Training in progress, step 20

README.md

@@ -4,8 +4,8 @@ library_name: transformers
 model_name: medgemma_lora
 tags:
 - generated_from_trainer
-- trl
 - sft
+- trl
 licence: license
 ---
 

adapter_config.json

@@ -25,8 +25,8 @@
   "rank_pattern": {},
   "revision": null,
   "target_modules": [
-    "v_proj",
-    "q_proj"
+    "q_proj",
+    "v_proj"
   ],
   "target_parameters": null,
   "task_type": "CAUSAL_LM",

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e0751ae67e7bdc676d7d24879cb0206affe0e7c4fabce5d3315dbf459e99a055
+oid sha256:043eb1e01faea6da2934972c4f8e8407fbcf31eb2b96afe4cc2a446c7b81fdbf
 size 25826856

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b5ba96187501fd47aa8a6fa34ce3ce3b7d1630e332fa48d80537f465630c7f0a
-size 6225
+oid sha256:79ea4b5c62dd2fb12515e971775d549f78e488ab8af5dcc925ab3710d08c4764
+size 6289

training_args.json

@@ -5,15 +5,15 @@
   "lora_target_modules": "q_proj,v_proj",
   "tuning_strategy": "lora",
   "num_trainable_layers": 5,
-  "output_dir": "checkpoints/pure/ALL/medgemma_lora",
+  "output_dir": "checkpoints/pure/Chest_xray_pa/medgemma_lora",
   "num_train_epochs": 3,
   "learning_rate": 0.0002,
-  "per_device_train_batch_size": 1,
-  "per_device_eval_batch_size": 8,
+  "per_device_train_batch_size": 16,
+  "per_device_eval_batch_size": 16,
   "gradient_accumulation_steps": 2,
   "logging_steps": 10,
-  "eval_steps": 200,
-  "save_steps": 2,
+  "eval_steps": 20,
+  "save_steps": 20,
   "warmup_ratio": 0.03,
   "weight_decay": 0.0,
   "max_grad_norm": 0.3,
@@ -24,7 +24,7 @@
   "gradient_checkpointing": true,
   "optim": "adamw_torch_fused",
   "ft_type": "SFT",
-  "data_name": "ALL",
+  "data_name": "Chest_xray_pa",
   "sft_type": "pure",
   "balanced": false
 }