Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
fix(integrations): include langgraph dependency in langchain and lang… (#257)
SOURAV SUMANcc901846ba3e
diff --git a/docs/langchain.md b/docs/langchain.mdindex 0c0b6768..f8b883df 100644--- a/docs/langchain.md+++ b/docs/langchain.md@@ -1,6 +1,6 @@# LangChain & LangGraph Integration-Laya provides fast (~33 ms), non-autoregressive decision components for **LangChain** and **LangGraph**:+Laya provides fast, non-autoregressive decision components for **LangChain** and **LangGraph** (single-question latency measured at **32.8 ms** with `laya-multilingual` and **39.5 ms** with `laya` on a Tesla T4 GPU; 193–464 ms on CPU):* **`LayaRouter`**: Conditional edge and branch router with confidence fallback gating.* **`LayaGuardrail`**: Sub-40ms inline screening for prompt injections, jailbreaks, and sensitive data.@@ -14,14 +14,16 @@ Supports both **local in-process inference** (`Agent` or `Router`) and **remote## Installation```bash-pip install "laya[langchain]"+pip install "laya[langchain]" # Installs both langchain-core and langgraph+# or+pip install "laya[langgraph]"```---## 1. LangGraph Conditional Edge Routing-In LangGraph, conditional edges determine which node executes next. Autoregressive LLMs take 500–2,000 ms to make this decision. `LayaRouter` runs in **~33 ms**:+In LangGraph, conditional edges determine which node executes next. Autoregressive LLMs take 500–2,000 ms to make this decision. `LayaRouter` runs in **~33 ms** (measured at 32.8 ms on `laya-multilingual` / 39.5 ms on `laya` English on a Tesla T4 GPU):```pythonfrom typing import TypedDictdiff --git a/examples/langchain_quickstart.py b/examples/langchain_quickstart.pyindex 34e5c36c..bc46ccdc 100644--- a/examples/langchain_quickstart.py+++ b/examples/langchain_quickstart.py@@ -91,3 +91,41 @@ print(f"Churn Risk: {enriched_state['triage']['churn_risk']}")# api_key="optional-secret-key",# criteria={...}# )+++# =====================================================================+# 5. Compiled LangGraph StateGraph Workflow+# =====================================================================+try:+ from langgraph.graph import StateGraph, END++ class AgentState(TypedDict):+ input: str+ response: str++ workflow = StateGraph(AgentState)+ workflow.add_node("billing_agent", lambda state: {"response": "Routing to Billing Specialist."})+ workflow.add_node("technical_agent", lambda state: {"response": "Routing to Technical Support Specialist."})+ workflow.add_node("sales_agent", lambda state: {"response": "Routing to Enterprise Sales Representative."})+ workflow.add_node("human_agent", lambda state: {"response": "Routing to Human Escalation Tier."})++ workflow.set_conditional_entry_point(+ router_node,+ {+ "billing_agent": "billing_agent",+ "technical_agent": "technical_agent",+ "sales_agent": "sales_agent",+ "human_agent": "human_agent",+ }+ )+ workflow.add_edge("billing_agent", END)+ workflow.add_edge("technical_agent", END)+ workflow.add_edge("sales_agent", END)+ workflow.add_edge("human_agent", END)++ app = workflow.compile()+ graph_res = app.invoke({"input": "I was charged twice on invoice #9821."})+ print("\nLangGraph StateGraph Result:", graph_res["response"])+except (ImportError, Exception) as e:+ print(f"\nNote: To run the compiled LangGraph StateGraph workflow, install 'laya[langchain]': {e}")+diff --git a/pyproject.toml b/pyproject.tomlindex a18490ca..82669ef5 100644--- a/pyproject.toml+++ b/pyproject.toml@@ -53,6 +53,11 @@ onnx = ["onnx", "onnxruntime"]# `pip install laya[langchain]` adds the LangChain / LangGraph integrations (laya.integrations).langchain = ["langchain-core>=0.2.0",+ "langgraph>=0.1.0",+]+langgraph = [+ "langchain-core>=0.2.0",+ "langgraph>=0.1.0",][project.scripts]diff --git a/tests/test_langchain.py b/tests/test_langchain.pyindex c309accf..ad12bf45 100644--- a/tests/test_langchain.py+++ b/tests/test_langchain.py@@ -114,6 +114,11 @@ check("router/fallback_on_low_confidence", router.invoke("lowconf question"), "h# LangGraph callable protocolcheck("router/callable_protocol", router({"messages": [DummyMessage("human", "refund please")]}), "billing")+# LangGraph conditional edge mapping simulation+mapping = {"billing": "BillingNode", "technical": "TechNode", "human_agent": "HumanNode"}+check("router/langgraph_edge_routing", mapping[router({"input": "I need an invoice refund"})], "BillingNode")+check("router/langgraph_edge_fallback", mapping[router({"input": "lowconf question"})], "HumanNode")+# --------------------------------------------------------------- 3. LayaGuardraildef mock_guard_response(state, questions):