See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: Qwen/Qwen2-0.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - 885d04d2490f6ef8_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/885d04d2490f6ef8_train_data.json
  type:
    field_input: Example
    field_instruction: '@partOfSpeech'
    field_output: Definition
    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: romainnn/70f19d31-85d8-45f7-9439-955b1b8bc7e1
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: 32
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 16
lora_target_linear: true
lora_target_modules:
- q_proj
- k_proj
- v_proj
lr_scheduler: cosine
max_grad_norm: 1.0
max_steps: 8832
micro_batch_size: 4
mlflow_experiment_name: /tmp/885d04d2490f6ef8_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
val_set_size: 0.041671875651123055
wandb_entity: null
wandb_mode: online
wandb_name: 36a8de66-ecec-4b15-ab19-b5863b4c1a11
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: 36a8de66-ecec-4b15-ab19-b5863b4c1a11
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

70f19d31-85d8-45f7-9439-955b1b8bc7e1

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

• Loss: 2.8708

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: 7187

Training results

Training Loss	Epoch	Step	Validation Loss
5.9812	0.0003	1	5.9869
2.9023	0.0278	100	3.2337
3.3221	0.0557	200	3.1458
3.0076	0.0835	300	3.1012
2.865	0.1113	400	3.0699
3.0561	0.1391	500	3.0413
3.1306	0.1670	600	3.0259
3.1343	0.1948	700	3.0144
2.9414	0.2226	800	2.9986
2.9757	0.2505	900	2.9911
3.2082	0.2783	1000	2.9812
2.8002	0.3061	1100	2.9719
2.7507	0.3339	1200	2.9659
2.6323	0.3618	1300	2.9565
3.0745	0.3896	1400	2.9542
3.0388	0.4174	1500	2.9435
2.9508	0.4453	1600	2.9356
2.7546	0.4731	1700	2.9307
2.6985	0.5009	1800	2.9260
2.7927	0.5288	1900	2.9199
2.841	0.5566	2000	2.9151
3.0443	0.5844	2100	2.9115
2.7652	0.6122	2200	2.9022
3.068	0.6401	2300	2.9016
2.8894	0.6679	2400	2.8940
3.0927	0.6957	2500	2.8911
2.8307	0.7236	2600	2.8894
2.7508	0.7514	2700	2.8839
3.1238	0.7792	2800	2.8822
2.9479	0.8070	2900	2.8736
2.8698	0.8349	3000	2.8748
2.7183	0.8627	3100	2.8701
2.7572	0.8905	3200	2.8613
2.9963	0.9184	3300	2.8580
2.6802	0.9462	3400	2.8532
2.8935	0.9740	3500	2.8504
2.6542	1.0019	3600	2.8475
2.4456	1.0297	3700	2.8660
2.5414	1.0576	3800	2.8708

Framework versions

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