See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-0.5B
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - a45c981e5c0cf094_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/a45c981e5c0cf094_train_data.json
  type:
    field_input: news
    field_instruction: prompt
    field_output: out
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 2
eval_max_new_tokens: 128
eval_steps: 100
eval_table_size: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
group_by_length: false
hub_model_id: Romain-XV/a1b5e284-ac1d-4c2f-8a4a-dee6cdfd2f50
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_best_model_at_end: true
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 128
lora_dropout: 0.3
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 64
lora_target_linear: true
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 4224
micro_batch_size: 4
mlflow_experiment_name: /tmp/a45c981e5c0cf094_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 2
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
resume_from_checkpoint: null
s2_attention: null
sample_packing: false
save_steps: 100
sequence_len: 1024
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
use_rslora: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: af5cfc74-3e25-40ef-bc0e-b0384655d5f2
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: af5cfc74-3e25-40ef-bc0e-b0384655d5f2
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

a1b5e284-ac1d-4c2f-8a4a-dee6cdfd2f50

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: 1.0687

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: 8
• total_train_batch_size: 32
• 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: 1190

Training results

Training Loss	Epoch	Step	Validation Loss
1.5909	0.0017	1	1.5854
1.2535	0.1681	100	1.1993
1.1752	0.3363	200	1.1645
1.2031	0.5044	300	1.1371
1.1425	0.6726	400	1.1188
1.114	0.8407	500	1.1036
0.9845	1.0090	600	1.0898
0.9108	1.1772	700	1.0903
0.9669	1.3453	800	1.0849
0.951	1.5135	900	1.0761
0.8827	1.6816	1000	1.0706
0.9931	1.8497	1100	1.0687

Framework versions

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