Upload final model from sft-gemma4-st3-reversed

README.md

@@ -1,18 +1,18 @@
 ---
-base_model: alwaysgood/gemma4-CPT-stage2
+base_model: alwaysgood/gemma4-st2
 library_name: transformers
 model_name: checkpoints
 tags:
 - generated_from_trainer
-- unsloth
-- trl
 - sft
+- trl
+- unsloth
 licence: license
 ---
 
 # Model Card for checkpoints
 
-This model is a fine-tuned version of [alwaysgood/gemma4-CPT-stage2](https://huggingface.co/alwaysgood/gemma4-CPT-stage2).
+This model is a fine-tuned version of [alwaysgood/gemma4-st2](https://huggingface.co/alwaysgood/gemma4-st2).
 It has been trained using [TRL](https://github.com/huggingface/trl).
 
 ## Quick start
@@ -28,7 +28,7 @@ print(output["generated_text"])
 
 ## Training procedure
 
-[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/hiloong/instruction-tuning/runs/2fa83aug) 
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/hiloong/instruction-tuning/runs/wddqpt5q) 
 
 
 This model was trained with SFT.

all_results.json

@@ -1,12 +1,12 @@
 {
     "epoch": 1.0,
-    "eval_loss": 1.1339046955108643,
-    "eval_runtime": 9.804,
-    "eval_samples_per_second": 7.344,
-    "eval_steps_per_second": 0.918,
+    "eval_loss": 1.1255158185958862,
+    "eval_runtime": 7.6733,
+    "eval_samples_per_second": 9.383,
+    "eval_steps_per_second": 1.173,
     "total_flos": 1.054823986446048e+16,
-    "train_loss": 1.2565322134229873,
-    "train_runtime": 223.9359,
-    "train_samples_per_second": 15.911,
-    "train_steps_per_second": 0.402
+    "train_loss": 1.251299254099528,
+    "train_runtime": 241.5055,
+    "train_samples_per_second": 14.753,
+    "train_steps_per_second": 0.373
 }

config.json

@@ -53,7 +53,7 @@
   "eos_token_id": 1,
   "image_token_id": 258880,
   "initializer_range": 0.02,
-  "model_name": "alwaysgood/gemma4-CPT-stage2",
+  "model_name": "alwaysgood/gemma4-st2",
   "model_type": "gemma4",
   "pad_token_id": 0,
   "text_config": {

eval_results.json

@@ -1,7 +1,7 @@
 {
     "epoch": 1.0,
-    "eval_loss": 1.1339046955108643,
-    "eval_runtime": 9.804,
-    "eval_samples_per_second": 7.344,
-    "eval_steps_per_second": 0.918
+    "eval_loss": 1.1255158185958862,
+    "eval_runtime": 7.6733,
+    "eval_samples_per_second": 9.383,
+    "eval_steps_per_second": 1.173
 }

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7cbe467656fb4a2f0ecdca311de7728db6fb37214ed0ba9e2ee26e90474178dc
+oid sha256:2801109bf570b752396c0b0188bf531de3886b87c5f18a13dde556a390ab2e2f
 size 10208852910

train_results.json

@@ -1,8 +1,8 @@
 {
     "epoch": 1.0,
     "total_flos": 1.054823986446048e+16,
-    "train_loss": 1.2565322134229873,
-    "train_runtime": 223.9359,
-    "train_samples_per_second": 15.911,
-    "train_steps_per_second": 0.402
+    "train_loss": 1.251299254099528,
+    "train_runtime": 241.5055,
+    "train_samples_per_second": 14.753,
+    "train_steps_per_second": 0.373
 }

trainer_state.json

@@ -11,83 +11,83 @@
   "log_history": [
     {
       "epoch": 0.11204481792717087,
-      "grad_norm": 38.5,
+      "grad_norm": 29.25,
       "learning_rate": 1e-05,
-      "loss": 1.4108799934387206,
+      "loss": 1.3767690658569336,
       "step": 10
     },
     {
       "epoch": 0.22408963585434175,
-      "grad_norm": 24.125,
+      "grad_norm": 23.0,
       "learning_rate": 9.628619846344453e-06,
-      "loss": 1.277059268951416,
+      "loss": 1.287515354156494,
       "step": 20
     },
     {
       "epoch": 0.33613445378151263,
-      "grad_norm": 13.875,
+      "grad_norm": 18.0,
       "learning_rate": 8.569648672789496e-06,
-      "loss": 1.2860784530639648,
+      "loss": 1.2844683647155761,
       "step": 30
     },
     {
       "epoch": 0.4481792717086835,
-      "grad_norm": 16.5,
+      "grad_norm": 14.3125,
       "learning_rate": 6.980398830195785e-06,
-      "loss": 1.2376164436340331,
+      "loss": 1.2283113479614258,
       "step": 40
     },
     {
       "epoch": 0.5602240896358543,
-      "grad_norm": 10.75,
+      "grad_norm": 14.8125,
       "learning_rate": 5.096956658859122e-06,
-      "loss": 1.1717830657958985,
+      "loss": 1.1659849166870118,
       "step": 50
     },
     {
       "epoch": 0.6722689075630253,
-      "grad_norm": 19.375,
+      "grad_norm": 15.125,
       "learning_rate": 3.1991113759764493e-06,
-      "loss": 1.2510300636291505,
+      "loss": 1.2485424995422363,
       "step": 60
     },
     {
       "epoch": 0.7843137254901961,
-      "grad_norm": 13.125,
+      "grad_norm": 12.875,
       "learning_rate": 1.5687918106563326e-06,
-      "loss": 1.2010592460632323,
+      "loss": 1.2000310897827149,
       "step": 70
     },
     {
       "epoch": 0.896358543417367,
-      "grad_norm": 30.0,
+      "grad_norm": 17.5,
       "learning_rate": 4.481852951692672e-07,
-      "loss": 1.2197324752807617,
+      "loss": 1.219639205932617,
       "step": 80
     },
     {
       "epoch": 1.0,
-      "grad_norm": 53.25,
+      "grad_norm": 53.0,
       "learning_rate": 3.760237478849793e-09,
-      "loss": 1.253550910949707,
+      "loss": 1.250431442260742,
       "step": 90
     },
     {
       "epoch": 1.0,
-      "eval_loss": 1.1343798637390137,
-      "eval_runtime": 8.3336,
-      "eval_samples_per_second": 8.64,
-      "eval_steps_per_second": 1.08,
+      "eval_loss": 1.125481128692627,
+      "eval_runtime": 7.4107,
+      "eval_samples_per_second": 9.716,
+      "eval_steps_per_second": 1.214,
       "step": 90
     },
     {
       "epoch": 1.0,
       "step": 90,
       "total_flos": 1.054823986446048e+16,
-      "train_loss": 1.2565322134229873,
-      "train_runtime": 223.9359,
-      "train_samples_per_second": 15.911,
-      "train_steps_per_second": 0.402
+      "train_loss": 1.251299254099528,
+      "train_runtime": 241.5055,
+      "train_samples_per_second": 14.753,
+      "train_steps_per_second": 0.373
     }
   ],
   "logging_steps": 10,

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4d59593836281d181d61beef9744d72ba2292a9efe5d259c83f3e332c8f135b5
+oid sha256:4fc372f6e22c3b97a824f3cd3eb9fd8288225b67895aaa0386f45245b33d0292
 size 5777