update requires

Former-commit-id: cae0e688ddcead370821e126c192bddc53ff6017

src/llamafactory/__init__.py

@@ -20,17 +20,17 @@ Level:
 
 Dependency graph:
   main:
-    transformers>=4.41.2,<=4.45.2
+    transformers>=4.41.2,<=4.46.0
     datasets>=2.16.0,<=2.21.0
-    accelerate>=0.30.1,<=0.34.2
+    accelerate>=0.34.0,<=1.0.1
     peft>=0.11.1,<=0.12.0
     trl>=0.8.6,<=0.9.6
   attention:
     transformers>=4.42.4 (gemma+fa2)
   longlora:
-    transformers>=4.41.2,<=4.45.2
+    transformers>=4.41.2,<=4.46.0
   packing:
-    transformers>=4.41.2,<=4.45.2
+    transformers>=4.41.2,<=4.46.0
 
 Disable version checking: DISABLE_VERSION_CHECK=1
 Enable VRAM recording: RECORD_VRAM=1

src/llamafactory/extras/misc.py

@@ -79,9 +79,9 @@ def check_dependencies() -> None:
     if os.environ.get("DISABLE_VERSION_CHECK", "0").lower() in ["true", "1"]:
         logger.warning("Version checking has been disabled, may lead to unexpected behaviors.")
     else:
-        require_version("transformers>=4.41.2,<=4.45.2", "To fix: pip install transformers>=4.41.2,<=4.45.2")
+        require_version("transformers>=4.41.2,<=4.46.0", "To fix: pip install transformers>=4.41.2,<=4.46.0")
         require_version("datasets>=2.16.0,<=2.21.0", "To fix: pip install datasets>=2.16.0,<=2.21.0")
-        require_version("accelerate>=0.30.1,<=0.34.2", "To fix: pip install accelerate>=0.30.1,<=0.34.2")
+        require_version("accelerate>=0.34.0,<=1.0.1", "To fix: pip install accelerate>=0.34.0,<=1.0.1")
         require_version("peft>=0.11.1,<=0.12.0", "To fix: pip install peft>=0.11.1,<=0.12.0")
         require_version("trl>=0.8.6,<=0.9.6", "To fix: pip install trl>=0.8.6,<=0.9.6")
 

src/llamafactory/model/model_utils/longlora.py

@@ -86,7 +86,7 @@ def llama_attention_forward(
 
     if getattr(self.config, "group_size_ratio", None) and self.training:  # shift
         groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
-        assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
+        assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
         num_groups = q_len // groupsz
 
         def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -195,7 +195,7 @@ def llama_flash_attention_2_forward(
 
     if getattr(self.config, "group_size_ratio", None) and self.training:  # shift
         groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
-        assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
+        assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
         num_groups = q_len // groupsz
 
         def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -301,7 +301,7 @@ def llama_sdpa_attention_forward(
 
     if getattr(self.config, "group_size_ratio", None) and self.training:  # shift
         groupsz = int(q_len * getattr(self.config, "group_size_ratio"))
-        assert q_len % groupsz == 0, "q_len {} should be divisible by group size {}.".format(q_len, groupsz)
+        assert q_len % groupsz == 0, f"q_len {q_len} should be divisible by group size {groupsz}."
         num_groups = q_len // groupsz
 
         def shift(state: "torch.Tensor") -> "torch.Tensor":
@@ -353,7 +353,7 @@ def llama_sdpa_attention_forward(
 
 
 def _apply_llama_patch() -> None:
-    require_version("transformers>=4.41.2,<=4.45.2", "To fix: pip install transformers>=4.41.2,<=4.45.2")
+    require_version("transformers>=4.41.2,<=4.46.0", "To fix: pip install transformers>=4.41.2,<=4.46.0")
     LlamaAttention.forward = llama_attention_forward
     LlamaFlashAttention2.forward = llama_flash_attention_2_forward
     LlamaSdpaAttention.forward = llama_sdpa_attention_forward

src/llamafactory/model/model_utils/packing.py

@@ -114,7 +114,7 @@ def get_unpad_data(attention_mask: "torch.Tensor") -> Tuple["torch.Tensor", "tor
 
 
 def _patch_for_block_diag_attn(model_type: str) -> None:
-    require_version("transformers>=4.41.2,<=4.45.2", "To fix: pip install transformers>=4.41.2,<=4.45.2")
+    require_version("transformers>=4.41.2,<=4.46.0", "To fix: pip install transformers>=4.41.2,<=4.46.0")
     if is_transformers_version_greater_than_4_43():
         import transformers.modeling_flash_attention_utils