Update vlm_inference.py

vlm_inference.py

@@ -166,10 +166,6 @@ def vlm_infer_stream(
     logits = model.llm.vocab_projection(x)[:, -1, :] / temperature
     probs = F.softmax(logits, dim=-1)
     next_token = torch.multinomial(probs, 1)
-    if int(next_token.item()) in stop_ids:
-        print("DEBUG first token in stop_ids")
-        yield "[EOS]"
-        return
 
     acc, last = [], ""
 
@@ -177,7 +173,6 @@ def vlm_infer_stream(
         # sampled from prefill
         tid = int(next_token.item())
         if tid in stop_ids:
-            yield "[EOS]"
             break