project-llm-trainer 0.3__py3-none-any.whl → 0.3.1__py3-none-any.whl

llm_trainer/checkpoint.py

@@ -144,7 +144,7 @@ def load_checkpoint_for_eval(
 def save_steps(global_steps: int, lr_scheduler: Optional[LRScheduler] = None):
     # 暂时只保存主进程的
     if TrainerTools().parallel.is_main_process:
-        steps_checkpoint_name = f"{os.environ.get('CHECKPOINT_NAME', DEFAULT_CHECKPOINT_NAME)}.steps"
+        steps_checkpoint_name = f"{os.environ.get('LOG_DIR', './')}steps.pt"
         ckpt = {'global_steps': global_steps, 'lr_steps': lr_scheduler.cur_steps}
         torch.save(ckpt, steps_checkpoint_name)
 
@@ -153,7 +153,7 @@ def load_steps(
         default_global_steps: int = 0,
         default_lr_steps: int = 0
 ) -> Tuple[Optional[int], Optional[int]]:
-    steps_checkpoint_name = f"{os.environ.get('CHECKPOINT_NAME', DEFAULT_CHECKPOINT_NAME)}.steps"
+    steps_checkpoint_name = f"{os.environ.get('LOG_DIR', './')}steps.pt"
     if os.path.exists(steps_checkpoint_name):
         ckpt = torch.load(steps_checkpoint_name, weights_only=True)
         return ckpt['global_steps'], ckpt['lr_steps']

{project_llm_trainer-0.3.dist-info → project_llm_trainer-0.3.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: project_llm_trainer
-Version: 0.3
+Version: 0.3.1
 Summary: LLM and VLM trainer
 Author: qibin
 Author-email: qibin0506@gmail.com

{project_llm_trainer-0.3.dist-info → project_llm_trainer-0.3.1.dist-info}/RECORD

@@ -1,5 +1,5 @@
 llm_trainer/__init__.py,sha256=HWgtTEVeQSnZmEyYQm2K6eFEG4X2QAoigMlB5Z2tcXE,260
-llm_trainer/checkpoint.py,sha256=iTbnmVrT0Ql4DpD178UI95zCmfBUdYtoJS5wIvf8_4k,6099
+llm_trainer/checkpoint.py,sha256=Dlkcit0o7Gx6S9QUrIrVp2pTurP9X0zVA7w7ImSuVQU,6049
 llm_trainer/dataset.py,sha256=uz1TTd87ikf7CZPdGxmR95TSQTFWPPTilgWLBWO46_I,3916
 llm_trainer/dcp.py,sha256=PkD97DyrOtoTKn4FJsfL3VqAy4dxufgjdzJEz8-Cnoc,3635
 llm_trainer/dpo_trainer.py,sha256=6rm8Jq0rI0xazcl_bCOun8rnd34Tb_PKgezowhwoiCM,13150
@@ -21,14 +21,14 @@ llm_trainer/tools.py,sha256=AhfjN9oln5Pyif1SgCWwgQg-Q5acTCd9xpz4L26QUjA,3039
 llm_trainer/train_configs.py,sha256=FAlylSYVeh_oJGTy2fcMNUV8JLD6B70hMuk-iKx14iI,15748
 llm_trainer/trainer.py,sha256=mq51d-2ADUpcWCArszhYnOSTveatt3_x43hcC7IZgYk,24330
 llm_trainer/utils.py,sha256=04XiMENVotNgbNRBn9wadHu-cJHPxj0Xq-zzLJmNgZQ,8062
-project_llm_trainer-0.3.data/scripts/calc_intermediate_size,sha256=AggpgNHokJiJMbEtVdOnolqr_4bH3i1UYuZNEAzC2Gc,460
-project_llm_trainer-0.3.data/scripts/ddp_train,sha256=x81AasaN2-9TwARFFF1l7iV1LmfMQ0bLw0i_CGbOwSw,299
-project_llm_trainer-0.3.data/scripts/ds_train,sha256=qL3qc3TcedBCw98UZUjW07ONcErRawLE1HymW2AmscA,265
-project_llm_trainer-0.3.data/scripts/plot_loss,sha256=MzFcdJESlVr1srj4Td6-AxPGUKkfB_QEcJwm0Bd-5fU,910
-project_llm_trainer-0.3.data/scripts/plot_lr,sha256=w_7XR_x3KYYyboeOVAeu_I4fveLFI-C0wBmRrNlmWUI,894
-project_llm_trainer-0.3.data/scripts/py_train,sha256=tOp9TquORQeU8XN5H7OVIk5O0Ypwi34p_GENxTwgwdk,265
-project_llm_trainer-0.3.data/scripts/smart_train,sha256=Pmt4Q0to4Hoz82iB9uFPZuz7uahNUbfE7FR1940EBy8,716
-project_llm_trainer-0.3.dist-info/METADATA,sha256=P64NiFbJzSd4QkFJ5udQ4qMyHUorPp3ex4F3eIdtVdU,193
-project_llm_trainer-0.3.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
-project_llm_trainer-0.3.dist-info/top_level.txt,sha256=LtRFg28i0QIG7iBCD2t095oSco99LCtkijibS9cMGik,12
-project_llm_trainer-0.3.dist-info/RECORD,,
+project_llm_trainer-0.3.1.data/scripts/calc_intermediate_size,sha256=AggpgNHokJiJMbEtVdOnolqr_4bH3i1UYuZNEAzC2Gc,460
+project_llm_trainer-0.3.1.data/scripts/ddp_train,sha256=x81AasaN2-9TwARFFF1l7iV1LmfMQ0bLw0i_CGbOwSw,299
+project_llm_trainer-0.3.1.data/scripts/ds_train,sha256=qL3qc3TcedBCw98UZUjW07ONcErRawLE1HymW2AmscA,265
+project_llm_trainer-0.3.1.data/scripts/plot_loss,sha256=MzFcdJESlVr1srj4Td6-AxPGUKkfB_QEcJwm0Bd-5fU,910
+project_llm_trainer-0.3.1.data/scripts/plot_lr,sha256=w_7XR_x3KYYyboeOVAeu_I4fveLFI-C0wBmRrNlmWUI,894
+project_llm_trainer-0.3.1.data/scripts/py_train,sha256=tOp9TquORQeU8XN5H7OVIk5O0Ypwi34p_GENxTwgwdk,265
+project_llm_trainer-0.3.1.data/scripts/smart_train,sha256=Pmt4Q0to4Hoz82iB9uFPZuz7uahNUbfE7FR1940EBy8,716
+project_llm_trainer-0.3.1.dist-info/METADATA,sha256=LJl2lNqTIIQZpTt7iVqzQJ2NhAvTUOwS9w44_XxIn0Y,195
+project_llm_trainer-0.3.1.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+project_llm_trainer-0.3.1.dist-info/top_level.txt,sha256=LtRFg28i0QIG7iBCD2t095oSco99LCtkijibS9cMGik,12
+project_llm_trainer-0.3.1.dist-info/RECORD,,