See axolotl config

axolotl version: 0.4.1

adapter: qlora
auto_resume_from_checkpoints: true
base_model: Vikhrmodels/Vikhr-7B-instruct_0.4
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - eab93cbd8a081369_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/eab93cbd8a081369_train_data.json
  type:
    field_input: info
    field_instruction: classical
    field_output: modern
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
eval_max_new_tokens: 128
eval_steps: 50
eval_table_size: null
evals_per_epoch: null
flash_attention: true
fp16: false
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 16
gradient_checkpointing: true
group_by_length: false
hub_model_id: error577/9b376dea-cf02-48e4-9582-5c0c293e48f1
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0002
load_in_4bit: true
load_in_8bit: false
local_rank: null
logging_steps: 1
lora_alpha: 128
lora_dropout: 0.05
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: null
micro_batch_size: 1
mlflow_experiment_name: /tmp/eab93cbd8a081369_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 3
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: 50
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.0005
wandb_entity: null
wandb_mode: online
wandb_name: a8f4f147-a1f8-43a0-8042-6b029c99bc63
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: a8f4f147-a1f8-43a0-8042-6b029c99bc63
warmup_steps: 100
weight_decay: 0.01
xformers_attention: null

9b376dea-cf02-48e4-9582-5c0c293e48f1

This model is a fine-tuned version of Vikhrmodels/Vikhr-7B-instruct_0.4 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.9280

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: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 16
• total_train_batch_size: 16
• 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: 100
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
3.8857	0.0000	1	4.3076
2.7847	0.0008	50	2.4657
2.0785	0.0016	100	2.4571
2.4357	0.0025	150	2.3172
2.2992	0.0033	200	2.2529
2.6233	0.0041	250	2.1813
1.9701	0.0049	300	2.1299
2.2295	0.0058	350	2.1234
2.4176	0.0066	400	2.0745
2.4644	0.0074	450	2.0456
2.051	0.0082	500	2.0536
2.8109	0.0091	550	2.0406
1.911	0.0099	600	2.0164
2.3129	0.0107	650	1.9792
1.5766	0.0115	700	1.9626
2.246	0.0123	750	1.9866
1.9065	0.0132	800	1.9530
2.2483	0.0140	850	1.9408
2.4119	0.0148	900	1.9647
2.3117	0.0156	950	1.9133
2.2847	0.0165	1000	1.9454
1.8987	0.0173	1050	1.9336
1.7661	0.0181	1100	1.9280

Framework versions

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