This model is a fine-tuned version of Qwen/Qwen2.5-Coder-14B-Instruct on the fim_midtrain_v1, the fim_midtrain_v2, the fim_midtrain_v3_multi_pairs, the fim_midtrain_v3_multi_pairs_0317, the fim_midtrain_v3_multi_triples and the fim_midtrain_v3_multi_triples_0317 datasets.
RoPE config fix
This upload keeps the original weights and adds top-level "rope_theta": 1000000.0 to config.json so Qwen2 loaders in current Transformers/vLLM versions do not fall back to the default 10000.0 RoPE base.
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: 1
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 16
total_train_batch_size: 128
total_eval_batch_size: 64
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
