Best model at step 0 with eval_loss=1.6147 | WandB Run: https://wandb.ai/tuandung/Qwen2.5-Coder-1.5B-Instruct-LoRA-Training/runs/9c0w9tu1

README.md

@@ -18,7 +18,7 @@ should probably proofread and complete it, then remove this comment. -->
 This model is a fine-tuned version of [unsloth/qwen2.5-coder-1.5b-instruct-bnb-4bit](https://huggingface.co/unsloth/qwen2.5-coder-1.5b-instruct-bnb-4bit) on an unknown dataset.
 It achieves the following results on the evaluation set:
 - eval_loss: 1.6147
-- eval_runtime: 25.1219
+- eval_runtime: 25.1177
 - eval_samples_per_second: 3.583
 - eval_steps_per_second: 0.478
 - epoch: 0
@@ -42,16 +42,16 @@ More information needed
 
 The following hyperparameters were used during training:
 - learning_rate: 5e-05
-- train_batch_size: 20
+- train_batch_size: 16
 - eval_batch_size: 8
 - seed: 42
 - gradient_accumulation_steps: 2
-- total_train_batch_size: 40
+- total_train_batch_size: 32
 - optimizer: Use lion_8bit and the args are:
 No additional optimizer arguments
 - lr_scheduler_type: cosine
-- lr_scheduler_warmup_steps: 25
-- num_epochs: 3
+- lr_scheduler_warmup_steps: 73
+- num_epochs: 7
 - mixed_precision_training: Native AMP
 
 ### Framework versions

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9878c99c5737dc3a0e54256c19b7bb707f7cbb263d22c0e46db557fed8635366
+oid sha256:92a14eb7bb7c09bccd05835290ee1bd1f196ba2454f0de1e0fc4748e7d62d1da
 size 5560