| method: lora_sft |
| base_model_id: meta-llama/Llama-3.1-8B-Instruct |
| seed: 42 |
| 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__s42/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__s42/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__s42/checkpoint-1 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-1 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 1.272672 |
| eval_runtime: 21.2481 |
| eval_samples_per_second: 3.812 |
| eval_steps_per_second: 0.988 |
| eval_perplexity: 3.570381 |
| hf_revision: step-00001 |
| hf_commit: 916eb24b57f8d2dcfb110497b8f3c613e8ea753d |
| - step: 3 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-3 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-3 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 1.204667 |
| eval_runtime: 11.0516 |
| eval_samples_per_second: 7.329 |
| eval_steps_per_second: 1.9 |
| eval_perplexity: 3.335648 |
| hf_revision: step-00003 |
| hf_commit: 19fbaccdd2d5ddbd646298af3c5731887887adac |
| - step: 7 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-7 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-7 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 0.782293 |
| eval_runtime: 11.0496 |
| eval_samples_per_second: 7.331 |
| eval_steps_per_second: 1.901 |
| eval_perplexity: 2.186481 |
| hf_revision: step-00007 |
| hf_commit: b33504155bc707a35f2408a4ae4effed20d32b2f |
| - step: 14 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-14 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-14 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 0.465839 |
| eval_runtime: 11.0245 |
| eval_samples_per_second: 7.347 |
| eval_steps_per_second: 1.905 |
| eval_perplexity: 1.59335 |
| hf_revision: step-00014 |
| hf_commit: 5d85cecbbd00ca48efc94708f26059429e4d7ce4 |
| - step: 29 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-29 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-29 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 0.380428 |
| eval_runtime: 11.014 |
| eval_samples_per_second: 7.354 |
| eval_steps_per_second: 1.907 |
| eval_perplexity: 1.46291 |
| hf_revision: step-00029 |
| hf_commit: 9cc7d1a239b3fd77169323eca3a6792165dc09d9 |
| - step: 57 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-57 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-57 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 0.262946 |
| eval_runtime: 11.0386 |
| eval_samples_per_second: 7.338 |
| eval_steps_per_second: 1.902 |
| eval_perplexity: 1.300756 |
| hf_revision: step-00057 |
| hf_commit: 2a0898b8e223f1d0f87d4440f016b84194e3b4f0 |
| - step: 114 |
| dir: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-114 |
| artifact: /capstor/scratch/cscs/tkwiecinski/amr-fma/train/Llama-3.1-8B-Instruct/lora_sft/sdpo_tooluse/p1_sft_math_tooluse__s42/checkpoint-114 |
| metadata: |
| source: trainer_on_save |
| metrics: |
| eval_loss: 0.146911 |
| eval_runtime: 11.0978 |
| eval_samples_per_second: 7.299 |
| eval_steps_per_second: 1.892 |
| eval_perplexity: 1.158251 |
| hf_revision: step-00114 |
| hf_commit: 538668900639172bb147b158862aa0a2e62d1318 |
| wandb_run_id: l635jthe |
| wandb_eval_run_ids: {} |
| hf_repo_id: tkwiecinski/amr-fma-Llama-3.1-8B-Instruct-lora_sft-sdpo_tooluse-p1_sft_math_tooluse-s42 |
|
|
