End of training

README.md

@@ -22,7 +22,7 @@ axolotl version: `0.10.0.dev0`
 ```yaml
 base_model: minpeter/pretrained-tiny-ko
 
-chat_template: tokenizer_default_fallback_chatml
+chat_template: chatml
 datasets:
   - path: lemon-mint/Korean-FineTome-100k
     type: chat_template
@@ -46,12 +46,16 @@ output_dir: ./ouputs/ko-tiny-exp
 wandb_project: "axolotl"
 wandb_entity: "kasfiekfs-e"
 
-sequence_len: 2048
+save_steps: 200
+warmup_steps: 100
+eval_steps: 200
+
+sequence_len: 512
 sample_packing: true
 pad_to_sequence_len: true
 
 gradient_accumulation_steps: 4
-micro_batch_size: 16
+micro_batch_size: 56
 
 optimizer: paged_adamw_8bit
 lr_scheduler: cosine
@@ -60,6 +64,15 @@ learning_rate: 2e-5
 bf16: auto
 tf32: false
 
+added_tokens_overrides:
+  128001: "<|im_end|>"
+  128002: "<|im_start|>"
+
+special_tokens:
+  bos_token: <|begin_of_text|>
+  eos_token: <|im_end|>
+  pad_token: <|im_end|>
+
 gradient_checkpointing: true
 gradient_checkpointing_kwargs:
   use_reentrant: false
@@ -67,10 +80,7 @@ resume_from_checkpoint:
 logging_steps: 1
 flash_attention: true
 
-warmup_steps: 100
 num_epochs: 2
-evals_per_epoch: 2
-saves_per_epoch: 1
 weight_decay: 0.0
 
 ```
@@ -81,7 +91,7 @@ weight_decay: 0.0
 
 This model is a fine-tuned version of [minpeter/pretrained-tiny-ko](https://huggingface.co/minpeter/pretrained-tiny-ko) on the lemon-mint/Korean-FineTome-100k and the lemon-mint/smol-koreantalk datasets.
 It achieves the following results on the evaluation set:
-- Loss: 2.8226
+- Loss: 3.6623
 
 ## Model description
 
@@ -101,24 +111,24 @@ More information needed
 
 The following hyperparameters were used during training:
 - learning_rate: 2e-05
-- train_batch_size: 16
-- eval_batch_size: 16
+- train_batch_size: 56
+- eval_batch_size: 56
 - seed: 42
+- distributed_type: multi-GPU
+- num_devices: 4
 - gradient_accumulation_steps: 4
-- total_train_batch_size: 64
+- total_train_batch_size: 896
+- total_eval_batch_size: 224
 - optimizer: Use OptimizerNames.PAGED_ADAMW_8BIT 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: 100
-- training_steps: 1498
+- training_steps: 48
 
 ### Training results
 
 | Training Loss | Epoch  | Step | Validation Loss |
 |:-------------:|:------:|:----:|:---------------:|
-| 3.0001        | 0.0013 | 1    | 2.9904          |
-| 2.8288        | 0.5002 | 375  | 2.8669          |
-| 2.8188        | 1.0    | 750  | 2.8255          |
-| 2.8012        | 1.5002 | 1125 | 2.8226          |
+| 3.6868        | 0.0404 | 1    | 3.6623          |
 
 
 ### Framework versions

generation_config.json

@@ -1,8 +1,8 @@
 {
   "_from_model_config": true,
-  "bos_token_id": 128256,
+  "bos_token_id": 128000,
   "do_sample": true,
-  "eos_token_id": 128257,
+  "eos_token_id": 128001,
   "transformers_version": "4.51.3",
   "use_cache": false
 }