Explain tiktoken library details

README_BEDROCK.md

@@ -0,0 +1,57 @@
+# AWS Bedrock Agent Examples
+
+Two approaches are included:
+
+- Converse API lightweight agent with tool use: `bedrock_converse_agent.py`
+- Managed Agents for Bedrock scripts: `bedrock_managed_agents.py`
+
+## Setup
+
+1) Python deps
+
+```bash
+python -m venv .venv && source .venv/bin/activate
+pip install -r requirements.txt
+```
+
+2) AWS credentials
+
+- Configure credentials with permissions for `bedrock` and `bedrock-runtime`.
+- Set `AWS_REGION` or `AWS_DEFAULT_REGION`.
+- Optionally set `BEDROCK_MODEL_ID` to override the default model.
+
+## Converse API agent
+
+Run a prompt through a small agent loop with a demo tool `get_weather`.
+
+```bash
+python bedrock_converse_agent.py --prompt "What's the weather in London in F?"
+```
+
+Options:
+
+- `--model-id`: override model (default: `anthropic.claude-3-5-sonnet-20240620-v1:0`)
+- `--region`: AWS region override
+
+## Managed Agents for Bedrock
+
+Create, prepare, and invoke a managed Agent.
+
+```bash
+# Create agent
+python bedrock_managed_agents.py create-agent \
+  --name demo-agent \
+  --foundation-model anthropic.claude-3-5-sonnet-20240620-v1:0 \
+  --instruction "You are a helpful assistant."
+
+# Prepare (deploy) agent
+python bedrock_managed_agents.py prepare-agent --agent-id <AGENT_ID>
+
+# Invoke agent
+python bedrock_managed_agents.py invoke-agent --agent-id <AGENT_ID> --input-text "Hello"
+```
+
+Notes:
+
+- Ensure your account has access to the chosen foundation model.
+- For production, add tools/knowledge bases/guardrails to the managed agent as needed.

bedrock_converse_agent.py

@@ -0,0 +1,166 @@
+import json
+import os
+from typing import Any, Dict, List, Callable, Optional, Tuple
+
+import boto3
+from botocore.config import Config
+import click
+from rich.console import Console
+from rich.panel import Panel
+
+
+console = Console()
+
+
+def get_bedrock_runtime_client(region_name: Optional[str] = None):
+	region = region_name or os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION")
+	if not region:
+		raise RuntimeError("AWS region not set. Provide --region or set AWS_REGION/AWS_DEFAULT_REGION.")
+	return boto3.client("bedrock-runtime", region_name=region, config=Config(retries={"max_attempts": 10}))
+
+
+def get_default_model_id() -> str:
+	# Popular high-quality model on Bedrock; change if your account lacks access
+	return os.getenv("BEDROCK_MODEL_ID", "anthropic.claude-3-5-sonnet-20240620-v1:0")
+
+
+class Tool:
+	def __init__(self, name: str, description: str, json_schema: Dict[str, Any], impl: Callable[[Dict[str, Any]], Dict[str, Any]]):
+		self.name = name
+		self.description = description
+		self.json_schema = json_schema
+		self.impl = impl
+
+	def to_tool_spec(self) -> Dict[str, Any]:
+		return {
+			"toolSpec": {
+				"name": self.name,
+				"description": self.description,
+				"inputSchema": {"json": self.json_schema},
+			}
+		}
+
+
+def get_weather_impl(args: Dict[str, Any]) -> Dict[str, Any]:
+	city = str(args.get("city", ""))
+	unit = str(args.get("unit", "c")).lower()
+	if unit not in {"c", "f"}:
+		unit = "c"
+	# Dummy data; in real use, call a weather API here
+	base_temp_c = 22.3
+	if unit == "f":
+		temp = base_temp_c * 9 / 5 + 32
+	else:
+		temp = base_temp_c
+	return {"city": city, "unit": unit, "temperature": round(temp, 1), "conditions": "Sunny"}
+
+
+def build_tool_registry() -> Dict[str, Tool]:
+	return {
+		"get_weather": Tool(
+			name="get_weather",
+			description="Get current weather for a city.",
+			json_schema={
+				"type": "object",
+				"properties": {
+					"city": {"type": "string", "description": "City name, e.g., London"},
+					"unit": {"type": "string", "enum": ["c", "f"], "default": "c"},
+				},
+				"required": ["city"],
+			},
+			impl=get_weather_impl,
+		),
+	}
+
+
+def extract_tool_uses_from_content(content: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
+	tool_uses: List[Dict[str, Any]] = []
+	for block in content:
+		if "toolUse" in block:
+			tool_uses.append(block["toolUse"])
+	return tool_uses
+
+
+def build_tool_results(tool_uses: List[Dict[str, Any]], registry: Dict[str, Tool]) -> List[Dict[str, Any]]:
+	results: List[Dict[str, Any]] = []
+	for tu in tool_uses:
+		name = tu.get("name")
+		tool_use_id = tu.get("toolUseId")
+		inputs = tu.get("input") or {}
+		impl = registry.get(name)
+		if not impl:
+			results.append({
+				"toolResult": {
+					"toolUseId": tool_use_id,
+					"content": [{"text": f"Tool '{name}' not implemented."}],
+					"status": "error",
+				}
+			})
+			continue
+		try:
+			output_json = impl.impl(inputs)
+			results.append({
+				"toolResult": {
+					"toolUseId": tool_use_id,
+					"content": [{"json": output_json}],
+					"status": "success",
+				}
+			})
+		except Exception as exc:  # pragma: no cover (defensive)
+			results.append({
+				"toolResult": {
+					"toolUseId": tool_use_id,
+					"content": [{"text": f"Tool '{name}' failed: {exc}"}],
+					"status": "error",
+				}
+			})
+	return results
+
+
+def run_converse_agent(prompt: str, model_id: Optional[str] = None, region: Optional[str] = None, system_prompt: Optional[str] = None) -> Tuple[str, List[Dict[str, Any]]]:
+	client = get_bedrock_runtime_client(region)
+	model = model_id or get_default_model_id()
+	tools = build_tool_registry()
+	tool_config = {"tools": [t.to_tool_spec() for t in tools.values()]}
+	system_blocks = [{"text": system_prompt or "You are a concise, helpful agent. Use tools when beneficial."}]
+	messages: List[Dict[str, Any]] = [{"role": "user", "content": [{"text": prompt}]}]
+
+	final_text: str = ""
+	for _ in range(5):
+		resp = client.converse(modelId=model, system=system_blocks, toolConfig=tool_config, messages=messages)
+		assistant_msg = resp.get("output", {}).get("message", {})
+		stop_reason = resp.get("stopReason")
+		content = assistant_msg.get("content", [])
+
+		# Gather any assistant text so far
+		for block in content:
+			if "text" in block:
+				final_text += block["text"]
+
+		tool_uses = extract_tool_uses_from_content(content)
+		if tool_uses:
+			results = build_tool_results(tool_uses, tools)
+			messages.append(assistant_msg)
+			messages.append({"role": "user", "content": results})
+			continue
+
+		# If no tool use requested, or completed after tool results
+		if stop_reason in {"end_turn", "max_tokens"} or not tool_uses:
+			break
+
+	return final_text.strip(), messages
+
+
+@click.command()
+@click.option("--prompt", required=True, help="User prompt to send to the agent")
+@click.option("--model-id", default=lambda: get_default_model_id(), show_default=True, help="Bedrock model ID")
+@click.option("--region", default=lambda: os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION") or "", help="AWS region")
+def main(prompt: str, model_id: str, region: str):
+	"""Run a lightweight agent loop on Bedrock using the Converse API."""
+	final_text, _ = run_converse_agent(prompt=prompt, model_id=model_id or None, region=region or None)
+	console.print(Panel.fit(final_text or "(no content)", title="Assistant", subtitle=model_id))
+
+
+if __name__ == "__main__":
+	main()
+

