AndyGulp/qwen-baseline

See axolotl config

axolotl version: 0.13.0.dev0

base_model: Qwen/Qwen2.5-Coder-7B-Instruct
bf16: true
dataset_prepared_path: last_run_prepared

# Dataset configuration for instruction/input/output format
datasets:
- chat_template: tokenizer_default
  field_messages: messages
  message_field_content: content
  message_field_role: role
  path: data_clean.jsonl
  roles:
    assistant:
    - assistant
    system:
    - system
    user:
    - user
  type: chat_template

debug: null
deepspeed: /osmosis/zero2.json
early_stopping_patience: null
flash_attention: true
fp16: null
fsdp: null
fsdp_config: null
gradient_accumulation_steps: 8
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
group_by_length: true
learning_rate: 1e-5
liger_fused_linear_cross_entropy: true
liger_glu_activation: true
liger_layer_norm: true
liger_rms_norm: true
liger_rope: true

logging_steps: 1

lr_scheduler: cosine
max_grad_norm: 1
micro_batch_size: 32
model_type: AutoModelForCausalLM
num_epochs: 1
optimizer: adamw_torch
output_dir: ./out-qwen-instruct
pad_to_sequence_len: true
plugins:
- axolotl.integrations.liger.LigerPlugin
resume_from_checkpoint: null
sample_packing: false
save_steps: 100
save_total_limit: 100
sequence_len: 4096
# special_tokens:
#   eos_token: <|im_end|>
strict: false
tf32: true
tokenizer_type: AutoTokenizer
train_on_inputs: false
trust_remote_code: true
val_set_size: 0.05
wandb_project: seedcoder
wandb_log_model: null
wandb_name: qwen-2-5-coder-7b-full-ft
wandb_watch: null
warmup_ratio: 0.05
weight_decay: 0.01
xformers_attention: null

out-qwen-instruct

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-7B-Instruct on the data_clean.jsonl dataset.

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 1024
• total_eval_batch_size: 128
• 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: 2
• training_steps: 31

Training results

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.3.0
• Tokenizers 0.22.1