Fix for <end_of_turn> token during inference (#622)

quic-akuruvil · web-flow · commit b014f72b2773 · 2025-11-19T09:07:08.000+05:30
Fix for this JIRA from Imagine team

Signed-off-by: Ann Kuruvilla &lt;akuruvil@qti.qualcomm.com&gt;
diff --git a/examples/gemma3_example/gemma3_mm.py b/examples/gemma3_example/gemma3_mm.py
@@ -105,5 +105,5 @@
     )
     inputs["pixel_values"] = inputs["pixel_values"].to(torch.float32)
     output = qeff_model.generate(inputs=inputs, generation_len=100)
-    print(tokenizer.batch_decode(output.generated_ids))
+    print(tokenizer.batch_decode(output.generated_ids, skip_special_tokens=True))
     print(output)

Original file line number	Diff line number	Diff line change
`@@ -105,5 +105,5 @@`
`105`	`105`	`)`
`106`	`106`	`inputs["pixel_values"] = inputs["pixel_values"].to(torch.float32)`
`107`	`107`	`output = qeff_model.generate(inputs=inputs, generation_len=100)`
`108`		`- print(tokenizer.batch_decode(output.generated_ids))`
	`108`	`+ print(tokenizer.batch_decode(output.generated_ids, skip_special_tokens=True))`
`109`	`109`	`print(output)`