bedrock_managed_agents.py

@@ -0,0 +1,73 @@
+import json
+import os
+from typing import Any, Dict, Optional
+
+import boto3
+from botocore.config import Config
+import click
+from rich.console import Console
+from rich.panel import Panel
+
+
+console = Console()
+
+
+def get_bedrock_clients(region_name: Optional[str] = None):
+	region = region_name or os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION")
+	if not region:
+		raise RuntimeError("AWS region not set. Provide --region or set AWS_REGION/AWS_DEFAULT_REGION.")
+	return (
+		boto3.client("bedrock", region_name=region, config=Config(retries={"max_attempts": 10})),
+		boto3.client("bedrock-runtime", region_name=region, config=Config(retries={"max_attempts": 10})),
+	)
+
+
+@click.group()
+def cli():
+	"""Scripts to create and invoke an Agent for Amazon Bedrock."""
+	pass
+
+
+@cli.command("create-agent")
+@click.option("--name", required=True, help="Agent name")
+@click.option("--foundation-model", required=True, help="Model ID, e.g., anthropic.claude-3-5-sonnet-20240620-v1:0")
+@click.option("--instruction", required=True, help="System prompt / instruction for the agent")
+@click.option("--region", default=lambda: os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION") or "", help="AWS region")
+def create_agent(name: str, foundation_model: str, instruction: str, region: str):
+	bedrock, _ = get_bedrock_clients(region)
+	resp = bedrock.create_agent(
+		agentName=name,
+		foundationModel=foundation_model,
+		instruction=instruction,
+		offGuardrail=False,
+	)
+	agent_id = resp["agent"].get("agentId")
+	console.print(Panel.fit(f"Created agent: {agent_id}", title="Create Agent"))
+	console.print(json.dumps(resp, indent=2))
+
+
+@cli.command("prepare-agent")
+@click.option("--agent-id", required=True, help="Agent ID from create-agent")
+@click.option("--region", default=lambda: os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION") or "", help="AWS region")
+def prepare_agent(agent_id: str, region: str):
+	bedrock, _ = get_bedrock_clients(region)
+	resp = bedrock.prepare_agent(agentId=agent_id)
+	console.print(Panel.fit(f"Preparing agent: {agent_id}", title="Prepare Agent"))
+	console.print(json.dumps(resp, indent=2))
+
+
+@cli.command("invoke-agent")
+@click.option("--agent-id", required=True, help="Agent ID")
+@click.option("--input-text", required=True, help="User input")
+@click.option("--session-id", default=None, help="Optional session ID for continuity")
+@click.option("--region", default=lambda: os.getenv("AWS_REGION") or os.getenv("AWS_DEFAULT_REGION") or "", help="AWS region")
+def invoke_agent(agent_id: str, input_text: str, session_id: Optional[str], region: str):
+	bedrock, bedrock_runtime = get_bedrock_clients(region)
+	resp = bedrock_runtime.invoke_agent(agentId=agent_id, sessionId=session_id or "session-1", inputText=input_text)
+	completion = resp.get("completion")
+	console.print(Panel.fit(completion or "(no content)", title="Agent Response"))
+
+
+if __name__ == "__main__":
+	cli()
+

requirements.txt

@@ -1,3 +1,9 @@
+boto3>=1.34.140
+botocore>=1.34.140
+click>=8.1.7
+rich>=13.7.1
+python-dotenv>=1.0.1
+typing-extensions>=4.12.2
 torch==2.3.1
 torchvision==0.18.1
 diffusers==0.11.1