diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index fc589c137..309ca6acb 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -65,7 +65,7 @@ To create a new release:
    - Click "Run workflow"
    - Select the version bump type:
      - `patch`: Bug fixes and minor changes (0.3.13 → 0.3.14)
-     - `minor`: New features and non-breaking changes (0.3.13 → 0.4.0)  
+     - `minor`: New features and non-breaking changes (0.3.13 → 0.4.0)
      - `major`: Breaking changes (0.3.13 → 1.0.0)
 
 3. **Edit the draft release notes**:
diff --git a/dev/new_models/prompts.json b/dev/new_models/prompts.json
index 6648f0b4b..938571de9 100644
--- a/dev/new_models/prompts.json
+++ b/dev/new_models/prompts.json
@@ -1 +1 @@
-["respond with 'yes', 'no', 'maybe'", "respond with 'maybe', 'yes', 'no'", "respond with 'no', 'yes', 'maybe'", "respond with 'yes', 'maybe', 'no'", "respond with yes or no", "respond with maybe or no", "respond with no or maybe", "respond with no or yes", "respond with yes or no", "respond with yes, no, maybe", "respond with maybe, yes, no", "respond with no, yes, maybe", "respond with yes, maybe, no", "respond with yes or no", "respond with maybe or no", "respond with no or maybe", "respond with no or yes", "respond with yes or no", "just respond with 'yes', 'no', 'maybe'", "just respond with 'maybe', 'yes', 'no'", "just respond with 'no', 'yes', 'maybe'", "just respond with 'yes', 'maybe', 'no'", "just respond with yes or no", "just respond with maybe or no", "just respond with no or maybe", "just respond with no or yes", "just respond with yes or no", "just respond with yes, no, maybe", "just respond with maybe, yes, no", "just respond with no, yes, maybe", "just respond with yes, maybe, no", "just respond with yes or no", "just respond with maybe or no", "just respond with no or maybe", "just respond with no or yes", "just respond with yes or no"]
\ No newline at end of file
+["respond with 'yes', 'no', 'maybe'", "respond with 'maybe', 'yes', 'no'", "respond with 'no', 'yes', 'maybe'", "respond with 'yes', 'maybe', 'no'", "respond with yes or no", "respond with maybe or no", "respond with no or maybe", "respond with no or yes", "respond with yes or no", "respond with yes, no, maybe", "respond with maybe, yes, no", "respond with no, yes, maybe", "respond with yes, maybe, no", "respond with yes or no", "respond with maybe or no", "respond with no or maybe", "respond with no or yes", "respond with yes or no", "just respond with 'yes', 'no', 'maybe'", "just respond with 'maybe', 'yes', 'no'", "just respond with 'no', 'yes', 'maybe'", "just respond with 'yes', 'maybe', 'no'", "just respond with yes or no", "just respond with maybe or no", "just respond with no or maybe", "just respond with no or yes", "just respond with yes or no", "just respond with yes, no, maybe", "just respond with maybe, yes, no", "just respond with no, yes, maybe", "just respond with yes, maybe, no", "just respond with yes or no", "just respond with maybe or no", "just respond with no or maybe", "just respond with no or yes", "just respond with yes or no"]
diff --git a/docs/experimental/gspo.mdx b/docs/experimental/gspo.mdx
index 511cde4fc..3b93ab49e 100644
--- a/docs/experimental/gspo.mdx
+++ b/docs/experimental/gspo.mdx
@@ -14,7 +14,7 @@ GSPO was introduced by the Qwen team to train state-of-the-art models including
 ## Key Benefits
 
 - **Stable Training**: Maintains stable training processes and resolves stability challenges in large MoE models
-- **Efficient Scaling**: Achieves higher training efficiency and continues improving with increased computational resources  
+- **Efficient Scaling**: Achieves higher training efficiency and continues improving with increased computational resources
 - **Infrastructure-Friendly**: More tolerant of precision discrepancies, eliminating the need for complex strategies like "Routing Replay"
 
 ## How It Works
diff --git a/docs/proposals/backend-first-training-api.md b/docs/proposals/backend-first-training-api.md
index a07dafd7d..938466e1a 100644
--- a/docs/proposals/backend-first-training-api.md
+++ b/docs/proposals/backend-first-training-api.md
@@ -96,21 +96,21 @@ class ServerlessBackend(Backend):
         *,
         # Core training args
         learning_rate: float = 5e-6,
-        
+
         # RL algorithm settings
         ppo: bool = False,
         epsilon: float | None = None,  # defaults based on ppo
         epsilon_high: float | None = None,
-        
+
         # Advantage computation
         advantage_balance: float = 0.0,
         scale_rewards: bool = True,
-        
+
         # Importance sampling
         importance_sampling_level: Literal["token", "sequence", "average", "geometric_average"] = "token",
         max_negative_advantage_importance_sampling_weight: float | None = None,
         mask_prob_ratio: bool = False,
-        
+
         # Experimental
         kimi_k2_tau: float | None = None,
         precalculate_logprobs: bool = False,
@@ -139,17 +139,17 @@ class LocalBackend(Backend):
         mask_prob_ratio: bool = False,
         kimi_k2_tau: float | None = None,
         precalculate_logprobs: bool = False,
-        
+
         # LocalBackend-specific
         allow_training_without_logprobs: bool = False,
         plot_tensors: bool = False,
         truncated_importance_sampling: float | None = None,
         scale_learning_rate_by_reward_std_dev: bool = False,
         logprob_calculation_chunk_size: int = 1024,
-        
+
         # Checkpoint behavior
         save_checkpoint: bool = True,
-        
+
         verbose: bool = False,
     ) -> TrainResult:
         ...
@@ -172,16 +172,16 @@ class TinkerBackend(Backend):
         advantage_balance: float = 0.0,
         scale_rewards: bool = True,
         importance_sampling_level: Literal["token", "sequence", "average", "geometric_average"] = "token",
-        
+
         # Tinker-specific checkpoint behavior
         save_checkpoint: bool = True,
         deploy_checkpoint: bool = False,  # Push to Tinker inference
-        
+
         # Tinker-specific training options
         train_mlp: bool = True,
         train_attn: bool = True,
         train_unembed: bool = False,
-        
+
         verbose: bool = False,
     ) -> TrainResult:
         ...
diff --git a/examples/codenames/codenames_words.json b/examples/codenames/codenames_words.json
index 45990b8c3..de850b8bf 100644
--- a/examples/codenames/codenames_words.json
+++ b/examples/codenames/codenames_words.json
@@ -399,4 +399,4 @@
     "witch",
     "worm",
     "yard"
-]
\ No newline at end of file
+]
diff --git a/examples/codenames/dictionary.json b/examples/codenames/dictionary.json
index 5553f3886..d20e5c8e6 100644
--- a/examples/codenames/dictionary.json
+++ b/examples/codenames/dictionary.json
@@ -466549,4 +466549,4 @@
     "zz",
     "zzt",
     "zzz"
-]
\ No newline at end of file
+]