mssongit/Qwen2.5-14B-SFT-LoRA-6e-5

See axolotl config

axolotl version: 0.5.0

base_model: Qwen/Qwen2.5-14B-Instruct

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: output.jsonl
    type: alpaca

special_tokens:
  bos_token:
  eos_token: "<|im_end|>"
  pad_token: "<|endoftext|>"

dataset_prepared_path:
val_set_size: 0.05
output_dir: ./outputs/lora-out

sequence_len: 4096
sample_packing: false
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: axolotl_gmatrix
wandb_entity: mssong
wandb_watch:
wandb_run_id: 
wandb_log_model:

gradient_accumulation_steps: 2
micro_batch_size: 1
num_epochs: 3
optimizer:
lr_scheduler: cosine
learning_rate: 0.00006
train_on_inputs:
group_by_length: false
bf16: true
fp16: false
tf32: false
gradient_checkpointing: true
early_stopping_patience: 4
local_rank:
logging_steps: 100
xformers_attention:
flash_attention: true
warmup_ratio: 0.05
#warmup_steps: 100
eval_steps: 100
save_steps: 100
save_total_limit: 2
eval_sample_packing:
debug:
deepspeed:
weight_decay: 0.01
fsdp:
fsdp_config:
trust_remote_code: true

outputs/lora-out

This model is a fine-tuned version of Qwen/Qwen2.5-14B-Instruct on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0875

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: 6e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 2
• optimizer: Use OptimizerNames.ADAMW_HF 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: 162
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0009	1	0.4243
0.2465	0.0923	100	0.1338
0.0425	0.1847	200	0.1110
0.0333	0.2770	300	0.1051
0.0319	0.3693	400	0.0933
0.0257	0.4617	500	0.0886
0.0245	0.5540	600	0.0898
0.0262	0.6464	700	0.0889
0.025	0.7387	800	0.0827
0.0221	0.8310	900	0.0813
0.0207	0.9234	1000	0.0901
0.0219	1.0157	1100	0.0878
0.0132	1.1080	1200	0.0890
0.0154	1.2004	1300	0.0875

Framework versions

• PEFT 0.13.2
• Transformers 4.46.1
• Pytorch 2.3.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.3