Benchmark Updates (#382)

benchmark/gsm8k/bench_sglang.py

@@ -73,7 +73,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = few_shot_gsm8k.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     preds = []

benchmark/hellaswag/bench_sglang.py

@@ -61,7 +61,7 @@ def main(args):
     # Run requests
     tic = time.time()
     rets = few_shot_hellaswag.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     preds = [choices[i].index(rets[i]["answer"]) for i in range(len(rets))]
     latency = time.time() - tic
 

benchmark/json_decode_regex/bench_sglang.py

@@ -63,7 +63,7 @@ def main(args):
 
     # Run requests
     tic = time.time()
-    states = json_decode.run_batch(arguments, temperature=0, num_threads=args.parallel)
+    states = json_decode.run_batch(arguments, temperature=0, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     # Compute accuracy

benchmark/json_jump_forward/bench_sglang.py

@@ -72,7 +72,7 @@ def bench_city_doc(args):
         arguments,
         temperature=0,
         num_threads=args.parallel,
-        progress_bar=(args.parallel == 1),
+        progress_bar=True,
     )
     latency = time.time() - tic
 
@@ -96,7 +96,7 @@ def bench_character(args):
         arguments,
         temperature=0,
         num_threads=args.parallel,
-        progress_bar=(args.parallel == 1),
+        progress_bar=True,
     )
     latency = time.time() - tic
 

benchmark/line_retrieval/bench_sglang.py

@@ -61,7 +61,7 @@ def eval_model(args, line_obj, num_hoops, src_indices, dst_percents):
 
     tic = time.time()
     states = line_retrieval.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     corrects = []

benchmark/llm_judge/bench_sglang.py

@@ -54,7 +54,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = multi_dimension_judge.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     print(f"Latency: {latency:.3f}")

benchmark/long_json_decode/bench_sglang.py

@@ -36,7 +36,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = json_decode.run_batch(
-        arguments, temperature=0, num_threads=args.parallel)
+        arguments, temperature=0, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     # Compute accuracy 

benchmark/mtbench/bench_sglang.py

@@ -60,7 +60,9 @@ def main(args):
         arguments,
         temperature=0,
         max_new_tokens=256,
-        num_threads=args.parallel)
+        num_threads=args.parallel,
+        progress_bar=True,
+    )
     answers = [[s["answer_1"], s["answer_2"]] for s in rets]
     latency = time.time() - tic
 

benchmark/multi_chain_reasoning/README.md

@@ -7,7 +7,7 @@ wget https://raw.githubusercontent.com/openai/grade-school-math/master/grade_sch
 
