End of training

README.md

@@ -64,7 +64,7 @@ lora_model_dir: null
 lora_r: 8
 lora_target_linear: true
 lr_scheduler: cosine
-max_steps: 1
+max_steps: 50
 micro_batch_size: 2
 mlflow_experiment_name: /tmp/6db463ae20360f31_train_data.json
 model_type: AutoModelForCausalLM
@@ -91,7 +91,7 @@ wandb_name: ee3aef79-53ae-409c-8fc7-27d9f89cced9
 wandb_project: Gradients-On-Demand
 wandb_run: your_name
 wandb_runid: ee3aef79-53ae-409c-8fc7-27d9f89cced9
-warmup_steps: 1
+warmup_steps: 10
 weight_decay: 0.0
 xformers_attention: null
 
@@ -102,6 +102,8 @@ xformers_attention: null
 # e5b09d94-c8c8-740a-7e35-1898127b854e
 
 This model is a fine-tuned version of [Korabbit/llama-2-ko-7b](https://huggingface.co/Korabbit/llama-2-ko-7b) on the None dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.8009
 
 ## Model description
 
@@ -128,14 +130,19 @@ The following hyperparameters were used during training:
 - 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: 2
-- training_steps: 1
+- lr_scheduler_warmup_steps: 10
+- training_steps: 50
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
 | No log        | 0.0035 | 1    | 7.8489          |
+| 7.2117        | 0.0354 | 10   | 5.4993          |
+| 1.4235        | 0.0707 | 20   | 0.8844          |
+| 1.0252        | 0.1061 | 30   | 0.8173          |
+| 0.7385        | 0.1415 | 40   | 0.8148          |
+| 0.855         | 0.1768 | 50   | 0.8009          |
 
 
 ### Framework versions

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a318a32231629b28e53f60d16b205d94c9ad99f1a4b609b2bf9c56f4fdbf4303
+oid sha256:63d9f8b77ff924a50d40da01f8962fc76eaee3134536843cd9b9634aef5bbc16
 size 80115210