Qwen2.5-14B-CPT-lora

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.5.0

# /home/az/Bureau/FINE_TUNING_VASTAI_CLM+SFT_SUR_BASE_MODEL/02-fineTune-CLM/axolotl_config.yaml

base_model: Qwen/Qwen2.5-14B 
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

seed: 42

load_in_8bit: false
load_in_4bit: false

# Dataset CLM pur
datasets:
  - path: philipperen55/dataset50CFT
    data_files: dataset50CFT.jsonl
    type: completion
    field: text

dataset_prepared_path: /workspace/prepared_data
val_set_size: 0.003
output_dir: /workspace/output

# Séquence et packing
sequence_len: 2048
sample_packing: false
pad_to_sequence_len: false
group_by_length: true
train_on_eos: true

# cela fonctionne pour Qwen, il met automatiquement le eot # refaire un test pour ministal, et sans doute mettre append_concat_token: true !!!!
append_concat_token: false
add_eos_token: false

# LoRA
adapter: lora
lora_r: 128
lora_alpha: 256
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj
  
# Training
micro_batch_size: 3
gradient_accumulation_steps: 32
num_epochs: 1
learning_rate: 2e-4
lr_scheduler: constant_with_warmup 

warmup_ratio: 0.03
optimizer: adamw_torch_fused

# Précision
bf16: true
fp16: false
tf32: true

weight_decay: 0.0
max_grad_norm: 1.0

flash_attention: true
sdp_attention: false

overrides_of_model_config:
  attn_implementation: flash_attention_2  # forcer FA2 jusqu’au modèle
  use_cache: false

  
gradient_checkpointing: false #true

# Logging et sauvegardes
logging_steps: 10
save_steps: 500
save_total_limit: 1
save_only_model: true
save_safetensors: true


eval_strategy: steps      
eval_steps: 500   


# WandB
wandb_project: Qwen2.5-14B-CPT-lora

# Hub
hub_model_id: philipperen55/Qwen2.5-14B-CPT-lora
push_to_hub: true
hub_strategy: every_save

</details><br>

Qwen2.5-14B-CPT-lora

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

• Loss: 1.9034

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: 3
• eval_batch_size: 3
• seed: 42
• gradient_accumulation_steps: 32
• total_train_batch_size: 96
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 110
• num_epochs: 1

Training results

Framework versions

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