Create lc_32k

lc_32k

@@ -0,0 +1,326 @@
+checkpoints:
+  checkpoint_interval: 2000
+  checkpoints_path: /scratch/elie-checkpoints-stage2-lc
+  checkpoints_path_is_shared_file_system: false
+  load_lr_scheduler: true
+  load_optimizer: true
+  resume_checkpoint_path: s3://smollm3/tp-fix-final-pre-training/elie-lc-prolong-think-chatml-mix01
+  save_final_state: true
+  save_initial_state: false
+data_stages:
+- data:
+    dataset:
+      dataset_read_path: # replace some programming languages with stack-edu-real, add finemath4+ and infiwebmath4+ and MegaMath, donwsample some sources like SE
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fineweb-edu
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/dclm
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/pes2o
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/wiki
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/mwiki
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/cosmopedia2
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stackexchange
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-fra_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-spa_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-deu_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-ita_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-por_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-cmn_Hani
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-rus_Cyrl
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-arb_Arab
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-jpn_Jpan
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-kor_Hang
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hin_Deva
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-tha_Thai
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-vie_Latn
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/fw2-hq-ell_Grek
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/infiwebmath-3plus
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/finemath-3plus
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/infiwebmath-4plus
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/finemath-4plus
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/megamath-web-pro
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/megamath-qa-qwen
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/megamath-text-code-block
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Python
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Java
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-JavaScript
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-C
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Cpp
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-C-Sharp
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-PHP
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-TypeScript
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Swift
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-SQL
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Ruby
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Markdown
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-HTML
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Rust
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Go
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/stack-edu-real-Shell
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/pull-requests
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/kaggle
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/jupyter-scripts
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/github-issues
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/openmathinstruct-2
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/openmathreasoning-4k
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/opencodereasoning-4k
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/natural_reasoning
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/problem-solving
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/short-context/2students
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/long-context/reasoning_nochatml_17M_4k_to_32k-correct_cpu_tok_4k_to_32k
+      - /fsx/elie_bakouch/smollm3_training/1004-nn/data/tokenized_data/long-context/reasoning_tulu_chatml_580k_all_lengths_cpu_tok_4k_to_32k
+      dataset_folder:
+      - s3://smollm3/datasets/llama_tokenized_32769/fineweb-edu/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/dclm/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/pes2o/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/wiki/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/mwiki/standard
