base_model: "Qwen/Qwen2-0.5B"
model_type: "AutoModelForCausalLM"
tokenizer_type: "AutoTokenizer"

load_in_8bit: false
load_in_4bit: true
strict: false
chat_template: "llama3"

datasets:
  - path: "/workspace/input_data/train_data.json"
    format: "custom"
    type:
      system_prompt: ""
      system_format: "{system}"
      field_instruction: "prompt"
      field_output: "question"
      no_input_format: "{instruction}"
      format: "{instruction}"
    ds_type: "json"
    data_files:
      - "train_data.json"

dataset_prepared_path: null
val_set_size: 0.04
output_dir: "miner_id_24"

sequence_len: 1024
sample_packing: false
pad_to_sequence_len: true
trust_remote_code: true

adapter: "lora"
lora_model_dir: null
lora_r: 64
lora_alpha: 128
lora_dropout: 0.3
lora_target_linear: true
lora_fan_in_fan_out: null

gradient_accumulation_steps: 6
micro_batch_size: 4
optimizer: "adamw_bnb_8bit"
lr_scheduler: "cosine"
learning_rate: 0.0002
num_epochs: 3
max_steps: 2
train_on_inputs: false
group_by_length: false

bf16: true
fp16: null
tf32: true
max_grad_norm: 1.0
gradient_checkpointing: true
early_stopping_patience: 4

save_steps: 100
eval_steps: 100
resume_from_checkpoint: null
local_rank: null
logging_steps: 1

xformers_attention: null
flash_attention: true
s2_attention: null

load_best_model_at_end: true

wandb_project: "Gradients-On-Demand"
wandb_entity: null
wandb_mode: "online"
wandb_run: "your_name"
wandb_runid: "c29f8be0-1d6a-40dd-83f1-f1d58697725a"

hub_model_id: "Alphatao/a266814f-d053-48ed-8aa8-1dc9cf4fd72f"
hub_repo: null
hub_strategy: "checkpoint"
hub_token: null

warmup_steps: 10
eval_table_size: null
eval_max_new_tokens: 128

debug: null
deepspeed: null
weight_decay: 0.0
fsdp: null
fsdp_config: null

wandb_name: "c29f8be0-1d6a-40dd-83f1-f1d58697725a"
lora_target_modules: ["q_proj", "k_proj", "v_proj"]
mlflow_experiment_name: "/tmp/train_data.json"

a266814f-d053-48ed-8aa8-1dc9cf4fd72f

This model is a fine-tuned version of Qwen/Qwen2-0.5B on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.4452

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.0002
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 6
total_train_batch_size: 24
optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_steps: 10
training_steps: 2

Training results

Training Loss	Epoch	Step	Validation Loss
0.26	0.0006	1	0.4452

Framework versions

PEFT 0.13.2
Transformers 4.46.0
Pytorch 2.5.0+cu124
Datasets 3.0.1
Tokenizers 0.20.1