[Fix] Support Qwen3-Coder-Next hybrid architecture in model inspector

.githooks/commit-msg

@@ -0,0 +1,52 @@
+#!/usr/bin/env bash
+#
+# commit-msg hook: enforces [Tag] Description format on commit messages.
+# Allowed tags: Feature, Fix, Refactor, Docs, Test, CI, Chore, Perf
+# Minimum description length: 10 characters
+#
+
+COMMIT_MSG_FILE="$1"
+COMMIT_MSG=$(head -1 "$COMMIT_MSG_FILE")
+
+# Skip merge commits
+if echo "$COMMIT_MSG" | grep -qE '^Merge '; then
+  exit 0
+fi
+
+# Reject Co-Authored-By or other trailers in the subject line
+if echo "$COMMIT_MSG" | grep -qiE '(Co-Authored-By|Signed-off-by|Reviewed-by):'; then
+  echo ""
+  echo "ERROR: Trailer found in the commit subject line."
+  echo "  Move 'Co-Authored-By:', 'Signed-off-by:', etc. to the commit body"
+  echo "  (separated from the subject by a blank line)."
+  echo ""
+  echo "  Example:"
+  echo "    [Feature] Add new authentication flow"
+  echo "    "
+  echo "    Co-Authored-By: Name <email>"
+  echo ""
+  exit 1
+fi
+
+# Validate [Tag] Description format
+PATTERN='^\[(Feature|Fix|Refactor|Docs|Test|CI|Chore|Perf)\] .{10,}'
+
+if ! echo "$COMMIT_MSG" | grep -qE "$PATTERN"; then
+  echo ""
+  echo "ERROR: Commit message does not follow the required format."
+  echo ""
+  echo "  Required: [Tag] Description (at least 10 characters)"
+  echo ""
+  echo "  Allowed tags: [Feature], [Fix], [Refactor], [Docs], [Test], [CI], [Chore], [Perf]"
+  echo ""
+  echo "  Examples:"
+  echo "    [Feature] Add user authentication endpoint"
+  echo "    [Fix] Resolve race condition in scheduler"
+  echo "    [Refactor] Simplify model loading pipeline"
+  echo ""
+  echo "  Your message: $COMMIT_MSG"
+  echo ""
+  exit 1
+fi
+
+exit 0

.githooks/pre-push

@@ -0,0 +1,42 @@
+#!/usr/bin/env bash
+#
+# pre-push hook: warns on direct pushes to main and validates last commit message.
+#
+
+REMOTE="$1"
+BRANCH=$(git symbolic-ref --short HEAD 2>/dev/null)
+
+# Warn if pushing directly to main
+if [ "$BRANCH" = "main" ] || [ "$BRANCH" = "master" ]; then
+  echo ""
+  echo "WARNING: You are pushing directly to '$BRANCH'."
+  echo "  Consider using a feature branch and opening a pull request instead."
+  echo ""
+fi
+
+# Validate the last commit message format (catches --no-verify commits)
+LAST_MSG=$(git log -1 --format='%s')
+
+# Skip merge commits
+if echo "$LAST_MSG" | grep -qE '^Merge '; then
+  exit 0
+fi
+
+PATTERN='^\[(Feature|Fix|Refactor|Docs|Test|CI|Chore|Perf)\] .{10,}'
+
+if ! echo "$LAST_MSG" | grep -qE "$PATTERN"; then
+  echo ""
+  echo "ERROR: Last commit message does not follow the required format."
+  echo ""
+  echo "  Required: [Tag] Description (at least 10 characters)"
+  echo "  Allowed tags: [Feature], [Fix], [Refactor], [Docs], [Test], [CI], [Chore], [Perf]"
+  echo ""
+  echo "  Last commit: $LAST_MSG"
+  echo ""
+  echo "  Please amend your commit message before pushing:"
+  echo "    git commit --amend"
+  echo ""
+  exit 1
+fi
+
+exit 0

.github/pull_request_template.md

@@ -0,0 +1,3 @@
+## Summary
+<!-- Describe what this PR does and why -->
+-

.github/workflows/pr-lint.yml

@@ -0,0 +1,45 @@
+name: PR Lint
+
+on:
+  pull_request:
+    types: [opened, edited, synchronize]
+
+jobs:
+  validate-pr:
+    runs-on: ubuntu-latest
+    steps:
+      - name: Validate PR title
+        env:
+          PR_TITLE: ${{ github.event.pull_request.title }}
+        run: |
+          PATTERN='^\[(Feature|Fix|Refactor|Docs|Test|CI|Chore|Perf)\] .{10,}'
+          if ! echo "$PR_TITLE" | grep -qE "$PATTERN"; then
+            echo "::error::PR title must match format: [Tag] Description (at least 10 chars)"
+            echo ""
+            echo "Allowed tags: [Feature], [Fix], [Refactor], [Docs], [Test], [CI], [Chore], [Perf]"
+            echo ""
+            echo "Examples:"
+            echo "  [Feature] Add user authentication endpoint"
+            echo "  [Fix] Resolve race condition in scheduler"
+            echo ""
+            echo "Your title: $PR_TITLE"
+            exit 1
+          fi
+          echo "PR title is valid."
+
+      - name: Validate PR body
+        env:
+          PR_BODY: ${{ github.event.pull_request.body }}
+        run: |
+          if ! echo "$PR_BODY" | grep -q '## Summary'; then
+            echo "::error::PR body must contain a '## Summary' section."
+            exit 1
+          fi
+
+          # Check that Summary section has content beyond the template placeholder
+          SUMMARY_CONTENT=$(echo "$PR_BODY" | sed -n '/## Summary/,/^## /p' | grep -v '## Summary' | grep -v '^## ' | grep -v '<!--.*-->' | sed '/^[[:space:]]*$/d' | sed '/^-$/d')
+          if [ -z "$SUMMARY_CONTENT" ]; then
+            echo "::error::PR Summary section must contain a description (not just the template placeholder)."
+            exit 1
+          fi
+          echo "PR body is valid."

