qwen25-coder-32b-lora-sft

This model is a fine-tuned version of Qwen/Qwen2.5-Coder-32B on the endsky/sera-4.5-django-t2-recall05-toolcalls dataset. It achieves the following results on the evaluation set:

• Loss: 0.5649

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• total_eval_batch_size: 4
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 0.03
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
0.3451	0.0299	200	0.4959
0.3057	0.0599	400	0.4804
0.2954	0.0898	600	0.4790
0.2619	0.1198	800	0.4820
0.2838	0.1497	1000	0.4822
0.2632	0.1797	1200	0.4846
0.2637	0.2096	1400	0.4863
0.2301	0.2395	1600	0.4905
0.2479	0.2695	1800	0.4920
0.2380	0.2994	2000	0.4946
0.2504	0.3294	2200	0.4936
0.2273	0.3593	2400	0.5033
0.2382	0.3892	2600	0.5026
0.2491	0.4192	2800	0.5058
0.2369	0.4491	3000	0.5141
0.2173	0.4791	3200	0.5111
0.2254	0.5090	3400	0.5156
0.2091	0.5390	3600	0.5232
0.2192	0.5689	3800	0.5277
0.2280	0.5988	4000	0.5324
0.2012	0.6288	4200	0.5339
0.2277	0.6587	4400	0.5407
0.2314	0.6887	4600	0.5442
0.2280	0.7186	4800	0.5511
0.2208	0.7485	5000	0.5498
0.2399	0.7785	5200	0.5569
0.2248	0.8084	5400	0.5603
0.2249	0.8384	5600	0.5616
0.2243	0.8683	5800	0.5640
0.2562	0.8983	6000	0.5646
0.2246	0.9282	6200	0.5638
0.2333	0.9581	6400	0.5644
0.2171	0.9881	6600	0.5648

Framework versions

• PEFT 0.18.1
• Transformers 5.2.0
• Pytorch 2.11.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2