See axolotl config

axolotl version: 0.4.1

adapter: lora
base_model: unsloth/Qwen2.5-Math-1.5B-Instruct
bf16: true
chat_template: llama3
dataset_prepared_path: null
datasets:
- data_files:
  - e6ce2c5be55a88f5_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/e6ce2c5be55a88f5_train_data.json
  type:
    field_input: input_context
    field_instruction: instruction
    field_output: errors
    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/3005f4db-caca-4ec5-8a89-6c347fc709c8
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: 3060
micro_batch_size: 4
mlflow_experiment_name: /tmp/e6ce2c5be55a88f5_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: 0.04
wandb_entity: null
wandb_mode: online
wandb_name: d4f3db6a-bb4d-48c5-85a7-800d37c5ba55
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: d4f3db6a-bb4d-48c5-85a7-800d37c5ba55
warmup_steps: 10
weight_decay: 0.0
xformers_attention: null

3005f4db-caca-4ec5-8a89-6c347fc709c8

This model is a fine-tuned version of unsloth/Qwen2.5-Math-1.5B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8402

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: 2531

Training results

Training Loss	Epoch	Step	Validation Loss
2.58	0.0008	1	2.6726
1.202	0.0790	100	1.1156
0.8173	0.1580	200	1.0485
1.024	0.2371	300	1.0102
1.0818	0.3161	400	0.9853
0.932	0.3951	500	0.9615
1.1392	0.4741	600	0.9458
1.0125	0.5531	700	0.9289
0.8807	0.6322	800	0.9135
0.9482	0.7112	900	0.9032
0.8576	0.7902	1000	0.8924
0.8979	0.8692	1100	0.8834
0.9764	0.9482	1200	0.8754
0.8257	1.0273	1300	0.8716
0.7471	1.1063	1400	0.8663
0.6291	1.1853	1500	0.8608
0.7818	1.2643	1600	0.8569
0.8119	1.3433	1700	0.8531
0.8916	1.4224	1800	0.8494
0.8509	1.5014	1900	0.8464
0.702	1.5804	2000	0.8441
0.8121	1.6594	2100	0.8423
0.768	1.7384	2200	0.8413
0.6225	1.8175	2300	0.8406
0.9079	1.8965	2400	0.8402
0.8679	1.9755	2500	0.8402

Framework versions

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