 ### Benchmark sglang
 ```
-python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000
+python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000  --schedule-conservativeness 1.3
 ```
 
 ```

benchmark/multi_chain_reasoning/bench_sglang.py

@@ -86,7 +86,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = multi_chain_gsm8k.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     preds = []

benchmark/multi_document_qa/bench_sglang.py

@@ -43,7 +43,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = multi_document_qa.run_batch(
-        arguments, temperature=0, num_threads=args.parallel)
+        arguments, temperature=0, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     # Compute accuracy 

benchmark/multi_turn_chat/bench_sglang.py

@@ -29,7 +29,7 @@ def main(args):
 
     tic = time.time()
     states = multi_turns.run_batch(
-        multi_qas, temperature=0, backend=backend, num_threads=args.parallel
+        multi_qas, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True
     )
     latency = time.time() - tic
 

benchmark/react/bench_sglang.py

@@ -110,7 +110,9 @@ def main(args):
     tic = time.time()
     states = webthink.run_batch(arguments,
                                 temperature=0,
-                                num_threads=args.parallel)
+                                num_threads=args.parallel,
+                                progress_bar=True,
+                                )
     latency = time.time() - tic
 
     # Compute accuracy

benchmark/tip_suggestion/.gitignore

@@ -0,0 +1 @@
+!topic.jsonl

benchmark/tip_suggestion/bench_sglang.py

@@ -59,7 +59,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = suggest_tips.run_batch(
-        arguments, temperature=0, num_threads=args.parallel)
+        arguments, temperature=0, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
 
     # Compute accuracy

benchmark/tip_suggestion/topic.jsonl

@@ -0,0 +1,50 @@
+{"topic": "organizing a successful charity event", "number": 6}
+{"topic": "improving personal credit scores", "number": 7}
+{"topic": "staying motivated during job searches", "number": 5}
+{"topic": "maintaining a work-life balance", "number": 9}
+{"topic": "reducing carbon footprint at home", "number": 8}
+{"topic": "starting a book club", "number": 5}
+{"topic": "learning to play a musical instrument", "number": 7}
+{"topic": "getting into freelance writing", "number": 6}
+{"topic": "beginner yoga poses", "number": 8}
+{"topic": "preparing for graduate school exams", "number": 5}
+{"topic": "exploring minimalist living", "number": 9}
+{"topic": "effective grocery shopping", "number": 7}
+{"topic": "winter camping", "number": 5}
+{"topic": "starting a podcast on a budget", "number": 8}
+{"topic": "creating a capsule wardrobe", "number": 6}
+{"topic": "improving your writing skills", "number": 7}
+{"topic": "learning a new software quickly", "number": 9}
+{"topic": "reducing anxiety before public speaking", "number": 5}
+{"topic": "planning a solo travel adventure", "number": 8}
+{"topic": "beginner skateboarders", "number": 6}
+{"topic": "studying abroad", "number": 7}
+{"topic": "planting a vegetable garden", "number": 5}
+{"topic": "adopting a shelter pet", "number": 9}
+{"topic": "learning to cook ethnic cuisines", "number": 8}
+{"topic": "effective conflict resolution", "number": 5}
+{"topic": "starting a vlog", "number": 7}
+{"topic": "keeping a daily journal", "number": 6}
+{"topic": "improving sleep hygiene", "number": 8}
+{"topic": "beginner mountain climbers", "number": 5}
+{"topic": "creating a mobile app", "number": 9}
+{"topic": "maintaining a saltwater aquarium", "number": 7}
+{"topic": "preparing for a baby's arrival", "number": 6}
+{"topic": "writing a fantasy novel", "number": 5}
+{"topic": "effective team leadership", "number": 8}
+{"topic": "making a documentary film", "number": 9}
+{"topic": "learning about historical events", "number": 7}
+{"topic": "baking gluten-free treats", "number": 6}
+{"topic": "improving mental arithmetic skills", "number": 5}
+{"topic": "building a treehouse", "number": 8}
+{"topic": "getting started with watercolor painting", "number": 9}
+{"topic": "creating a YouTube tutorial series", "number": 7}
+{"topic": "landscape photography", "number": 5}
+{"topic": "navigating cultural differences", "number": 6}
+{"topic": "preparing for a marathon", "number": 8}
+{"topic": "building an online business", "number": 9}
+{"topic": "learning to dance at home", "number": 5}
+{"topic": "self-publishing a book", "number": 7}
+{"topic": "starting an urban farm", "number": 6}
+{"topic": "improving your memory", "number": 8}
+{"topic": "creating a personal brand online", "number": 9}

benchmark/tree_of_thought_deep/bench_sglang.py

@@ -112,7 +112,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = tree_search.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
     answers_text = []
     for s in states:

benchmark/tree_of_thought_v0/bench_sglang.py

@@ -102,7 +102,7 @@ def main(args):
     # Run requests
     tic = time.time()
     states = tree_search.run_batch(
-        arguments, temperature=0, backend=backend, num_threads=args.parallel)
+        arguments, temperature=0, backend=backend, num_threads=args.parallel, progress_bar=True)
     latency = time.time() - tic
     answers_text = []
     for s in states: