Revert "checkpoint model on first step callback (#2906)"

This reverts commit 10ba1622f7.
This commit is contained in:
Dan Saunders
2025-07-15 15:01:12 -04:00
committed by GitHub
parent 10ba1622f7
commit 6f6d917a99
146 changed files with 9 additions and 419 deletions

View File

@@ -69,7 +69,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"use_tensorboard": True,
"bf16": True,
"save_first_step": False,
}
)
@@ -136,7 +135,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"use_tensorboard": True,
"bf16": True,
"save_first_step": False,
}
)
@@ -212,7 +210,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"use_tensorboard": True,
"bf16": True,
"save_first_step": False,
}
)
@@ -292,7 +289,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"use_tensorboard": True,
"bf16": True,
"save_first_step": False,
}
)
@@ -369,7 +365,6 @@ class TestMultiGPULlama:
},
"use_tensorboard": True,
"seed": 42,
"save_first_step": False,
}
)
@@ -447,7 +442,6 @@ class TestMultiGPULlama:
"fsdp_auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
},
"use_tensorboard": True,
"save_first_step": False,
}
)
@@ -526,7 +520,6 @@ class TestMultiGPULlama:
"fsdp_reshard_after_forward": fsdp_reshard_after_forward,
},
"use_tensorboard": True,
"save_first_step": False,
}
)
if attention_backend == "flash":
@@ -612,7 +605,6 @@ class TestMultiGPULlama:
"fsdp_auto_wrap_policy": "TRANSFORMER_BASED_WRAP",
},
"use_tensorboard": True,
"save_first_step": False,
}
)
@@ -697,7 +689,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"deepspeed": str(AXOLOTL_ROOT / deepspeed),
"use_tensorboard": True,
"save_first_step": False,
**adapter,
}
)
@@ -774,7 +765,6 @@ class TestMultiGPULlama:
"deepspeed": str(AXOLOTL_ROOT / "deepspeed_configs/zero2.json"),
"use_tensorboard": True,
"seed": 42,
"save_first_step": False,
**adapter,
}
)
@@ -850,7 +840,6 @@ class TestMultiGPULlama:
"flash_attention": True,
"deepspeed": str(AXOLOTL_ROOT / "deepspeed_configs/zero1.json"),
"use_tensorboard": True,
"save_first_step": False,
**adapter,
}
)
@@ -919,7 +908,6 @@ class TestMultiGPULlama:
"save_safetensors": True,
# "deepspeed": str(AXOLOTL_ROOT / "deepspeed_configs/zero1.json"),
"use_tensorboard": True,
"save_first_step": False,
}
)