You can deploy this model and start earning money today!

philipperen55/
Qwen2.5-14B-style-SFT-v3

Name: Qwen2.5-14B-style-SFT-v3
Availability: InStock

generated_from_trainer

philipperen55/Qwen2.5-14B-style-MERGED-v2

adapter

Views

No views yet

See axolotl config

axolotl version: 0.5.0

yaml

# =============================================================================
# PHASE 2 V3 : SFT Réécriture - FORMAT COMPLETION (sans template Alpaca)
# =============================================================================

base_model: philipperen55/Qwen2.5-14B-style-MERGED-v2
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

load_in_8bit: false
load_in_4bit: false

# =============================================================================
# DATASET : Format completion pré-formaté (pas de template ajouté par Axolotl)
# =============================================================================
datasets:
  - path: philipperen55/dataset40FT
    data_files: dataset40FT.jsonl
    type: completion
    field: text

dataset_prepared_path: /workspace/prepared_data
val_set_size: 0.05
output_dir: /workspace/output

# =============================================================================
# SÉQUENCE - SANS PACKING
# =============================================================================
sequence_len: 2048
sample_packing: false
pad_to_sequence_len: true

# =============================================================================
# TOKEN EOS EXPLICITE
# =============================================================================
special_tokens:
  eos_token: "<|endoftext|>"

# =============================================================================
# LoRA
# =============================================================================
adapter: lora
lora_r: 32
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj

# =============================================================================
# TRAINING - 33K PAIRES, 2 EPOCHS, LR 1.5e-5
# =============================================================================
gradient_accumulation_steps: 2
micro_batch_size: 8
num_epochs: 2
learning_rate: 1.5e-5
lr_scheduler: cosine
warmup_ratio: 0.05
optimizer: adamw_torch

# =============================================================================
# PRÉCISION
# =============================================================================
bf16: true
fp16: false
tf32: true

# =============================================================================
# OPTIMISATIONS
# =============================================================================
flash_attention: false
gradient_checkpointing: true

# =============================================================================
# LOGGING ET SAUVEGARDES
# =============================================================================
logging_steps: 50
save_steps: 500
eval_steps: 500
save_total_limit: 8

# =============================================================================
# WEIGHTS & BIASES
# =============================================================================
wandb_project: Qwen2.5-14B-SFT-phase2-v3

# =============================================================================
# HUGGING FACE HUB
# =============================================================================
hub_model_id: philipperen55/Qwen2.5-14B-style-SFT-v3
push_to_hub: true
hub_strategy: every_save

Qwen2.5-14B-style-SFT-v3

This model is a fine-tuned version of philipperen55/Qwen2.5-14B-style-MERGED-v2 on the None dataset. It achieves the following results on the evaluation set:

Loss: 1.3685

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: 1.5e-05
train_batch_size: 8
eval_batch_size: 8
seed: 42
gradient_accumulation_steps: 2
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: 195
num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0005	1	1.6796
1.4216	0.2557	500	1.4063
1.3818	0.5114	1000	1.3881
1.3878	0.7671	1500	1.3792
1.3816	1.0228	2000	1.3741
1.3631	1.2784	2500	1.3711
1.3541	1.5341	3000	1.3693
1.3653	1.7898	3500	1.3685

Framework versions

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