PyPI - project-llm-trainer - Versions diffs - 0.5.11__py3-none-any.whl → 0.5.13__py3-none-any.whl - Mend

project-llm-trainer 0.5.11py3-none-any.whl → 0.5.13py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.

This version of project-llm-trainer might be problematic. Click here for more details.

Files changed (16) hide show

llm_trainer/dpo_trainer.py CHANGED Viewed

@@ -271,7 +271,5 @@ class DPOTrainer(Trainer):
                 TrainerTools().parallel.on_epoch_end(epoch)
                 self._on_epoch_end(tag=f'epoch:{epoch}')
-        # 等待checkpoint保存完成
-        time.sleep(10)
         TrainerTools().parallel.destroy()

llm_trainer/ds_checkpoint.py CHANGED Viewed

@@ -2,6 +2,7 @@ import os
 from glob import glob
 import shutil
 from torch import nn
+from .tools import TrainerTools
 try:
     import deepspeed
@@ -23,17 +24,20 @@ def save_ds_checkpoint(model: nn.Module):
     try:
         # 包括model、optimizer等状态
         model.save_checkpoint(save_dir=ckpt_dir)
-    except:
-        return
-    # 删除历史checkpoint
-    ckpt_paths = glob(os.path.join(ckpt_dir, "global_*"))
-    if len(ckpt_paths) > 2:
-        # 按修改时间排序，找到最旧的目录
-        oldest_ckpt = sorted(ckpt_paths, key=os.path.getmtime)[0]
-        try:
-            shutil.rmtree(oldest_ckpt)
-        except: ...
+    except: ...
+    TrainerTools().parallel.wait()
+    # 只在main rank上执行
+    if TrainerTools().parallel.is_main_process:
+        # 删除历史checkpoint
+        ckpt_paths = glob(os.path.join(ckpt_dir, "global_*"))
+        if len(ckpt_paths) > 2:
+            # 按修改时间排序，找到最旧的目录
+            oldest_ckpt = sorted(ckpt_paths, key=os.path.getmtime)[0]
+            try:
+                shutil.rmtree(oldest_ckpt)
+            except: ...
 def load_ds_checkpoint(

llm_trainer/grpo_trainer.py CHANGED Viewed

@@ -389,6 +389,4 @@ class GRPOTrainer(Trainer):
                 TrainerTools().parallel.on_epoch_end(epoch)
                 self._on_epoch_end(tag=f'epoch:{epoch}')
-        # 等待checkpoint保存完成
-        time.sleep(10)
         TrainerTools().parallel.destroy()

llm_trainer/train_configs.py CHANGED Viewed

@@ -136,8 +136,8 @@ class DPOConfig:
 @dataclass(kw_only=True)
 class GRPOConfig:
     grpo_steps: int = 1
-    clip_eps: float = 0.2
-    kl_weight: float = 0.01
+    clip_eps: float = 0.1
+    kl_weight: float = 0.04
     group_size: int = 12
     mixup_alpha: float = 1.0
     gen_max_new_tokens: Optional[int] = None

llm_trainer/trainer.py CHANGED Viewed

@@ -597,6 +597,4 @@ class Trainer:
                 TrainerTools().parallel.on_epoch_end(epoch)
                 self._on_epoch_end(tag=f'epoch:{epoch}')
-        # 等待checkpoint保存完成
-        time.sleep(10)
         TrainerTools().parallel.destroy()

{project_llm_trainer-0.5.11.dist-info → project_llm_trainer-0.5.13.dist-info}/METADATA RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: project_llm_trainer
-Version: 0.5.11
+Version: 0.5.13
 Summary: LLM and VLM trainer
 Author: qibin
 Author-email: qibin0506@gmail.com

{project_llm_trainer-0.5.11.dist-info → project_llm_trainer-0.5.13.dist-info}/RECORD RENAMED Viewed

