base_model: Qwen/Qwen2.5-7B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true
load_in_8bit: false
load_in_4bit: false

datasets:
  - path: furkancekic/turkish-finance-instructions-v2
    type: alpaca
    chat_template: chatml
    split: train

dataset_prepared_path: last_run_prepared
val_set_size: 0.02
output_dir: ./outputs/qwen7b-turkish-finance-v2

adapter: lora
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj

sequence_len: 1024
sample_packing: true
pad_to_sequence_len: true

micro_batch_size: 4
gradient_accumulation_steps: 4
num_epochs: 2
learning_rate: 1e-4
lr_scheduler: cosine
warmup_ratio: 0.05
weight_decay: 0.01
optimizer: adamw_torch

gradient_checkpointing: true
bf16: auto
tf32: true

logging_steps: 10
save_strategy: steps
save_steps: 200
save_total_limit: 3
eval_steps: 200

wandb_project: turkish-finance-slm-v2
wandb_run_id:
wandb_log_model:

train_on_inputs: false
special_tokens:
  pad_token: "<|endoftext|>"

outputs/qwen7b-turkish-finance-v2

This model is a fine-tuned version of Qwen/Qwen2.5-7B-Instruct on the furkancekic/turkish-finance-instructions-v2 dataset. It achieves the following results on the evaluation set:

Loss: 0.8145
Memory/max Active (gib): 21.94
Memory/max Allocated (gib): 21.94
Memory/device Reserved (gib): 25.91

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: 0.0001
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH 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: 114
training_steps: 2296

Training results

Training Loss	Epoch	Step	Validation Loss	Active (gib)	Allocated (gib)	Reserved (gib)
No log	0	0	6.1174	21.3	21.3	21.42
1.1934	0.1741	200	1.2171	21.94	21.94	26.14
1.0438	0.3482	400	1.0703	21.94	21.94	25.91
0.979	0.5223	600	0.9918	21.94	21.94	25.91
0.9547	0.6964	800	0.9400	21.94	21.94	25.91
0.9103	0.8705	1000	0.9004	21.94	21.94	25.91
0.8156	1.0444	1200	0.8716	21.94	21.94	25.91
0.799	1.2185	1400	0.8502	21.94	21.94	25.91
0.7954	1.3926	1600	0.8339	21.94	21.94	25.91
0.7699	1.5667	1800	0.8228	21.94	21.94	25.91
0.7713	1.7408	2000	0.8165	21.94	21.94	25.91
0.7638	1.9149	2200	0.8145	21.94	21.94	25.91

Framework versions

PEFT 0.17.1
Transformers 4.57.0
Pytorch 2.7.1+cu126
Datasets 4.0.0
Tokenizers 0.22.1

furkancekic/turkish-finance-qwen7b-v2-lora

outputs/qwen7b-turkish-finance-v2

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

furkancekic/
turkish-finance-qwen7b-v2-lora