minor: Add basic editorconfig and pre-commit hooks to enforce style for whitespaces (#1926)

docs/references/troubleshooting.md

@@ -3,9 +3,9 @@
 This page lists some common errors and tips for fixing them.
 
 ## CUDA out of memory
-If you see out of memory (OOM) errors, you can try to tune the following parameters.  
-If OOM happens during prefill, try to decrease `--chunked-prefill-size` to `4096` or `2048`.  
-If OOM happens during decoding, try to decrease `--max-running-requests`.  
+If you see out of memory (OOM) errors, you can try to tune the following parameters.
+If OOM happens during prefill, try to decrease `--chunked-prefill-size` to `4096` or `2048`.
+If OOM happens during decoding, try to decrease `--max-running-requests`.
 You can also try to decrease `--mem-fraction-static`, which reduces the memory usage of the KV cache memory pool and helps both prefill and decoding.
 
 ## CUDA error: an illegal memory access was encountered

docs/requirements.txt

@@ -14,4 +14,4 @@ sphinx-book-theme
 sphinx-copybutton
 sphinx-tabs
 sphinxcontrib-mermaid
-urllib3<2.0.0
+urllib3<2.0.0
\ No newline at end of file

examples/frontend_language/usage/llava_video/srt_example_llava_v.sh

@@ -33,7 +33,7 @@ CUR_NODES_IDX=$2
 
 VIDEO_DIR=$3
 
-MODEL_PATH=$4   
+MODEL_PATH=$4
 
 NUM_FRAMES=$5
 
@@ -73,16 +73,16 @@ for IDX in $(seq 1 $LOCAL_CHUNKS); do
     (
         START=$(((IDX-1) * GPUS_PER_CHUNK))
         LENGTH=$GPUS_PER_CHUNK # Length for slicing, not the end index
-        
+
         CHUNK_GPUS=(${GPULIST[@]:$START:$LENGTH})
-        
+
         # Convert the chunk GPUs array to a comma-separated string
         CHUNK_GPUS_STR=$(IFS=,; echo "${CHUNK_GPUS[*]}")
 
         LOCAL_IDX=$((CUR_NODES_IDX * LOCAL_CHUNKS + IDX))
 
         echo "Chunk $(($LOCAL_IDX - 1)) will run on GPUs $CHUNK_GPUS_STR"
-        
+
         # Calculate the port for this chunk. Ensure it's incremented by 5 for each chunk.
         PORT=$((10000 + RANDOM % 55536))
 
@@ -92,7 +92,7 @@ for IDX in $(seq 1 $LOCAL_CHUNKS); do
 
         while [ $RETRY_COUNT -lt $MAX_RETRIES ] && [ $COMMAND_STATUS -ne 0 ]; do
             echo "Running chunk $(($LOCAL_IDX - 1)) on GPUs $CHUNK_GPUS_STR with port $PORT. Attempt $(($RETRY_COUNT + 1))"
-            
+
 #!/bin/bash
             CUDA_VISIBLE_DEVICES=$CHUNK_GPUS_STR python3 srt_example_llava_v.py \
             --port $PORT \
@@ -102,10 +102,10 @@ for IDX in $(seq 1 $LOCAL_CHUNKS); do
             --video-dir $VIDEO_DIR \
             --model-path $MODEL_PATH \
             --num-frames $NUM_FRAMES #&
-            
+
             wait $!  # Wait for the process to finish and capture its exit status
             COMMAND_STATUS=$?
-            
+
             if [ $COMMAND_STATUS -ne 0 ]; then
                 echo "Execution failed for chunk $(($LOCAL_IDX - 1)), attempt $(($RETRY_COUNT + 1)). Retrying..."
                 RETRY_COUNT=$(($RETRY_COUNT + 1))
@@ -124,8 +124,8 @@ done
 
 wait
 
-cat work_dirs/llava_next_video_inference_results/final_results_chunk_*.csv > work_dirs/llava_next_video_inference_results/final_results_node_${CUR_NODES_IDX}.csv   
+cat work_dirs/llava_next_video_inference_results/final_results_chunk_*.csv > work_dirs/llava_next_video_inference_results/final_results_node_${CUR_NODES_IDX}.csv
 
 END_TIME=$(date +%s)  # Capture end time
 ELAPSED_TIME=$(($END_TIME - $START_TIME))
-echo "Total execution time: $ELAPSED_TIME seconds."
+echo "Total execution time: $ELAPSED_TIME seconds."
\ No newline at end of file

examples/frontend_language/usage/openai_chat_speculative.py

@@ -4,8 +4,8 @@ Usage:
 Show in "assistant" the desired answer format. Each "gen" term should have a stop token.
 The stream mode is not supported in speculative execution.
 
