method: lora_sft
base_model_id: meta-llama/Llama-3.1-8B-Instruct
seed: 43
exp_name: p1_sft_math_tooluse
git_commit: 8b979a30de6dfbf3b5a1052e42d8c0453b214d3f
dataset: lasgroup/SDPO
dataset_slug: sdpo_tooluse
manifest_path: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/manifest.yaml
tags:
  phase: P1
  domain: tool_use
hyperparams:
  model:
    base_model_id: meta-llama/Llama-3.1-8B-Instruct
    model_family: llama3
    target_modules:
    - q_proj
    - k_proj
    - v_proj
    - o_proj
    - gate_proj
    - up_proj
    - down_proj
  dataset:
    name: lasgroup/SDPO
    split: train
    text_field: prompt
    max_samples: null
    eval_samples: 256
    config: null
    domain: tool_use
    slug: sdpo_tooluse
    format: tooluse
    min_level: null
  sequence:
    max_length: 2048
    packing: true
  lora:
    r: 16
    alpha: 32
    dropout: 0.05
    target_modules:
    - q_proj
    - k_proj
    - v_proj
    - o_proj
    - gate_proj
    - up_proj
    - down_proj
  optimization:
    num_train_epochs: 3
    per_device_batch_size: 4
    gradient_accumulation_steps: 8
    learning_rate: 0.0002
    warmup_ratio: 0.05
    weight_decay: 0.01
    lr_scheduler_type: cosine
    max_grad_norm: 1.0
  checkpointing:
    num_checkpoints: 8
    save_total_limit: 64
    schedule: log
    save_steps: null
  runtime:
    logging_steps: 20
    bf16: true
    gradient_checkpointing: true
    wandb: true
    wandb_project: amr-fma-train
    hf_push: true
    hf_org: tkwiecinski
    hf_visibility: public
    force_restart: false
  sdpo: null
  evaluation:
    enabled: true
    eval_steps: 200
    strategy: steps
  prompt_style: null
  final_adapter_path: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/adapter_final
  total_steps: 114
checkpoints:
- step: 1
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-1
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-1
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 1.272672
    eval_runtime: 21.1731
    eval_samples_per_second: 3.826
    eval_steps_per_second: 0.992
    eval_perplexity: 3.570381
  hf_revision: step-00001
  hf_commit: 0d4986498ea3fc44e19079d35eee157277715f70
- step: 3
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-3
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-3
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 1.206122
    eval_runtime: 11.0618
    eval_samples_per_second: 7.323
    eval_steps_per_second: 1.898
    eval_perplexity: 3.340504
  hf_revision: step-00003
  hf_commit: 5848980eb5187a5b6a3c445cb411af3bd64e9680
- step: 7
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-7
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-7
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 0.781388
    eval_runtime: 11.0652
    eval_samples_per_second: 7.32
    eval_steps_per_second: 1.898
    eval_perplexity: 2.184502
  hf_revision: step-00007
  hf_commit: 3312a9ced0a76447d646da154b3009d3b37778b5
- step: 14
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-14
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-14
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 0.466285
    eval_runtime: 11.0629
    eval_samples_per_second: 7.322
    eval_steps_per_second: 1.898
    eval_perplexity: 1.594061
  hf_revision: step-00014
  hf_commit: 53df359874697e9811bf984311da87e0bf65903b
- step: 29
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-29
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-29
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 0.379769
    eval_runtime: 11.0591
    eval_samples_per_second: 7.324
    eval_steps_per_second: 1.899
    eval_perplexity: 1.461946
  hf_revision: step-00029
  hf_commit: 4531078bb439378d5a6bfdfdc88e4d4eef143a6d
- step: 57
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-57
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-57
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 0.271018
    eval_runtime: 11.0913
    eval_samples_per_second: 7.303
    eval_steps_per_second: 1.893
    eval_perplexity: 1.311298
  hf_revision: step-00057
  hf_commit: cb4e9fbb49d9ae4a48b123f49378459279fdc0fd
- step: 114
  dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-114
  artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s43/checkpoint-114
  metadata:
    source: trainer_on_save
  metrics:
    eval_loss: 0.148939
    eval_runtime: 11.059
    eval_samples_per_second: 7.324
    eval_steps_per_second: 1.899
    eval_perplexity: 1.160602
  hf_revision: step-00114
  hf_commit: f05f56d523c19166bda3280bb90fc989a0c28f23
wandb_run_id: 4l3ivdio
wandb_eval_run_ids: {}
hf_repo_id: tkwiecinski/amr-fma-Llama-3.1-8B-Instruct-lora_sft-sdpo_tooluse-p1_sft_math_tooluse-s43