+      - s3://smollm3/datasets/llama_tokenized_32769/cosmopedia2/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/stackexchange/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-fra_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-spa_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-deu_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-ita_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-por_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-cmn_Hani/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-rus_Cyrl/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-arb_Arab/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-jpn_Jpan/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-kor_Hang/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hin_Deva/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-tha_Thai/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-vie_Latn/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/fw2-hq-ell_Grek/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/infiwebmath-3plus/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/finemath-3plus/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/infiwebmath-4plus/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/finemath-4plus/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/megamath-web-pro/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/megamath-qa-qwen/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/megamath-text-code-block/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Python/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Java/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-JavaScript/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-C/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Cpp/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-C-Sharp/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-PHP/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-TypeScript/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Swift/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-SQL/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Ruby/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Markdown/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-HTML/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Rust/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Go/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/stack-edu-real-Shell/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/pull-requests/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/kaggle/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/jupyter-scripts/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/github-issues/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/openmathinstruct-2/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/openmathreasoning-4k/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/opencodereasoning-4k/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/natural_reasoning/standard/
+      - s3://smollm3/datasets/llama_tokenized_32769/problem-solving/tokenized/
+      - s3://smollm3/datasets/llama_tokenized_32769/2students/tokenized/
+      - s3://smollm3/datasets/elie_correct_llama_tokenized_32769/reasoning_nochatml_17M_4k_to_32k-correct_cpu_tok_4k_to_32k/tokenized/
+      - s3://smollm3/datasets/elie_correct_llama_tokenized_32769/reasoning_tulu_chatml_580k_all_lengths_cpu_tok_4k_to_32k/tokenized/
+      dataset_weights:
+      - 0.1253669
+      - 0.2686434
+      - 0.0017910 #pes2o
+      - 0.0001791 # Wiki
+      - 0.0071638 # mwiki
+      - 0.0035819 # cosmopedia
+      - 0.0008955 # SE
+      - 0.0161186
+      - 0.0197005
+      - 0.0205960
+      - 0.0111935
+      - 0.0040297
+      - 0.0089548
+      - 0.0089548
+      - 0.0080593
+      - 0.0028655
+      - 0.0028655
+      - 0.0028655
+      - 0.0028655
+      - 0.0000448 # downsample viet, too many epochs
+      - 0.0019701
+      - 0.0017910 # update infiwebmath3+
+      - 0.0017910 # update finemath3+
+      - 0.0179096 # add infiwebmath4+
+      - 0.0223869 # add finemath4+
+      - 0.0125367 # add MegamathWeb-Pro
+      - 0.0017910 # MegaMathQA-Qwen
+      - 0.0626834 # add MegaMath-Text-Code
+      - 0.0805930 # add stack-edu-real-shuffled-Python
+      - 0.0161186 # Java
+      - 0.0161186 # JS
+      - 0.0071638 # C
+      - 0.0394010 # add stack-edu-real-shuffled-Cpp
+      - 0.0053729 # C# Edu
+      - 0.0053729 # PHP
+      - 0.0026864 # TS
+      - 0.0017910  # Swift
+      - 0.0116412 # SQL
+      - 0.0008955 # Ruby
+      - 0.0044774 # Md
+      - 0.0091339 # HTML
+      - 0.0008955 # Rust
+      - 0.0004477 # Go
+      - 0.0053729 # Shell
+      - 0.0044774 # PRs
+      - 0.0005373 # kaggle
+      - 0.0107457 # notebooks
+      - 0.0035819 # issues
+      - 0.0223869 # OpenMathInstruct
+      - 0.0044774 # OpenMathReasoning
