# 16-bit QAT Training Configuration for IasoQL
base_model: XGenerationLab/XiYanSQL-QwenCoder-7B-2504
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
trust_remote_code: true

# 16-bit QAT Configuration
qat: true
qat_config:
  bits: 16
  group_size: 128
  desc_act: false

# LoRA Configuration
adapter: lora
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj

# Dataset Configuration
datasets:
  - path: json
    data_files: xiyan-sql-clickhouse-training-dataset-enhanced.jsonl
    type: chat_template
    chat_template: qwen_25

# Training Configuration
sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

# Training Hyperparameters
micro_batch_size: 2
gradient_accumulation_steps: 4
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.00002

# Memory optimization
gradient_checkpointing: true
bf16: auto
tf32: false
load_in_4bit: true
load_in_8bit: true
flash_attention: true

# Logging
logging_steps: 10
save_steps: 25
save_total_limit: 2
warmup_steps: 10
output_dir: ./iasoql-16bit-outputs

# HuggingFace Hub - will upload manually after training
# hub_model_id: vivekkrishnan/iasoql-16bit-qat-lora
# push_to_hub: true
# hub_strategy: checkpoint

# Chat Template
chat_template: qwen_25

# Special Tokens
special_tokens:
  bos_token: "<|im_start|>"
  eos_token: "<|im_end|>"
  pad_token: "<|im_end|>"

# Settings
seed: 42

iasoql-16bit-outputs

This model is a fine-tuned version of XGenerationLab/XiYanSQL-QwenCoder-7B-2504 on the json 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: 2e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 8
optimizer: Use OptimizerNames.ADAMW_BNB 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: 10
num_epochs: 2.0

Training results

Framework versions

PEFT 0.14.0
Transformers 4.49.0
Pytorch 2.5.1+cu124
Datasets 3.2.0
Tokenizers 0.21.2

vivkris/iasoql-16bit-qat-lora

iasoql-16bit-outputs

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

vivkris/
iasoql-16bit-qat-lora