-E.g. 
-correct: 
+E.g.
+correct:
     sgl.assistant("\nName:" + sgl.gen("name", stop="\n") + "\nBirthday:" + sgl.gen("birthday", stop="\n") + "\nJob:" + sgl.gen("job", stop="\n"))
 incorrect:
     s += sgl.assistant("\nName:" + sgl.gen("name", stop="\n"))

examples/frontend_language/usage/triton/Dockerfile

@@ -7,4 +7,4 @@ RUN git clone https://github.com/sgl-project/sglang.git
 WORKDIR /opt/sglang
 RUN pip install --upgrade pip && \
     pip install -e "python[all]" && \
-    pip install datasets
+    pip install datasets
\ No newline at end of file

examples/frontend_language/usage/triton/README.md

@@ -32,4 +32,4 @@ curl -X POST http://localhost:8000/v2/models/character_generation/generate \
   "INPUT_TEXT": ["harry"]
 }'
 
-```
+```
\ No newline at end of file

examples/runtime/engine/input_ids.py

@@ -21,7 +21,7 @@ def main():
     # Tokenize inputs
     tokenizer = get_tokenizer(MODEL_PATH)
     token_ids_list = [tokenizer.encode(prompt) for prompt in prompts]
-    
+
     # Create an LLM.
     # You can also specify `skip_tokenizer_init=True`, but it requires explicit detokenization at the end
     llm = sgl.Engine(model_path=MODEL_PATH)
@@ -36,4 +36,4 @@ def main():
 # The __main__ condition is necessary here because we use "spawn" to create subprocesses
 # Spawn starts a fresh program every time, if there is no __main__, it will run into infinite loop to keep spawning processes from sgl.Engine
 if __name__ == "__main__":
-    main()
+    main()
\ No newline at end of file

