llmflowstack 1.2.4__tar.gz → 1.2.5__tar.gz

{llmflowstack-1.2.4 → llmflowstack-1.2.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: llmflowstack
-Version: 1.2.4
+Version: 1.2.5
 Summary: LLMFlowStack is a framework for training and using LLMs (LLaMA, GPT-OSS, Gemma, ...). Supports DAPT, fine-tuning, and distributed inference. Public fork without institution-specific components.
 Author-email: Gustavo Henrique Ferreira Cruz <gustavohferreiracruz@gmail.com>
 License: MIT

{llmflowstack-1.2.4 → llmflowstack-1.2.5}/llmflowstack/decoders/BaseDecoder.py

@@ -292,6 +292,7 @@ class BaseDecoder(ABC):
 			warmup_ratio=params.warmup_ratio,
 			lr_scheduler_type="cosine_with_min_lr",
 			lr_scheduler_kwargs={"min_lr_rate": 0.1},
+			label_smoothing_factor=params.label_smoothing_factor,
 			output_dir=None,
 			save_strategy="no",
 			logging_steps=params.logging_steps
@@ -437,6 +438,7 @@ class BaseDecoder(ABC):
 			warmup_ratio=params.warmup_ratio,
 			lr_scheduler_type="cosine_with_min_lr",
 			lr_scheduler_kwargs={"min_lr_rate": 0.1},
+			label_smoothing_factor=params.label_smoothing_factor,
 			output_dir=None,
 			save_strategy="no",
 			logging_steps=params.logging_steps

{llmflowstack-1.2.4 → llmflowstack-1.2.5}/llmflowstack/schemas/params.py

@@ -17,6 +17,7 @@ class TrainParams:
     "sgd"
   ] = "adamw_torch"
   logging_steps: int = 1
+  label_smoothing_factor: float = 0
 
 @dataclass
 class GenerationBeamsParams:

{llmflowstack-1.2.4 → llmflowstack-1.2.5}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "llmflowstack"
-version = "1.2.4"
+version = "1.2.5"
 authors = [
   { name = "Gustavo Henrique Ferreira Cruz", email = "gustavohferreiracruz@gmail.com" }
 ]