Minor update to docs

finetuning/specialists/resource-efficient/README.md

@@ -33,7 +33,7 @@ Fixed parameters:
 - training and validation batch size - `1`
 - minimum number of training "samples" for training on the provided images - min. **`50`** (oversample while min. 50 training samples not found) (this is done to avoid the exhaustive time constraints while training with only 1 training sample)
 - learning rate: `1e-5`
-- optimizer: `Adam`
+- optimizer: `AdamW`
 - lr scheduler: `ReduceLRonPlateau`
 - early stopping: `10`
 - patch shape: `(512, 512)`

finetuning/specialists/resource-efficient/run_resource_efficient_finetuning.py

@@ -15,6 +15,7 @@ def base_slurm_script(env_name, partition, cpu_mem, cpu_cores, gpu_name=None):
 #SBATCH --mem {cpu_mem}
 #SBATCH -p {partition}
 #SBATCH -t 2-00:00:00
+#SBATCH --job-name micro-sam-resource-efficient-finetuning
 """
     if gpu_name is not None:
         base_script += f"#SBATCH -G {gpu_name}:1 \n"
@@ -29,7 +30,7 @@ def base_slurm_script(env_name, partition, cpu_mem, cpu_cores, gpu_name=None):
 
 def write_batch_sript(
     env_name, partition, cpu_mem, cpu_cores, gpu_name, input_path, save_root,
-    model_type, n_objects, n_images, script_name, freeze, lora,
+    model_type, n_objects, n_images, script_name, freeze, lora, dry,
 ):
     assert model_type in ["vit_t", "vit_b", "vit_t_lm", "vit_b_lm"]
 
@@ -52,7 +53,7 @@ def write_batch_sript(
 
     # Whether to use LoRA-based finetuning
     # NOTE: We use rank as 4 for LoRA.
-    if lora is not None:
+    if lora:
         python_script += "--lora_rank 4 "
 
     if gpu_name is not None:
@@ -82,8 +83,9 @@ def write_batch_sript(
     with open(script_name, "w") as f:
         f.write(batch_script)
 
-    cmd = ["sbatch", script_name]
-    subprocess.run(cmd)
+    if not dry:
+        cmd = ["sbatch", script_name]
+        subprocess.run(cmd)
 
 
 def get_batch_script_names(tmp_folder):
@@ -102,6 +104,7 @@ def main(args):
 
     all_n_images = [1, 2, 5, 10]
     use_lora = [False, True]
+
     for (n_images, lora) in itertools.product(all_n_images, use_lora):
         # We cannot use LoRA and freeze the image encoder at the same time.
         if lora and args.freeze == "image_encoder":
@@ -121,6 +124,7 @@ def main(args):
             script_name=get_batch_script_names(tmp_folder),
             freeze=args.freeze,
             lora=lora,
+            dry=args.dry,
         )
 
 
@@ -141,7 +145,9 @@ def main(args):
     parser.add_argument("--partition", type=str, required=True, help="Name of the partition for running the job.")
     parser.add_argument("--mem", type=str, required=True, help="Amount of cpu memory.")
     parser.add_argument("-c", "--cpu_cores", type=int, required=True, help="Number of cpu cores.")
-    parser.add_argument("-G", "--gpu_name", type=str, default=None, help="The GPI resources used for finetuning.")
+    parser.add_argument("-G", "--gpu_name", type=str, default=None, help="The GPU resources used for finetuning.")
+
+    parser.add_argument("--dry", action="store_true", help="Whether to avoid submitting the configured scripts.")
 
     args = parser.parse_args()
     main(args)