 230310}
|
| 84 |
+
{"step": 620, "max_steps": 763, "train_loss": 6.070086, "learning_rate": 0.0001072506818584584, "grad_norm": 0.365415096282959, "tokens_seen": 81264640, "tokens_per_second": 266189}
|
| 85 |
+
{"step": 630, "max_steps": 763, "train_loss": 6.068446, "learning_rate": 0.00010104036896927688, "grad_norm": 0.357563316822052, "tokens_seen": 82575360, "tokens_per_second": 266300}
|
| 86 |
+
{"step": 640, "max_steps": 763, "train_loss": 6.156944, "learning_rate": 9.523387961330576e-05, "grad_norm": 0.39240700006484985, "tokens_seen": 83886080, "tokens_per_second": 266265}
|
| 87 |
+
{"step": 650, "max_steps": 763, "train_loss": 6.172819, "learning_rate": 8.984145488429902e-05, "grad_norm": 0.3248313069343567, "tokens_seen": 85196800, "tokens_per_second": 266358}
|
| 88 |
+
{"step": 660, "max_steps": 763, "train_loss": 6.079392, "learning_rate": 8.487260557678771e-05, "grad_norm": 0.4324984550476074, "tokens_seen": 86507520, "tokens_per_second": 266305}
|
| 89 |
+
{"step": 670, "max_steps": 763, "train_loss": 6.055832, "learning_rate": 8.033609541158293e-05, "grad_norm": 0.35276126861572266, "tokens_seen": 87818240, "tokens_per_second": 266321}
|
| 90 |
+
{"step": 680, "max_steps": 763, "train_loss": 6.067945, "learning_rate": 7.623992557891675e-05, "grad_norm": 0.38677510619163513, "tokens_seen": 89128960, "tokens_per_second": 266280}
|
| 91 |
+
{"step": 690, "max_steps": 763, "train_loss": 6.053312, "learning_rate": 7.25913206264835e-05, "grad_norm": 0.3388923406600952, "tokens_seen": 90439680, "tokens_per_second": 266248}
|
| 92 |
+
{"step": 700, "max_steps": 763, "train_loss": 6.094653, "learning_rate": 6.939671571727054e-05, "grad_norm": 0.32606759667396545, "tokens_seen": 91750400, "tokens_per_second": 266236}
|
| 93 |
+
{"step": 700, "validation_loss": 6.900528, "validation_perplexity": 992.7987}
|
| 94 |
+
{"step": 710, "max_steps": 763, "train_loss": 6.094742, "learning_rate": 6.6661745279652e-05, "grad_norm": 0.36600637435913086, "tokens_seen": 93061120, "tokens_per_second": 230352}
|
| 95 |
+
{"step": 720, "max_steps": 763, "train_loss": 6.077293, "learning_rate": 6.439123306976511e-05, "grad_norm": 0.35945576429367065, "tokens_seen": 94371840, "tokens_per_second": 266274}
|
| 96 |
+
{"step": 730, "max_steps": 763, "train_loss": 6.124818, "learning_rate": 6.258918366369462e-05, "grad_norm": 0.3347718417644501, "tokens_seen": 95682560, "tokens_per_second": 266338}
|
| 97 |
+
{"step": 740, "max_steps": 763, "train_loss": 6.070718, "learning_rate": 6.125877539447194e-05, "grad_norm": 0.44693854451179504, "tokens_seen": 96993280, "tokens_per_second": 266348}
|
| 98 |
+
{"step": 750, "max_steps": 763, "train_loss": 6.03677, "learning_rate": 6.0402354746346246e-05, "grad_norm": 0.3649730980396271, "tokens_seen": 98304000, "tokens_per_second": 266274}
|
| 99 |
+
Saved checkpoint: runs/kawk50m-v2c-100m-5090/checkpoint-00000750
|
| 100 |
+
{"step": 760, "max_steps": 763, "train_loss": 6.019629, "learning_rate": 6.002143221621371e-05, "grad_norm": 0.41180509328842163, "tokens_seen": 99614720, "tokens_per_second": 237759}
|
| 101 |
+
Final checkpoint: runs/kawk50m-v2c-100m-5090/checkpoint-00000763
|
reports/v2c-100m/v2c-byte-evaluation.json
ADDED
|
@@ -0,0 +1,17 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2c-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2c-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"inputs": [
|
| 5 |
+
"/workspace/KAWK50M/data/clean-v2-full/validation-00000.jsonl"
|
| 6 |
+
],
|
| 7 |
+
"documents": 500,
|
| 8 |
+
"characters": 792512,
|
| 9 |
+
"utf8_bytes": 1888545,
|
| 10 |
+
"predicted_tokens": 418575,
|
| 11 |
+
"context_length": 1024,
|
| 12 |
+
"token_nll": 7.422225550706402,
|
| 13 |
+
"token_perplexity": 1672.752162939873,
|
| 14 |
+
"nats_per_character": 3.920140086064227,
|
| 15 |
+
"bits_per_utf8_byte": 2.373310906671906,
|
| 16 |
+
"note": "Compare bits_per_utf8_byte only when checkpoints are evaluated on the exact same ordered raw-text documents."