+      - 0.0004477 # OpenCodeReasoning
+      - 0.0008955 # Natural reasoning
+      - 0.0026864 # Olmo Problem solving
+      - 0.0026864 # Olmo two students
+      - 0.09     # 0.1 * 0.2 * 0.75 DCLM 16k
+      - 0.01     # 0.1 * 0.2 * 0.75 DCLM 16k
+      pad_samples_to_global_batch_size: false
+      return_positions: true
+      token_size_in_bytes: 4
+      tokenizer_name: HuggingFaceTB/SmolLM3-11T-32k-v1-remote-code
+      use_old_brrr_dataloader: false
+      vocab_size: 128256
+    num_loading_workers: 0
+    seed: 6
+  name: lc stage
+  start_training_step: 1
+general:
+  benchmark_csv_path: null
+  consumed_train_samples: null
+  ignore_sanity_checks: true
+  project: smollm3-3B-final
+  run: elie-lc-prolong-think-chatml-mix01
+  seed: 6
+  step: null
+logging:
+  iteration_step_info_interval: 1
+  log_level: info
+  log_level_replica: info
+model:
+  ddp_bucket_cap_mb: 50
+  dtype: bfloat16
+  init_method:
+    std: 0.02
+  make_vocab_size_divisible_by: 1
+  model_config:
+    _attn_implementation: llama3_ring_attention
+    _fused_rms_norm: true
+    _fused_rotary_emb: true
+    _use_doc_masking: true
+    _use_qkv_packed: true
+    ring_attn_heads_k_stride: 1
+    attention_bias: false
+    bos_token_id: 128000
+    eos_token_id: 128001
+    flex_attention_mask: null
+    hidden_act: silu
+    hidden_size: 2048
+    initializer_range: 0.02
+    intermediate_size: 11008
+    is_qwen2_config: true
+    max_position_embeddings: 32768
+    moe_config: null
+    num_attention_heads: 16
+    num_hidden_layers: 36
+    num_key_value_heads: 4
+    pad_token_id: null
+    pretraining_tp: 2
+    rms_norm_eps: 1.0e-06
+    rope_interleaved: false
+    rope_scaling: null
+    rope_theta: 2000000.0
+    sliding_window_size: null
+    tie_word_embeddings: true
+    use_cache: true
+    vocab_size: 128256
+    z_loss_coefficient: 1.0e-05
+    z_loss_enabled: false
+    no_rope_layer: 4
+optimizer:
+  accumulate_grad_in_fp32: true
+  clip_grad: 1.0
+  learning_rate_scheduler:
+    learning_rate: 0.00002
+    lr_decay_starting_step: 1000
+    lr_decay_steps: 19000
+    lr_decay_style: cosine
+    lr_warmup_steps: 1000
+    lr_warmup_style: linear
+    min_decay_lr: 0
+  optimizer_factory:
+    adam_beta1: 0.9
+    adam_beta2: 0.95
+    adam_eps: 1.0e-08
+    name: adamW
+    torch_adam_is_fused: true
+  weight_decay: 0.1
+  weight_decay_exclude_named_params:
+  - .*token_embedding.*
+  zero_stage: 0
+parallelism:
+  context_parallel_size: 4
+  dp: 12
+  expert_parallel_size: 1
+  moe_layer_recompute: false
+  pp: 1
+  pp_engine: 1f1b
+  recompute_layer: false
+  tp: 2
+  tp_linear_async_communication: true
+  tp_mode: REDUCE_SCATTER
+  tp_recompute_allgather: true
+s3_upload:
+  remove_after_upload: true
+  s5cmd_concurrency: 5
+  s5cmd_numworkers: 16
+  s5cmd_path: /fsx/elie_bakouch/smollm3_training/1004-nn/1004-hope/bin/s5cmd
+  upload_s3_path: s3://smollm3/tp-fix-final-pre-training/elie-lc-prolong-think-chatml-mix01
+tokenizer:
+  tokenizer_max_length: 32768
+  tokenizer_name_or_path: HuggingFaceTB/SmolLM3-11T-32k-v1-remote-code
+  tokenizer_revision: null
+metrics_logging:
+  log_level: 1
+  log_detail_interval: 200
+tokens:
+  batch_accumulation_per_replica: 6
+  limit_test_batches: 0
+  limit_val_batches: 0
+  micro_batch_size: 1
+  sequence_length: 32768
+  train_steps: 20000
+  val_check_interval: 100
+lighteval:
+  slurm_script_dir: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron/_final1004/eval_results/launch-config"  # Default path for launch scripts
+  logs_path: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron/_final1004/eval_results/logs"  # Default path for evaluation logs
+  local_checkpoint_dir: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron/_final1004/evals-ckpt"  # Default path for temporary checkpoint storage. Will store under {local_checkpoint_dir}/{run_name}/{step}
+  nanotron_path: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron"
+  output_dir: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron/_final1004/eval_results/results"
+  s3_save_path: "s3://smollm3/eval-prod"
+  upload_to_wandb: true
+  wandb_project: smollm3-3B-evals
+  wandb_entity: huggingface
+  parallelism:
+    dp: 4
+    pp: 1
+    tp: 2
+    tp_linear_async_communication: true
+  batch_size: 8  # Optional batch size for evaluation
+  eval_config_override: "/fsx/elie_bakouch/smollm3_training/1004-nn/nanotron/_final1004/smollm3_eval_decay.yaml"  # The evaluation config file to use
+  eval_interval: 400000
+  eval_interval_file: null
+  slurm:
+    gpus_per_node: 8
+    partition: "hopper-prod"
+    hf_cache: "/fsx/elie_bakouch/.cache/huggingface"
+    cpus_per_task: 88
+    qos: "normal"
+    time: "01:59:00"
+    reservation: smollm  # Optional reservation name, can be null