Training in progress, step 2000

.ipynb_checkpoints/upload-checkpoint.py

@@ -23,7 +23,7 @@ TRAINER_STATE_FILENAME = "trainer_state.json"
 LOSS_PLOT_FILENAME = "loss.png"
 
 # Plotting Configuration
-LOSS_SMOOTHING_WINDOW = 30
+LOSS_SMOOTHING_WINDOW = 40
 
 # Monitoring Configuration
 CHECKPOINT_DIR_PATTERN = re.compile(r"^checkpoint-(\d+)$")
@@ -120,10 +120,15 @@ library_name: peft
     """.strip()
     return readme_template
 
+import json
+import numpy as np
+import matplotlib.pyplot as plt
+from pathlib import Path
+
 def plot_loss_from_json(
     json_file_path: Path,
     output_image_path: Path,
-    smooth_steps: int = LOSS_SMOOTHING_WINDOW
+    smooth_steps: int = 10  # Example default value for smooth_steps
 ):
     """
     Reads training log data from a JSON file (trainer_state.json),
@@ -152,7 +157,7 @@ def plot_loss_from_json(
         print(f"An unexpected error occurred while reading {json_file_path}: {e}")
         return
 
-    log_history = data.get("log_history") # Use .get for safer access
+    log_history = data.get("log_history")  # Use .get for safer access
     if not isinstance(log_history, list):
         print(f"Error: 'log_history' key not found or not a list in {json_file_path}")
         return
@@ -184,23 +189,23 @@ def plot_loss_from_json(
     # Calculate Running Average
     smoothed_losses = None
     smoothed_steps = None
-    apply_smoothing = smooth_steps > 0 and len(losses) >= smooth_steps
+    apply_smoothing = smooth_steps > 0
 
     if apply_smoothing:
         try:
             weights = np.ones(smooth_steps) / smooth_steps
-            smoothed_losses = np.convolve(losses, weights, mode='valid')
-            smoothed_steps = steps[smooth_steps - 1:] # Steps corresponding to the smoothed values
+            smoothed_losses = np.convolve(losses, weights, mode='full')[:len(losses)]
+            smoothed_steps = steps  # Steps corresponding to the smoothed values
             print(f"Calculated smoothed loss over {len(smoothed_steps)} points.")
         except Exception as e:
             print(f"Warning: Could not calculate smoothed loss. Error: {e}")
-            apply_smoothing = False # Disable if calculation fails
+            apply_smoothing = False  # Disable if calculation fails
     elif smooth_steps > 0:
         print(f"Warning: Not enough data points ({len(losses)}) for smoothing window ({smooth_steps}). Skipping smoothing.")
 
     # Plotting
-    plt.style.use('seaborn-v0_8-darkgrid') # Use a nice style
-    plt.figure(figsize=(10, 6)) # Standard figure size
+    plt.style.use('seaborn-v0_8-darkgrid')  # Use a nice style
+    plt.figure(figsize=(10, 6))  # Standard figure size
 
     plt.plot(steps, losses, linestyle='-', color='skyblue', alpha=0.5, label='Original Loss')
 
@@ -212,7 +217,7 @@ def plot_loss_from_json(
     plt.ylabel("Loss")
     plt.title("Training Loss Progression")
     plt.legend()
-    plt.tight_layout() # Adjust layout
+    plt.tight_layout()  # Adjust layout
 
     # Saving
     try:
@@ -221,7 +226,7 @@ def plot_loss_from_json(
     except Exception as e:
         print(f"Error saving plot to {output_image_path}: {e}")
     finally:
-        plt.close() # Ensure figure is closed to free memory
+        plt.close()  # Ensure figure is closed to free memory
 
 def prepare_checkpoint_folder(checkpoint_path: Path, checkpoint_number: int):
     """

adapter_config.json

@@ -1,7 +1,7 @@
 {
   "alpha_pattern": {},
   "auto_mapping": null,
-  "base_model_name_or_path": "Qwen/Qwen2.5-7B-Instruct",
+  "base_model_name_or_path": "./Qwen-2.5-7B-Instruct",
   "bias": "none",
   "eva_config": null,
   "exclude_modules": null,

adapter_model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fcedc657eb350f98486269232f8cd45740cc1f77f201c5bced88439de5d4e75f
+oid sha256:c7b7c7e22d18bb221ca73885bb0c6ded7c4cad69581153ddbe61260268f902cb
 size 161533192

upload.py

@@ -23,7 +23,7 @@ TRAINER_STATE_FILENAME = "trainer_state.json"
 LOSS_PLOT_FILENAME = "loss.png"
 
