Upload qwen2_5_vl_full_sft.yaml
Browse files
sft_related/qwen2_5_vl_full_sft.yaml
CHANGED
|
@@ -1,7 +1,7 @@
|
|
| 1 |
### model
|
| 2 |
model_name_or_path: Qwen/Qwen2.5-VL-3B-Instruct
|
| 3 |
-
|
| 4 |
-
|
| 5 |
trust_remote_code: true
|
| 6 |
|
| 7 |
### method
|
|
@@ -22,7 +22,7 @@ overwrite_cache: true
|
|
| 22 |
preprocessing_num_workers: 16
|
| 23 |
|
| 24 |
### output
|
| 25 |
-
output_dir: saves/qwen2_5_vl-3b/full/sft
|
| 26 |
logging_steps: 10
|
| 27 |
save_steps: 100
|
| 28 |
plot_loss: true
|
|
@@ -36,10 +36,4 @@ num_train_epochs: 3.0
|
|
| 36 |
lr_scheduler_type: cosine
|
| 37 |
warmup_ratio: 0.1
|
| 38 |
bf16: true
|
| 39 |
-
ddp_timeout: 180000000
|
| 40 |
-
|
| 41 |
-
### eval
|
| 42 |
-
# val_size: 0.1
|
| 43 |
-
# per_device_eval_batch_size: 1
|
| 44 |
-
# eval_strategy: steps
|
| 45 |
-
# eval_steps: 500
|
|
|
|
| 1 |
### model
|
| 2 |
model_name_or_path: Qwen/Qwen2.5-VL-3B-Instruct
|
| 3 |
+
image_max_pixels: 262144
|
| 4 |
+
video_max_pixels: 16384
|
| 5 |
trust_remote_code: true
|
| 6 |
|
| 7 |
### method
|
|
|
|
| 22 |
preprocessing_num_workers: 16
|
| 23 |
|
| 24 |
### output
|
| 25 |
+
output_dir: saves/qwen2_5_vl-3b/full/sft-noPixel
|
| 26 |
logging_steps: 10
|
| 27 |
save_steps: 100
|
| 28 |
plot_loss: true
|
|
|
|
| 36 |
lr_scheduler_type: cosine
|
| 37 |
warmup_ratio: 0.1
|
| 38 |
bf16: true
|
| 39 |
+
ddp_timeout: 180000000
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|