• Model creator: https://huggingface.co/Basticooler/
• Original model: https://huggingface.co/Basticooler/ML4Science-meditron/

Original model description:

• generated_from_trainer model-index:
• name: mloscratch/homes/bbernath/meditron_instruct/instruction_tuned_model_with_ml4science_data results: []

---

See axolotl config

axolotl version: 0.4.0

yaml

1base_model: OpenMeditron/Meditron3-8B 2bf16: auto 3output_dir: /mloscratch/homes/bbernath/meditron_instruct/instruction_tuned_model_with_ml4science_data 4chat_template: llama3 5datasets: 6 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/mixtures/Instruction_tuning_mixture.jsonl 7 type: chat_template 8 ds_type: json 9 split: train 10 field_messages: conversations 11 message_field_role: from 12 message_field_content: value 13 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/replay_data/datasets/pubmed_3B.jsonl 14 type: completion 15 ds_type: json 16 field: text 17 sample_ratio: 0.1 18 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/replay_data/datasets/amboss_article.jsonl 19 type: completion 20 ds_type: json 21 field: text 22 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/replay_data/datasets/medmcqa.jsonl 23 type: chat_template 24 ds_type: json 25 split: train 26 field_messages: conversations 27 message_field_role: from 28 message_field_content: value 29 sample_ratio: 0.5 30 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/replay_data/datasets/pubmedqa.jsonl 31 type: chat_template 32 ds_type: json 33 split: train 34 field_messages: conversations 35 message_field_role: from 36 message_field_content: value 37 sample_ratio: 0.2 38 - path: /mloscratch/homes/bbernath/meditron_instruct/datasets/replay_data/datasets/medqa.jsonl 39 type: chat_template 40 ds_type: json 41 split: train 42 field_messages: conversations 43 message_field_role: from 44 message_field_content: value 45shuffle_merged_datasets: true 46dataset_processes: 64 47flash_attention: true 48sequence_len: 8192 49gradient_accumulation_steps: 2 50micro_batch_size: 2 51train_on_inputs: false 52group_by_length: false 53pad_to_sequence_len: true 54sample_packing: true 55optimizer: adamw_torch 56optim_args: 57 fused: true 58cosine_min_lr_ratio: 0.1 59learning_rate: 1.0e-5 60warmup_ratio: 0.0 61weight_decay: 0.05 62gradient_checkpointing: true 63gradient_checkpointing_kwargs: 64 use_reentrant: false 65load_in_4bit: false 66load_in_8bit: false 67logging_steps: 1 68num_epochs: 1 69# saves_per_epoch: 1 70# evals_per_epoch: 2 71eval_set_size: 0.0 72eval_table_size: null 73lr_scheduler: cosine 74max_grad_norm: 1.0 75resume_from_checkpoint: null 76special_tokens: 77 pad_token: <|end_of_text|> 78tf32: false 79tokenizer_type: AutoTokenizer 80type: LlamaForCausalLM 81seed: 42 82flash_attn_rms_norm: true 83flash_attn_fuse_qkv: false 84early_stopping_patience: 0 85eval_steps: 3000 86save_steps: 3000 87load_best_model_at_end: true 88xformers_attention: null 89distributed: 90 world_size: 5 91 backend: nccl 92deepspeed: /mloscratch/homes/bbernath/meditron_instruct/axolotl_config/ds_config.json 93 94wandb_project: Meditron DDX 95wandb_entity: alexs-team 96wandb_name: Instruction_tune_Meditron_8b_with_ML4Science_dataset_10000_first_try

mloscratch/homes/bbernath/meditron_instruct/instruction_tuned_model_with_ml4science_data

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

• learning_rate: 1e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 2
• total_train_batch_size: 24
• total_eval_batch_size: 12
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=fused=True
• lr_scheduler_type: cosine
• num_epochs: 1

Training results

Framework versions

• Transformers 4.46.1
• Pytorch 2.5.1+cu124
• Datasets 3.0.1
• Tokenizers 0.20.3