base_model: Qwen/Qwen3-1.7B-Base

datasets:
  - path: nq_text_compressed_axolotl/train_with_pseudo_axolotl.jsonl
    type: chat_template
    chat_template: tokenizer_default_fallback_chatml
    field_messages: conversations
    message_property_mappings:
      role: role
      content: content
    roles:
      assistant:
        - assistant
        - gpt
        - model
      user:
        - user
        - human
      system:
        - system

roles_to_train: ["assistant"]
train_on_eos: "turn"

shuffle_merged_datasets: true
output_dir: ./checkpoint/Qwen3-1.7B-nq_text_compressed-with_pseudo-lr1e-4-10epochs
sequence_len: 1024
sample_packing: true
pad_to_sequence_len: true
flash_attention: false
xformers_attention: false
flex_attention: false
sdp_attention: true

gradient_accumulation_steps: 32
micro_batch_size: 4
dataloader_num_workers: 2
num_epochs: 10

optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 1e-4
warmup_ratio: 0.1
weight_decay: 0.0
bf16: true
tf32: false
gradient_checkpointing: true

logging_steps: 1000
save_strategy: epoch
save_total_limit: 1

special_tokens:
  eos_token: <|im_end|>

wandb_project: "ICLGR-NQ"
wandb_entity: "abnerden0803-national-taiwan-university"
wandb_name: "qwen3-1.7b-nq_text_compressed-pseudo-lr1e-4-10epochs"
wandb_log_model:
wandb_watch:

val_set_size: 0.0

checkpoint/Qwen3-1.7B-nq_text_compressed-with_pseudo-lr1e-4-10epochs

This model is a fine-tuned version of Qwen/Qwen3-1.7B-Base on the nq_text_compressed_axolotl/train_with_pseudo_axolotl.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: 0.0001
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 32
total_train_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: 319
training_steps: 3190

Training results

Framework versions

Transformers 5.5.0
Pytorch 2.8.0+cu128
Datasets 4.5.0
Tokenizers 0.22.2

Abner0803/qwen_nq_compressed

checkpoint/Qwen3-1.7B-nq_text_compressed-with_pseudo-lr1e-4-10epochs

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Abner0803/
qwen_nq_compressed