End of training

README.md

@@ -65,7 +65,7 @@ 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/b52f39ac5b733496_train_data.json
 model_type: AutoModelForCausalLM
@@ -92,7 +92,7 @@ wandb_name: a0f466d2-eb38-4586-8de4-ff7ee0adc033
 wandb_project: Gradients-On-Demand
 wandb_run: your_name
 wandb_runid: a0f466d2-eb38-4586-8de4-ff7ee0adc033
-warmup_steps: 1
+warmup_steps: 10
 weight_decay: 0.0
 xformers_attention: null
 
@@ -103,6 +103,8 @@ xformers_attention: null
 # 53ce8117-8c3f-addf-1d54-7535c4325586
 
 This model is a fine-tuned version of [JackFram/llama-160m](https://huggingface.co/JackFram/llama-160m) on the None dataset.
+It achieves the following results on the evaluation set:
+- Loss: 0.5383
 
 ## Model description
 
@@ -129,14 +131,19 @@ 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        | 0.0004 | 1    | 0.9877          |
+| 0.7846        | 0.0043 | 10   | 0.8959          |
+| 0.6315        | 0.0086 | 20   | 0.6548          |
+| 0.4356        | 0.0128 | 30   | 0.5717          |
+| 0.3842        | 0.0171 | 40   | 0.5428          |
+| 0.3988        | 0.0214 | 50   | 0.5383          |
 
 
 ### Framework versions

adapter_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:655b84bd06ff092b4e0f50395341e20a6adf39f18e90265fe73f4b051cc258b5
+oid sha256:2e1345b743256040a8b24cad20f22e9c3dfd6d71b51d6fa1aaebb38f0354f1e9
 size 6843306