CLAUDE.md

@@ -10,6 +10,21 @@ SGLang is a high-performance serving framework for large language models (LLMs)
 - **sgl-kernel/** — Optimized CUDA/C++ compute kernels
 - **sgl-model-gateway/** — Rust-based routing/API gateway (OpenAI-compatible)
 
+## Setup
+
+After cloning, activate the git hooks:
+```bash
+bash setup-hooks.sh
+```
+
+## Commit & PR Conventions
+
+All commit messages and PR titles must follow: `[Tag] Description` .
+
+Allowed tags: `[Feature]`, `[Fix]`, `[Refactor]`, `[Docs]`, `[Test]`, `[CI]`, `[Chore]`, `[Perf]`
+
+Trailers like `Co-Authored-By:` is not allowed to show nerither in the commit body or the subject line.
+
 ## Build & Development Commands
 
 ### Python package (main runtime)
@@ -23,11 +38,6 @@ cd sglang/sgl-kernel && make build
 # Resource-limited: make build MAX_JOBS=2 CMAKE_ARGS="-DSGL_KERNEL_COMPILE_THREADS=1"
 ```
 
-### sgl-model-gateway (Rust)
-```bash
-cd sglang/sgl-model-gateway && make build
-```
-
 ### Running the server
 ```bash
 python -m sglang.launch_server --model-path meta-llama/Llama-2-7b-hf
@@ -40,30 +50,8 @@ pre-commit run --all-files
 ```
 Hooks: isort, ruff (F401/F821), black-jupyter, codespell, clang-format, nbstripout.
 
-### Testing
-```bash
-# Single test file
-python3 test/srt/test_srt_endpoint.py
-
-# Single test case
-python3 test/srt/test_srt_endpoint.py TestSRTEndpoint.test_simple_decode
-
-# Test suite (per-commit)
-python3 test/srt/run_suite.py --suite per-commit
-
-# sgl-kernel tests
-cd sglang/sgl-kernel && pytest tests/
-```
 
 ## Architecture
-
-### Frontend (`python/sglang/lang/`)
-Language-level API for composing LLM programs. Key files:
-- `api.py` — Public API exports (Engine, function, gen, etc.)
-- `interpreter.py` — Program interpreter
-- `ir.py` — Intermediate representation
-- `backend/` — Backend connectors (OpenAI, Anthropic, LiteLLM, VertexAI, SGLang runtime)
-
 ### Serving Runtime (`python/sglang/srt/`)
 The core inference engine. Major subsystems:
 - **`entrypoints/`** — Engine entry points, gRPC server
@@ -83,9 +71,6 @@ The core inference engine. Major subsystems:
 ### sgl-kernel (`sgl-kernel/`)
 CMake-based C++/CUDA project with Python bindings. Contains optimized kernels for attention, quantization, and other compute-intensive operations. Built via scikit-build-core.
 
-### sgl-model-gateway (`sgl-model-gateway/`)
-Rust async service (tokio/tonic) providing HTTP and gRPC routing, load balancing, and an OpenAI-compatible API layer. Has Python and Go bindings in `bindings/`.
-
 ### Key Design Patterns
 - **RadixAttention**: Prefix-aware KV cache reuse across requests
 - **Continuous batching**: Dynamic request scheduling for throughput
@@ -94,12 +79,4 @@ Rust async service (tokio/tonic) providing HTTP and gRPC routing, load balancing
 
 ## Version Management
 
-Versions are managed via `setuptools-scm` from git tags. Generated gRPC files (`*_pb2.py`, `*_pb2_grpc.py`) are excluded from linting.
-
-## Hardware Support
-
-NVIDIA CUDA (primary), AMD ROCm/HIP, Intel Gaudi, Ascend NPU, Google TPU (experimental). Hardware-specific code lives in `srt/hardware_backend/`.
-
-## Key Dependencies
-
-PyTorch >= 2.8.0, Transformers 4.57.1, Flash Attention, Triton (kernels), FastAPI/Uvicorn (HTTP), tonic (gRPC/Rust), outlines/xgrammar/llguidance (structured generation).
+Versions are managed via `setuptools-scm` from git tags. Generated gRPC files (`*_pb2.py`, `*_pb2_grpc.py`) are excluded from linting.

configs/qwen3_coder_next_config.json

@@ -0,0 +1,44 @@
+{
+  "architectures": [
+    "Qwen3NextForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0,
+  "bos_token_id": 151643,
+  "decoder_sparse_step": 1,
+  "eos_token_id": 151645,
+  "full_attention_interval": 4,
+  "head_dim": 256,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5120,
+  "linear_conv_kernel_dim": 4,
+  "linear_key_head_dim": 128,
+  "linear_num_key_heads": 16,
+  "linear_num_value_heads": 32,
+  "linear_value_head_dim": 128,
+  "max_position_embeddings": 262144,
+  "mlp_only_layers": [],
+  "model_type": "qwen3_next",
+  "moe_intermediate_size": 512,
+  "norm_topk_prob": true,
+  "num_attention_heads": 16,
+  "num_experts": 512,
+  "num_experts_per_tok": 10,
+  "num_hidden_layers": 48,
+  "num_key_value_heads": 2,
+  "output_router_logits": false,
+  "partial_rotary_factor": 0.25,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 5000000,
+  "router_aux_loss_coef": 0.001,
+  "shared_expert_intermediate_size": 512,
+  "tie_word_embeddings": false,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.57.0.dev0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}