 # Plotting Configuration
-LOSS_SMOOTHING_WINDOW = 30
+LOSS_SMOOTHING_WINDOW = 40
 
 # Monitoring Configuration
 CHECKPOINT_DIR_PATTERN = re.compile(r"^checkpoint-(\d+)$")
@@ -120,10 +120,15 @@ library_name: peft
     """.strip()
     return readme_template
 
+import json
+import numpy as np
+import matplotlib.pyplot as plt
+from pathlib import Path
+
 def plot_loss_from_json(
     json_file_path: Path,
     output_image_path: Path,
-    smooth_steps: int = LOSS_SMOOTHING_WINDOW
+    smooth_steps: int = 10  # Example default value for smooth_steps
 ):
     """
     Reads training log data from a JSON file (trainer_state.json),
@@ -152,7 +157,7 @@ def plot_loss_from_json(
         print(f"An unexpected error occurred while reading {json_file_path}: {e}")
         return
 
-    log_history = data.get("log_history") # Use .get for safer access
+    log_history = data.get("log_history")  # Use .get for safer access
     if not isinstance(log_history, list):
         print(f"Error: 'log_history' key not found or not a list in {json_file_path}")
         return
@@ -184,23 +189,23 @@ def plot_loss_from_json(
     # Calculate Running Average
     smoothed_losses = None
     smoothed_steps = None
-    apply_smoothing = smooth_steps > 0 and len(losses) >= smooth_steps
+    apply_smoothing = smooth_steps > 0
 
     if apply_smoothing:
         try:
             weights = np.ones(smooth_steps) / smooth_steps
-            smoothed_losses = np.convolve(losses, weights, mode='valid')
-            smoothed_steps = steps[smooth_steps - 1:] # Steps corresponding to the smoothed values
+            smoothed_losses = np.convolve(losses, weights, mode='full')[:len(losses)]
+            smoothed_steps = steps  # Steps corresponding to the smoothed values
             print(f"Calculated smoothed loss over {len(smoothed_steps)} points.")
         except Exception as e:
             print(f"Warning: Could not calculate smoothed loss. Error: {e}")
-            apply_smoothing = False # Disable if calculation fails
+            apply_smoothing = False  # Disable if calculation fails
     elif smooth_steps > 0:
         print(f"Warning: Not enough data points ({len(losses)}) for smoothing window ({smooth_steps}). Skipping smoothing.")
 
     # Plotting
-    plt.style.use('seaborn-v0_8-darkgrid') # Use a nice style
-    plt.figure(figsize=(10, 6)) # Standard figure size
+    plt.style.use('seaborn-v0_8-darkgrid')  # Use a nice style
+    plt.figure(figsize=(10, 6))  # Standard figure size
 
     plt.plot(steps, losses, linestyle='-', color='skyblue', alpha=0.5, label='Original Loss')
 
@@ -212,7 +217,7 @@ def plot_loss_from_json(
     plt.ylabel("Loss")
     plt.title("Training Loss Progression")
     plt.legend()
-    plt.tight_layout() # Adjust layout
+    plt.tight_layout()  # Adjust layout
 
     # Saving
     try:
@@ -221,7 +226,7 @@ def plot_loss_from_json(
     except Exception as e:
         print(f"Error saving plot to {output_image_path}: {e}")
     finally:
-        plt.close() # Ensure figure is closed to free memory
+        plt.close()  # Ensure figure is closed to free memory
 
 def prepare_checkpoint_folder(checkpoint_path: Path, checkpoint_number: int):
     """