Training in progress, step 150, checkpoint

checkpoint-150/README.md

@@ -216,4 +216,23 @@ The following `bitsandbytes` quantization config was used during training:
 ### Framework versions
 
 
+- PEFT 0.6.0
+## Training procedure
+
+
+The following `bitsandbytes` quantization config was used during training:
+- quant_method: bitsandbytes
+- load_in_8bit: True
+- load_in_4bit: False
+- llm_int8_threshold: 6.0
+- llm_int8_skip_modules: None
+- llm_int8_enable_fp32_cpu_offload: False
+- llm_int8_has_fp16_weight: False
+- bnb_4bit_quant_type: fp4
+- bnb_4bit_use_double_quant: False
+- bnb_4bit_compute_dtype: float32
+
+### Framework versions
+
+
 - PEFT 0.6.0

checkpoint-150/adapter_config.json

@@ -1,7 +1,7 @@
 {
   "alpha_pattern": {},
   "auto_mapping": null,
-  "base_model_name_or_path": "/kaggle/input/vietcuna-for-binary-classification/kalapa-vietcuna-3b",
+  "base_model_name_or_path": "vietcuna-3b-v2/kalapa-vietcuna-3b/",
   "bias": "none",
   "fan_in_fan_out": false,
   "inference_mode": true,
@@ -9,10 +9,10 @@
   "layers_pattern": null,
   "layers_to_transform": null,
   "lora_alpha": 32,
-  "lora_dropout": 0.1,
+  "lora_dropout": 0.05,
   "modules_to_save": null,
   "peft_type": "LORA",
-  "r": 8,
+  "r": 16,
   "rank_pattern": {},
   "revision": null,
   "target_modules": [

checkpoint-150/adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:953b5c0a434279eedbdc0b0b8a9aa74658e382b2c83dca7011218ad7a0902776
-size 9859800
+oid sha256:dddcd65ff6df9d9f677eb4f14a43db2eb07482d82b13c30695cff8601f54c15e
+size 19690328

checkpoint-150/optimizer.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8a6e29cc344423bb6f9262ac1c7ec4968cedc8ed87ac2cef49f98122da71942b
-size 42724
+oid sha256:2bf8a8e294ad91dad98cdc10260e4bc5df16015bdea464a7c37a94878393d035
+size 38087098

checkpoint-150/rng_state.pth

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cc73ce8c65e40ac6855f7b2e3e6248898fe7fdbbaf2d613f95e2b6c995ee1bf
+oid sha256:55e71ed3f6f3ad40fcbc3c5cfb0b3bc77f998518af052b57e64a7852561eecf3
 size 14244

checkpoint-150/scheduler.pt

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9393be481ed85e8190dd814ca7d2cf441f97ed5dbde384528496c6d547e316ce
+oid sha256:c5dfe32eefef4e0fa34f6f96b3b2ec5e3e27c4f5298718cbc335d8fcca2a95dc
 size 1064

checkpoint-150/special_tokens_map.json

@@ -2,6 +2,6 @@
   "bos_token": "<s>",
   "eos_token": "</s>",
   "pad_token": "<pad>",
-  "sep_token": "<s>",
+  "sep_token": "\n\n",
   "unk_token": "<unk>"
 }

checkpoint-150/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723

checkpoint-150/tokenizer_config.json

@@ -39,7 +39,7 @@
   "eos_token": "</s>",
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<pad>",
-  "sep_token": "<s>",
+  "sep_token": "\n\n",
   "tokenizer_class": "BloomTokenizer",
   "unk_token": "<unk>"
 }

checkpoint-150/trainer_state.json

@@ -1,7 +1,7 @@
 {
-  "best_metric": 0.522564172744751,
-  "best_model_checkpoint": "output/checkpoint-50",
-  "epoch": 1.0471204188481675,
+  "best_metric": 1.2631328105926514,
+  "best_model_checkpoint": "output/checkpoint-150",
+  "epoch": 0.7657945118059988,
   "eval_steps": 50,
   "global_step": 150,
   "is_hyper_param_search": false,
@@ -10,61 +10,61 @@
   "log_history": [
     {
       "epoch": 0.01,
-      "learning_rate": 1.0000000000000001e-07,
-      "loss": 0.4196,
+      "learning_rate": 2.777777777777778e-07,
+      "loss": 9.5569,
       "step": 1
     },
     {
-      "epoch": 0.35,
-      "learning_rate": 4.6e-06,
-      "loss": 0.4279,
+      "epoch": 0.26,
+      "learning_rate": 1.25e-05,
+      "loss": 6.958,
       "step": 50
     },
     {
-      "epoch": 0.35,
-      "eval_accuracy": 0.7657142857142857,
-      "eval_loss": 0.522564172744751,
-      "eval_runtime": 29.0326,
-      "eval_samples_per_second": 6.028,
-      "eval_steps_per_second": 1.516,
+      "epoch": 0.26,
+      "eval_accuracy": 0.42857142857142855,
+      "eval_loss": 6.581231117248535,
+      "eval_runtime": 28.2966,
+      "eval_samples_per_second": 6.184,
+      "eval_steps_per_second": 1.555,
       "step": 50
     },
     {
-      "epoch": 0.7,
-      "learning_rate": 8.86829268292683e-06,
-      "loss": 0.4089,
+      "epoch": 0.51,
+      "learning_rate": 2.6388888888888892e-05,
+      "loss": 1.6141,
       "step": 100
     },
     {
-      "epoch": 0.7,
-      "eval_accuracy": 0.7428571428571429,
-      "eval_loss": 0.5401451587677002,
-      "eval_runtime": 29.0003,
-      "eval_samples_per_second": 6.034,
-      "eval_steps_per_second": 1.517,
+      "epoch": 0.51,
+      "eval_accuracy": 0.5028571428571429,
+      "eval_loss": 1.6640697717666626,
+      "eval_runtime": 28.1139,
+      "eval_samples_per_second": 6.225,
+      "eval_steps_per_second": 1.565,
       "step": 100
     },
     {
-      "epoch": 1.05,
-      "learning_rate": 7.792682926829268e-06,
-      "loss": 0.4018,
+      "epoch": 0.77,
+      "learning_rate": 4.027777777777778e-05,
+      "loss": 0.8533,
       "step": 150
     },
     {
-      "epoch": 1.05,
-      "eval_accuracy": 0.7428571428571429,
-      "eval_loss": 0.5425500273704529,
-      "eval_runtime": 28.9813,
-      "eval_samples_per_second": 6.038,
-      "eval_steps_per_second": 1.518,
+      "epoch": 0.77,
+      "eval_accuracy": 0.6342857142857142,
+      "eval_loss": 1.2631328105926514,
+      "eval_runtime": 26.205,
+      "eval_samples_per_second": 6.678,
+      "eval_steps_per_second": 1.679,
       "step": 150
     }
   ],
   "logging_steps": 50,
-  "max_steps": 500,
-  "num_train_epochs": 4,
+  "max_steps": 1000,
+  "num_train_epochs": 6,
   "save_steps": 50,
-  "total_flos": 5.643009513984e+16,
+  "total_flos": 5.595400201863168e+16,
   "trial_name": null,
   "trial_params": null
 }

checkpoint-150/training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b1b0a94354c0f3dfee9c93e225dbd48ad6f856f98977ef38ff3415910328e64a
+oid sha256:5a0ef36d7c31265f68a3edc8ae53ce9272dcc90a79f261e29a2a46e040c77a33
 size 4600