Julian Bilcke
commited on
Commit
·
abc2df1
1
Parent(s):
19db9a3
fix typo
Browse files
vms/ui/project/services/training.py
CHANGED
|
@@ -1146,15 +1146,12 @@ class TrainingService:
|
|
| 1146 |
auto_recover = True # Always auto-recover on startup
|
| 1147 |
|
| 1148 |
if auto_recover:
|
| 1149 |
-
# Rest of the auto-recovery code remains unchanged
|
| 1150 |
try:
|
| 1151 |
-
# Use the internal model_type for the actual training
|
| 1152 |
-
# But keep model_type_display for the UI
|
| 1153 |
result = self.start_training(
|
| 1154 |
model_type=model_type_internal,
|
| 1155 |
lora_rank=params.get('lora_rank', DEFAULT_LORA_RANK_STR),
|
| 1156 |
lora_alpha=params.get('lora_alpha', DEFAULT_LORA_ALPHA_STR),
|
| 1157 |
-
|
| 1158 |
batch_size=params.get('batch_size', DEFAULT_BATCH_SIZE),
|
| 1159 |
learning_rate=params.get('learning_rate', DEFAULT_LEARNING_RATE),
|
| 1160 |
save_iterations=params.get('save_iterations', DEFAULT_SAVE_CHECKPOINT_EVERY_N_STEPS),
|
|
|
|
| 1146 |
auto_recover = True # Always auto-recover on startup
|
| 1147 |
|
| 1148 |
if auto_recover:
|
|
|
|
| 1149 |
try:
|
|
|
|
|
|
|
| 1150 |
result = self.start_training(
|
| 1151 |
model_type=model_type_internal,
|
| 1152 |
lora_rank=params.get('lora_rank', DEFAULT_LORA_RANK_STR),
|
| 1153 |
lora_alpha=params.get('lora_alpha', DEFAULT_LORA_ALPHA_STR),
|
| 1154 |
+
train_steps=params.get('train_steps', DEFAULT_NB_TRAINING_STEPS),
|
| 1155 |
batch_size=params.get('batch_size', DEFAULT_BATCH_SIZE),
|
| 1156 |
learning_rate=params.get('learning_rate', DEFAULT_LEARNING_RATE),
|
| 1157 |
save_iterations=params.get('save_iterations', DEFAULT_SAVE_CHECKPOINT_EVERY_N_STEPS),
|