@@ -1,11 +1,11 @@
 llm_trainer/__init__.py,sha256=HWgtTEVeQSnZmEyYQm2K6eFEG4X2QAoigMlB5Z2tcXE,260
 llm_trainer/checkpoint.py,sha256=Wh5CwceIajTgJ9i_mH3I1R9N2nOLFqVFmlEMkTiGcD4,4306
 llm_trainer/dataset.py,sha256=4QlOo0SFB5816BUYegQjgobUqTUMQvdmZMM_OEAMSjE,4347
-llm_trainer/dpo_trainer.py,sha256=1A_4QP2_xqM_YeqdXy-0RaMvEL80gim-pgnPQyHww9U,12052
-llm_trainer/ds_checkpoint.py,sha256=D092fkS1Up4QmpV9YCpqbSzfX_caCAeX-UiOrhOE1I8,1947
+llm_trainer/dpo_trainer.py,sha256=8LYRxviJKcB-rN_XprVsWr5YshU8KolggMm7irjbXvI,11990
+llm_trainer/ds_checkpoint.py,sha256=-QjtY8JPJMa1IGjMHLJQKrbWRwVqDQSa7OXOKYamwDo,2115
 llm_trainer/eval.py,sha256=ZyUfSo2Q8P-lrCdPEnGkoo5pGubd0AabREK5eMISRII,1109
 llm_trainer/generate_utils.py,sha256=wrZoG2g7CsOyG4sb3px9vURHQFV6_9j5kQmpFc5A8yg,15335
-llm_trainer/grpo_trainer.py,sha256=sCYjvksdm9f7TpN23KXuCmua_8VFTZEfVEcflL89P_I,16058
+llm_trainer/grpo_trainer.py,sha256=PVTlKOEJpI0AMlh7Siw_MHpLm9CAZepCAMjjSZF6eRU,15996
 llm_trainer/log.py,sha256=LxqTGRNZUGMTSQCePRpk-rYyxSnSIbT4kOdP8Fbzr0M,462
 llm_trainer/loss.py,sha256=NZCQeUXnLSj__mmDflE8g89KgE0emAJXIab0IERCLno,6023
 llm_trainer/parallel.py,sha256=DQu8GqEFxD99HQ6hKuIxxyKi-05dMO33eMhImYlPuOI,4468
@@ -17,17 +17,17 @@ llm_trainer/scheduler.py,sha256=LAI_0VxClsIQkix0bRoduRD4vPfVuIZDhZgTAT_KK8k,4901
 llm_trainer/sft_trainer.py,sha256=gxQA7T1o1QGUsHp2CX1Qb_fO5LppBJuNbc0H4ixCYUA,1783
 llm_trainer/tokenizer.py,sha256=SSpgXtb0e1NtQqRW0gCq09TTZi47umggy-Fh5EMHKJg,6708
 llm_trainer/tools.py,sha256=yF17lp6oOfLe2XJeKDQ1juZcbv-6vFamJSLwEeArduA,2975
-llm_trainer/train_configs.py,sha256=guV8xkG5TSGvYwFvsQV_mA8mDHLLVhL5L0xo_WMsMME,7347
-llm_trainer/trainer.py,sha256=U26dZc22nByfTZUzKeEiqqYVexBzgw0ep7N0Z2zIcWI,26141
+llm_trainer/train_configs.py,sha256=992wy0YhBG2WvxwdLEPL4_-JUl4NkwMPT-jj_BIHo6A,7347
+llm_trainer/trainer.py,sha256=FF75J-BRUp34No2TvQIgomvNozWYzVhDeOfaBgQLV9g,26079
 llm_trainer/utils.py,sha256=LWNhyQ0NDEZ9mZtk2Ryvh6EulvHIaUGIflugSpqmeFI,6791
-project_llm_trainer-0.5.11.data/scripts/calc_intermediate_size,sha256=AggpgNHokJiJMbEtVdOnolqr_4bH3i1UYuZNEAzC2Gc,460
-project_llm_trainer-0.5.11.data/scripts/ddp_train,sha256=x81AasaN2-9TwARFFF1l7iV1LmfMQ0bLw0i_CGbOwSw,299
-project_llm_trainer-0.5.11.data/scripts/ds_train,sha256=qL3qc3TcedBCw98UZUjW07ONcErRawLE1HymW2AmscA,265
-project_llm_trainer-0.5.11.data/scripts/plot_loss,sha256=MzFcdJESlVr1srj4Td6-AxPGUKkfB_QEcJwm0Bd-5fU,910
-project_llm_trainer-0.5.11.data/scripts/plot_lr,sha256=w_7XR_x3KYYyboeOVAeu_I4fveLFI-C0wBmRrNlmWUI,894
-project_llm_trainer-0.5.11.data/scripts/py_train,sha256=tOp9TquORQeU8XN5H7OVIk5O0Ypwi34p_GENxTwgwdk,265
-project_llm_trainer-0.5.11.data/scripts/smart_train,sha256=Pmt4Q0to4Hoz82iB9uFPZuz7uahNUbfE7FR1940EBy8,716
-project_llm_trainer-0.5.11.dist-info/METADATA,sha256=RAhT8VLTlO4Oyr9ocDxBvhulYed1JXWX_5GWWkiJ7go,196
-project_llm_trainer-0.5.11.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
-project_llm_trainer-0.5.11.dist-info/top_level.txt,sha256=LtRFg28i0QIG7iBCD2t095oSco99LCtkijibS9cMGik,12
-project_llm_trainer-0.5.11.dist-info/RECORD,,
+project_llm_trainer-0.5.13.data/scripts/calc_intermediate_size,sha256=AggpgNHokJiJMbEtVdOnolqr_4bH3i1UYuZNEAzC2Gc,460
+project_llm_trainer-0.5.13.data/scripts/ddp_train,sha256=x81AasaN2-9TwARFFF1l7iV1LmfMQ0bLw0i_CGbOwSw,299
+project_llm_trainer-0.5.13.data/scripts/ds_train,sha256=qL3qc3TcedBCw98UZUjW07ONcErRawLE1HymW2AmscA,265
+project_llm_trainer-0.5.13.data/scripts/plot_loss,sha256=MzFcdJESlVr1srj4Td6-AxPGUKkfB_QEcJwm0Bd-5fU,910
+project_llm_trainer-0.5.13.data/scripts/plot_lr,sha256=w_7XR_x3KYYyboeOVAeu_I4fveLFI-C0wBmRrNlmWUI,894
+project_llm_trainer-0.5.13.data/scripts/py_train,sha256=tOp9TquORQeU8XN5H7OVIk5O0Ypwi34p_GENxTwgwdk,265
+project_llm_trainer-0.5.13.data/scripts/smart_train,sha256=Pmt4Q0to4Hoz82iB9uFPZuz7uahNUbfE7FR1940EBy8,716
+project_llm_trainer-0.5.13.dist-info/METADATA,sha256=ROfaBOhsQK5yC6HMBa_4Tblg90TVadyKbjcNU-s3Imk,196
+project_llm_trainer-0.5.13.dist-info/WHEEL,sha256=Nw36Djuh_5VDukK0H78QzOX-_FQEo6V37m3nkm96gtU,91
+project_llm_trainer-0.5.13.dist-info/top_level.txt,sha256=LtRFg28i0QIG7iBCD2t095oSco99LCtkijibS9cMGik,12
+project_llm_trainer-0.5.13.dist-info/RECORD,,