Minor changes to readme

README.md

@@ -10,7 +10,9 @@ tags:
 - computational-social-science
 ---
 
-Recommended usage is with vLLM.
+
+
+# Example usage with vLLM
 
 ## Load the model and tokenizer
 
@@ -24,6 +26,17 @@ model = LLM(model="ahalt/event-attribute-extractor",
             gpu_memory_utilization=0.80)
 
 tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3-0.6B")
+
+
+sampling_params = SamplingParams(
+    temperature=0.5,       # Greedy decoding breaks Qwen
+    top_p=0.8,             # Qwen3 non-thinking recommendation  
+    top_k=20,              # Qwen3 recommendation
+    presence_penalty=1.5,  # Recommended for quantized models
+    min_p=0.0,
+    #guided_decoding=guided_decoding_params, # Optionally, set a JSON schema for contrained decoding
+    max_tokens=1024,
+)
 ```
 
 
@@ -73,8 +86,8 @@ In a joint statement published Saturday on social media, the National Anti-Corru
 
 event_type = "Investigate, charge, or prosecute"
 
-prompt = make_prompt(text, event_type, tokenizer2)
-output = model2.generate(prompt, sampling_params=sampling_params)
+prompt = make_prompt(text, event_type, tokenizer)
+output = model.generate(prompt, sampling_params=sampling_params)
 response = output[0].outputs[0].text.strip()
 
 [{"event_type": "Investigate, charge, or prosecute",