See axolotl config

axolotl version: 0.4.1

adapter: lora
auto_find_batch_size: true
base_model: NousResearch/Yarn-Llama-2-7b-128k
bf16: auto
chat_template: llama3
dataloader_num_workers: 12
dataset_prepared_path: null
datasets:
- data_files:
  - cff7ac798e6d5dcd_train_data.json
  ds_type: json
  format: custom
  path: /workspace/input_data/cff7ac798e6d5dcd_train_data.json
  type:
    field_input: input
    field_instruction: instruction
    field_output: response
    format: '{instruction} {input}'
    no_input_format: '{instruction}'
    system_format: '{system}'
    system_prompt: ''
debug: null
deepspeed: null
early_stopping_patience: 3
early_stopping_threshold: 0.001
eval_max_new_tokens: 128
eval_steps: 40
flash_attention: false
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 2
gradient_checkpointing: false
group_by_length: false
hub_model_id: mrferr3t/b66756af-9ce0-428c-afcd-509c8a65e711
hub_repo: null
hub_strategy: checkpoint
hub_token: null
learning_rate: 0.0003
load_in_4bit: false
load_in_8bit: false
local_rank: null
logging_steps: 100
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out: null
lora_model_dir: null
lora_r: 8
lora_target_linear: true
lr_scheduler: cosine
micro_batch_size: 32
mlflow_experiment_name: /tmp/cff7ac798e6d5dcd_train_data.json
model_type: AutoModelForCausalLM
num_epochs: 50
optimizer: adamw_bnb_8bit
output_dir: miner_id_24
pad_to_sequence_len: true
s2_attention: null
sample_packing: false
save_steps: 40
saves_per_epoch: 0
sequence_len: 512
strict: false
tf32: false
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_entity: null
wandb_mode: online
wandb_name: cf8d9384-56f2-40e9-8877-64c5c8e6e996
wandb_project: Gradients-On-Demand
wandb_run: your_name
wandb_runid: cf8d9384-56f2-40e9-8877-64c5c8e6e996
warmup_ratio: 0.05
weight_decay: 0.0
xformers_attention: null

b66756af-9ce0-428c-afcd-509c8a65e711

This model is a fine-tuned version of NousResearch/Yarn-Llama-2-7b-128k on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5108

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Use adamw_bnb_8bit 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: 212
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0029	1	2.6011
No log	0.1178	40	2.1022
No log	0.2356	80	1.6311
4.1759	0.3535	120	1.4511
4.1759	0.4713	160	1.3336
2.8347	0.5891	200	1.2712
2.8347	0.7069	240	1.2134
2.8347	0.8247	280	1.1423
2.5511	0.9426	320	1.0515
2.5511	1.0604	360	1.0495
1.8608	1.1782	400	0.9674
1.8608	1.2960	440	0.8957
1.8608	1.4138	480	0.8620
1.5922	1.5317	520	0.8250
1.5922	1.6495	560	0.7791
1.4457	1.7673	600	0.7836
1.4457	1.8851	640	0.7161
1.4457	2.0029	680	0.6826
1.2167	2.1208	720	0.6942
1.2167	2.2386	760	0.6890
0.813	2.3564	800	0.6791
0.813	2.4742	840	0.6476
0.813	2.5920	880	0.6133
0.8131	2.7099	920	0.5796
0.8131	2.8277	960	0.5654
0.7533	2.9455	1000	0.5553
0.7533	3.0633	1040	0.5500
0.7533	3.1811	1080	0.5648
0.4819	3.2990	1120	0.5604
0.4819	3.4168	1160	0.5394
0.4436	3.5346	1200	0.5321
0.4436	3.6524	1240	0.5429
0.4436	3.7703	1280	0.5055
0.5149	3.8881	1320	0.5246
0.5149	4.0059	1360	0.4923
0.3846	4.1237	1400	0.5086
0.3846	4.2415	1440	0.4987
0.3846	4.3594	1480	0.5108

Framework versions

• PEFT 0.13.2
• Transformers 4.46.0
• Pytorch 2.3.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1