[wip] almost there

graphs/openwebuilocalllms.gen_quant_graph.py

@@ -0,0 +1,53 @@
+import matplotlib.pyplot as plt
+
+# Your data: quantization level -> (memory usage, accuracy)
+data = {
+    "Q2_K": (3032, 74.29),
+    "Q3_K_S": (3495, 82.19),
+    "Q3_K_M": (3833, 93.29),
+    "Q4_0": (4460, 96.09),
+    "Q4_K_S": (4476, 97.38),
+    "Q4_K_M": (4693, 97.67),
+    "Q4_1": (4893, 97.18),
+    "Q5_0": (5354, 98.98),
+    "Q5_K_S": (5340, 99.08),
+    "Q5_K_M": (5468, 99.00),
+    "Q5_1": (5788, 99.16),
+    "Q6_K": (6291, 99.58),
+    "Q8_0": (8146, 99.93)
+}
+
+# Extract labels, memory usage, and accuracy
+labels = list(data.keys())
+memory_usage = [value[0] for value in data.values()]
+accuracy = [value[1] for value in data.values()]
+
+# Plot setup using a dark theme
+plt.style.use('dark_background')
+
+fig, ax1 = plt.subplots()
+plt.title('Quantization Levels of llama 3.1 8B')
+
+# Create two y-axes: one for memory usage and the other for accuracy
+color_memory = 'tab:cyan'
+ax1.set_xlabel('Quantization Level')
+ax1.set_ylabel('Memory Usage (MB)', color=color_memory)
+ax1.bar(labels, memory_usage, color=color_memory, alpha=0.8, label='Memory Usage')
+ax1.tick_params(axis='y', labelcolor=color_memory)
+
+# Second y-axis for accuracy
+ax2 = ax1.twinx()
+color_accuracy = 'tab:orange'
+ax2.set_ylabel('Accuracy (%)', color=color_accuracy)
+ax2.plot(labels, accuracy, color=color_accuracy, marker='o', linestyle='-', linewidth=2, markersize=8, label='Accuracy')
+ax2.tick_params(axis='y', labelcolor=color_accuracy)
+
+# Adding legends
+fig.tight_layout()  # To ensure the layout is tight
+lines1, labels1 = ax1.get_legend_handles_labels()
+lines2, labels2 = ax2.get_legend_handles_labels()
+ax1.legend(lines1 + lines2, labels1 + labels2, loc='upper left')
+
+
+# Show plot
+plt.show()