See axolotl config

axolotl version: 0.4.0

base_model: Qwen/Qwen1.5-0.5B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

# is_qwen_derived_model: true
trust_remote_code: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: OdiaGenAIdata/culturax-odia
    type: completion
dataset_prepared_path:
val_set_size: 0.05
output_dir: ./lora-out-qwen-0.5b-odia
hub_model_id: sam2ai/qwen_1.5_odia_0.5b

sequence_len: 2048  # supports up to 8192
sample_packing: false
pad_to_sequence_len:

adapter: qlora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: Qwen-completion-0.5b-odia
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 10
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention:

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

qwen_1.5_odia_0.5b

This model is a fine-tuned version of Qwen/Qwen1.5-0.5B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4242

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: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss
1.2821	0.0	1	1.2706
0.5906	0.25	1366	0.5987
0.531	0.5	2732	0.5510
0.5095	0.75	4098	0.5236
0.5027	1.0	5464	0.5054
0.5019	1.25	6830	0.4933
0.4798	1.5	8196	0.4845
0.4484	1.75	9562	0.4771
0.4526	2.0	10928	0.4704
0.4498	2.25	12294	0.4657
0.4508	2.5	13660	0.4608
0.4226	2.75	15026	0.4568
0.4161	3.0	16392	0.4539
0.4258	3.25	17758	0.4515
0.428	3.5	19124	0.4489
0.4748	3.75	20490	0.4459
0.4083	4.0	21856	0.4441
0.4278	4.25	23222	0.4423
0.3997	4.5	24588	0.4406
0.4581	4.75	25954	0.4386
0.378	5.0	27320	0.4372
0.4141	5.25	28686	0.4358
0.4017	5.5	30052	0.4344
0.4223	5.75	31418	0.4328
0.426	6.0	32784	0.4317
0.3967	6.25	34150	0.4310
0.3934	6.5	35516	0.4298
0.404	6.75	36882	0.4287
0.3874	7.0	38248	0.4282
0.384	7.25	39614	0.4275
0.3925	7.5	40980	0.4268
0.409	7.75	42346	0.4261
0.3891	8.0	43712	0.4256
0.41	8.25	45078	0.4253
0.3999	8.5	46444	0.4249
0.3874	8.75	47810	0.4247
0.3894	9.0	49176	0.4245
0.3827	9.25	50542	0.4244
0.3815	9.5	51908	0.4243
0.3816	9.75	53274	0.4242

Framework versions

• PEFT 0.8.2
• Transformers 4.37.0
• Pytorch 2.0.1+gita61a294
• Datasets 2.16.1
• Tokenizers 0.15.0