examples/runtime/engine/readme.md

@@ -37,4 +37,4 @@ curl -X POST http://localhost:8000/generate  -H "Content-Type: application/json"
 curl -X POST http://localhost:8000/generate_stream  -H "Content-Type: application/json"  -d '{"prompt": "The Transformer architecture is..."}' --no-buffer
 ```
 
-This will send both non-streaming and streaming requests to the server.
+This will send both non-streaming and streaming requests to the server.
\ No newline at end of file

examples/runtime/llava_onevision/http_llama3_llava_test.py

@@ -3,7 +3,7 @@ Usage:
 # Installing latest llava-next: pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
 # Installing latest sglang.
 
-# Endpoint Service CLI: 
+# Endpoint Service CLI:
 python -m sglang.launch_server --model-path lmms-lab/llama3-llava-next-8b --port=30000
 
 python3 http_llama3_llava_test.py

examples/runtime/llava_onevision/http_qwen_llava_test.py

@@ -3,7 +3,7 @@ Usage:
 # Installing latest llava-next: pip install git+https://github.com/LLaVA-VL/LLaVA-NeXT.git
 # Installing latest sglang.
 
-# Endpoint Service CLI: 
+# Endpoint Service CLI:
 python -m sglang.launch_server --model-path lmms-lab/llava-next-72b --port=30000 --tp-size=8
 
 python3 http_qwen_llava_test.py

python/sglang/srt/layers/quantization/base_config.py

@@ -134,4 +134,4 @@ def method_has_implemented_embedding(
     class_embedding = inspect.getattr_static(method_class, "embedding", None)
 
     return (class_embedding is not None
-            and class_embedding is not base_embedding)
+            and class_embedding is not base_embedding)
\ No newline at end of file

python/sglang/srt/layers/vocab_parallel_embedding.py

@@ -311,7 +311,7 @@ class VocabParallelEmbedding(torch.nn.Module):
     def get_sharded_to_full_mapping(self) -> Optional[List[int]]:
         """Get a mapping that can be used to reindex the gathered
         logits for sampling.
-        
+
         During sampling, we gather logits from all ranks. The relationship
         of index->token_id will follow the same format as outlined in the class
         docstring. However, after the gather, we want to reindex the final
@@ -483,4 +483,4 @@ class ParallelLMHead(VocabParallelEmbedding):
 
     def forward(self, input_):
         del input_
-        raise RuntimeError("LMHead's weights should be used in the sampler.")
+        raise RuntimeError("LMHead's weights should be used in the sampler.")
\ No newline at end of file

python/sglang/srt/managers/scheduler.py

@@ -838,7 +838,7 @@ class Scheduler:
         time_per_output_tokens_iter: List[float] = []
 
         # Request stats
-        #   Decode 
+        #   Decode
         gen_throughput: float = 0.0
         #   Latency
         time_e2e_requests: List[float] = []
@@ -866,11 +866,11 @@ class Scheduler:
                     time_waiting_requests.append(req.queued_time - req.created_time)
                     num_prompt_tokens_requests.append(len(req.origin_input_ids))
                     num_generation_tokens_requests.append(len(req.output_ids))
-                    finished_reason_requests.append(                            
+                    finished_reason_requests.append(
                             req.finished_reason.to_json()
                             if req.finished_reason is not None
                             else None)
-    
+
         return Stats(
             new_seq=new_seq,
             num_running_req=num_running_req,

python/sglang/srt/managers/tokenizer_manager.py

@@ -384,7 +384,7 @@ class TokenizerManager:
             obj.load_format = self.server_args.load_format
 
         if not self.model_update_lock.locked():
-        
+
             async with self.model_update_lock:
                 # wait for the previous generation requests to finish
                 while len(self.rid_to_state) > 0:

python/sglang/srt/metrics/metrics_collector.py

@@ -151,7 +151,7 @@ class Metrics:
                 0.005, 0.01, 0.015, 0.02, 0.025, 0.03, 0.04, 0.05, 0.075, 0.1, 0.15, 0.2, 0.3, 0.4, 0.5, 0.75,
                 1.0, 2.5
             ])
-        
+
         # Request Stats
         #   Metadata
         self.num_prompt_tokens_requests = Histogram(
@@ -253,7 +253,7 @@ class PrometheusMetricsCollector(MetricsCollector):
                             stats.time_to_first_tokens_iter)
         self._log_histogram(self.metrics.histogram_time_per_output_token,
                             stats.time_per_output_tokens_iter)
-        
+
         # self._log_gauge(self.metrics.gpu_cache_usage_sys, stats.gpu_cache_usage_sys)
         self._log_gauge(self.metrics.num_running_sys, stats.num_running_req)
         self._log_gauge(self.metrics.num_waiting_sys, stats.num_waiting_req)
@@ -294,4 +294,4 @@ def build_1_2_5_buckets(max_value: int) -> List[int]:
                 buckets.append(value)
             else:
                 return buckets
-        exponent += 1
+        exponent += 1
\ No newline at end of file

python/sglang/srt/metrics/metrics_types.py

@@ -54,4 +54,4 @@ class Stats:
     num_prompt_tokens_iter: int = 0
     num_generation_tokens_iter: int = 0
     time_to_first_tokens_iter: List[float] = field(default_factory=list)
-    time_per_output_tokens_iter: List[float] = field(default_factory=list)
+    time_per_output_tokens_iter: List[float] = field(default_factory=list)
\ No newline at end of file

python/sglang/srt/mm_utils.py

@@ -17,7 +17,7 @@ limitations under the License.
 """
 Utilities for multi-modal models.
 
-This python file mainly contains utilities that were used in the 
+This python file mainly contains utilities that were used in the
 image processing logic of llava-next including operations such as
 anyres and anyres_max
 

python/sglang/srt/models/gpt2.py

@@ -136,7 +136,7 @@ class GPT2Block(nn.Module):
         layer_id: int,
         config: GPT2Config,
         cache_config = None,
-        
+
         quant_config: Optional[QuantizationConfig] = None,
         prefix: str = "",
     ):
@@ -284,4 +284,4 @@ class GPT2LMHeadModel(nn.Module):
                                     default_weight_loader)
             weight_loader(param, loaded_weight)
 
-EntryClass = GPT2LMHeadModel
+EntryClass = GPT2LMHeadModel
\ No newline at end of file

python/sglang/srt/models/qwen2_vl.py

@@ -57,27 +57,27 @@ logger = init_logger(__name__)
 
 class Qwen2VLImageInputs(TypedDict):
     pixel_values: torch.Tensor
-    """Shape: 
+    """Shape:
     `(num_patches, num_channels * patch_size * patch_size)`
     """
 
     image_grid_thw: torch.Tensor
     """Shape: `(num_images, 3)`
-    
+
     This should be in `(grid_t, grid_h, grid_w)` format.
     """
 
 
 class Qwen2VLVideoInputs(TypedDict):
     pixel_values_videos: torch.Tensor
-    """Shape: 
-    `(num_patches, 
+    """Shape:
+    `(num_patches,
       num_channels * temporal_patch_size * patch_size * patch_size)`
     """
 
     video_grid_thw: torch.Tensor
     """Shape: `(num_videos, 3)`
-    
+
     This should be in `(grid_t, grid_h, grid_w)` format.
     """