|
| 17 |
+
}
|
reports/v2c-100m/v2c-evaluation.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"checkpoint": "/workspace/KAWK50M/runs/kawk50m-v2c-100m-5090/checkpoint-00000763",
|
| 3 |
+
"tokenizer": "/workspace/KAWK50M/runs/kawk50m-v2c-100m-5090/checkpoint-00000763/tokenizer",
|
| 4 |
+
"validation_loss": 6.885479140281677,
|
| 5 |
+
"validation_perplexity": 977.9701423572044,
|
| 6 |
+
"evaluated_tokens": 655360,
|
| 7 |
+
"samples": [
|
| 8 |
+
{
|
| 9 |
+
"prompt": "대한민국의 수도는",
|
| 10 |
+
"continuation": "의한과,의권의에지가될 아니라적인와의이을찾.\n신적으로 있는은\n 법인 사과성() \n동제 :1843-02-571690\n원 :2421091년1월0,도 93\n부일 :12559,700",
|
| 11 |
+
"full_text": "대한민국의 수도는의한과,의권의에지가될 아니라적인와의이을찾.\n신적으로 있는은\n 법인 사과성() \n동제 :1843-02-571690\n원 :2421091년1월0,도 93\n부일 :12559,700"
|
| 12 |
+
},
|
| 13 |
+
{
|
| 14 |
+
"prompt": "오늘 아침에 일어나서",
|
| 15 |
+
"continuation": "0%.\n이게 8 미만 6%인가1일3,3분2,월2시간 일2분5,1명4만원\n도산\n\n\n\n\n\n\n기내용제\n.답\n.-문\n..\n\n\n.작권\n\n\n\n성\n\n\n\n\n\n\n\n\n\n\n소지\n\n\n신부\n\n\n\n\n\n구\n\n무\n수\n\n세\n사\n\n보\n\n\n정\n\n\n자\n\n아\n상\n주/목\n\n고연\n제\n유\n기\n\n어시\n포\n원일\n\n장의\n인",
|
| 16 |
+
"full_text": "오늘 아침에 일어나서0%.\n이게 8 미만 6%인가1일3,3분2,월2시간 일2분5,1명4만원\n도산\n\n\n\n\n\n\n기내용제\n.답\n.-문\n..\n\n\n.작권\n\n\n\n성\n\n\n\n\n\n\n\n\n\n\n소지\n\n\n신부\n\n\n\n\n\n구\n\n무\n수\n\n세\n사\n\n보\n\n\n정\n\n\n자\n\n아\n상\n주/목\n\n고연\n제\n유\n기\n\n어시\n포\n원일\n\n장의\n인"
|
| 17 |
+
},
|
| 18 |
+
{
|
| 19 |
+
"prompt": "인공지능 언어 모델은",
|
| 20 |
+
"continuation": "적,의에하는도의과을한, 구로의를는이의의을어 것이가.\n제지,,이 ,,,,,,-\n,,,,,,--\n ,--\n고이,,,,,,,,,,,,-,,,,,,,-,-,-----,-,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,",
|
| 21 |
+
"full_text": "인공지능 언어 모델은적,의에하는도의과을한, 구로의를는이의의을어 것이가.\n제지,,이 ,,,,,,-\n,,,,,,--\n ,--\n고이,,,,,,,,,,,,-,,,,,,,-,-,-----,-,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,,"
|
| 22 |
+
},
|
| 23 |
+
{
|
| 24 |
+
"prompt": "비가 내리는 날에는",
|
| 25 |
+
"continuation": "과에 지이게오,은을,고도하는야,이를해,고를어.\n신사소의로의이으로 수 이에10이트-2385940261607516860800250138080010021906041110030101920800101200002년52월3351002년92901.0||이106000110002202 03||3",
|
| 26 |
+
"full_text": "비가 내리는 날에는과에 지이게오,은을,고도하는야,이를해,고를어.\n신사소의로의이으로 수 이에10이트-2385940261607516860800250138080010021906041110030101920800101200002년52월3351002년92901.0||이106000110002202 03||3"
|
| 27 |
+
}
|
| 28 |
+
],
|
| 29 |
+
"note": "This is a base-model continuation evaluation, not instruction QA."
|
| 30 |
+
}
|