Add OpenAI backend to the CI test (#869)

.github/workflows/pr-e2e-test.yml

@@ -32,7 +32,13 @@ jobs:
         pip install flashinfer -i https://flashinfer.ai/whl/cu121/torch2.3/ --force-reinstall
         pip install --upgrade transformers
 
-    - name: Launch server and run benchmark
+    - name: Test OpenAI Backend
+      run: |
+        export OPENAI_API_KEY=secrets.OPENAI_API_KEY
+        cd sglang/test/lang
+        python3 test_openai_backend.py
+
+    - name: Benchmark Serving
       run: |
         python3 -m sglang.launch_server --model /home/lmzheng/zhyncs/Meta-Llama-3.1-8B-Instruct --port 8413 --disable-radix-cache &
 

.github/workflows/release-docker.yml

-name: publish docker
+name: Release Docker
 on:
   push:
     branches:

.github/workflows/release-fake-tag.yml

-name: fake tag
+name: Release Fake Tag
 on:
   push:
     branches:

.github/workflows/release-github.yml

-name: release tag
+name: Release GitHub
 on:
   workflow_dispatch:
 jobs:

.github/workflows/release-pypi.yml

-name: publish to pypi
+name: Release PyPI
 on:
   push:
     branches:

docs/en/benchmark_results.md

-# Benchmark Results
-
-We tested our system on the following common LLM workloads and reported the achieved throughput:
-- **[MMLU](https://arxiv.org/abs/2009.03300)**: A 5-shot, multi-choice, multi-task benchmark.
-- **[HellaSwag](https://arxiv.org/abs/1905.07830)**: A 20-shot, multi-choice sentence completion benchmark.
-- **[ReAct Agent](https://arxiv.org/abs/2210.03629)**: An agent task using prompt traces collected from the original ReAct paper.
-- **[Tree-of-Thought](https://arxiv.org/pdf/2305.10601.pdf)**: A custom tree search-based prompt for solving GSM-8K problems.
-- **JSON Decode**: Extracting information from a Wikipedia page and outputting it in JSON format.
-- **Chat (short)**: A synthetic chat benchmark where each conversation includes 4 turns with short LLM outputs.
-- **Chat (long)**: A synthetic chat benchmark where each conversation includes 4 turns with long LLM outputs.
-- **[DSPy RAG](https://github.com/stanfordnlp/dspy)**: A retrieval-augmented generation pipeline in the DSPy tutorial.
-- **[LLaVA Bench](https://github.com/haotian-liu/LLaVA)**: Running LLaVA v1.5, a vision language model on the LLaVA-in-the-wild benchmark.
-
-We tested both Llama-7B on one NVIDIA A10G GPU (24GB) and Mixtral-8x7B on 8 NVIDIA A10G GPUs with tensor parallelism, using FP16 precision. We used vllm v0.2.5, guidance v0.1.8, Hugging Face TGI v1.3.0, and SGLang v0.1.5.
-
-- Llama-7B on NVIDIA A10G, FP16, Tensor Parallelism=1
-![llama_7b](../assets/llama_7b.jpg)
-
-- Mixtral-8x7B on NVIDIA A10G, FP16, Tensor Parallelism=8
-![mixtral_8x7b](../assets/mixtral_8x7b.jpg)
-
-The benchmark code is available [here](https://github.com/sgl-project/sglang/tree/main/benchmark).

docs/en/model_support.md

@@ -11,6 +11,6 @@ To port a model from vLLM to SGLang, you can compare these two files [SGLang LLa
   - Change `forward()` functions, and add `input_metadata`.
   - Add `EntryClass` at the end.
   - Test correctness by comparing the final logits and outputs of the two following commands:
-    - `python3 playground/reference_hf.py --model [new model]`
+    - `python3 scripts/playground/reference_hf.py --model [new model]`
     - `python3 -m sglang.bench_latency --model [new model] --correct --output-len 16 --trust-remote-code`
   - Update [Supported Models](https://github.com/sgl-project/sglang/tree/main?tab=readme-ov-file#supported-models) at [README](../README.md).

test/lang/test_bind_cache.py

+"""
+Usage:
+python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
+python3 test_bind_cache.py
+"""
+
 import unittest
 
 import sglang as sgl

test/lang/test_srt_backend.py

 """
-python3 -m sglang.launch_server --model-path meta-llama/Llama-2-7b-chat-hf --port 30000
+Usage:
+python3 -m sglang.launch_server --model-path meta-llama/Meta-Llama-3-8B-Instruct --port 30000
+python3 test_srt_backend.py
 """
 
 import json

test/lang/test_tracing.py

@@ -16,7 +16,7 @@ class TestTracing(unittest.TestCase):
             s += "A:" + sgl.gen("answer", stop="\n")
 
         tracer = few_shot_qa.trace()
-        print(tracer.last_node.print_graph_dfs() + "\n")
+        # print(tracer.last_node.print_graph_dfs() + "\n")
 
     def test_select(self):
         @sgl.function
@@ -26,7 +26,7 @@ class TestTracing(unittest.TestCase):
             s += "It is a city" + sgl.gen("description", stop=".")
 
         tracer = capital.trace()
-        print(tracer.last_node.print_graph_dfs() + "\n")
+        # print(tracer.last_node.print_graph_dfs() + "\n")
 
     def test_raise_warning(self):
         @sgl.function
@@ -66,11 +66,11 @@ class TestTracing(unittest.TestCase):
             s += "In summary" + sgl.gen("summary")
 
         compiled = tip_suggestion.compile()
-        compiled.print_graph()
+        # compiled.print_graph()
 
         sgl.set_default_backend(sgl.OpenAI("gpt-3.5-turbo-instruct"))
         state = compiled.run(topic="staying healthy")
-        print(state.text() + "\n")
+        # print(state.text() + "\n")
 
         states = compiled.run_batch(
             [
@@ -80,8 +80,8 @@ class TestTracing(unittest.TestCase):
             ],
             temperature=0,
         )
-        for s in states:
-            print(s.text() + "\n")
+        # for s in states:
+        #     print(s.text() + "\n")
 
     def test_role(self):
         @sgl.function
@@ -95,7 +95,7 @@ class TestTracing(unittest.TestCase):
         backend.chat_template = get_chat_template("llama-2-chat")
 
         compiled = multi_turn_chat.compile(backend=backend)
-        compiled.print_graph()
+        # compiled.print_graph()
 
     def test_fork(self):
         @sgl.function
@@ -118,10 +118,10 @@ class TestTracing(unittest.TestCase):
             s += "In summary" + sgl.gen("summary")
 
         tracer = tip_suggestion.trace()
-        print(tracer.last_node.print_graph_dfs())
+        # print(tracer.last_node.print_graph_dfs())
 
         a = tip_suggestion.run(backend=sgl.OpenAI("gpt-3.5-turbo-instruct"))
-        print(a.text())
+        # print(a.text())
 
 
 if __name__ == "__main__":