File size: 6,250 Bytes
f786ba3 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 | """Hub Scout — an AGENTS.md-driven agent.
The agent's entire instruction set lives in AGENTS.md, loaded verbatim at
startup and passed as the system prompt. app.py only wires tools and the
chat loop; edit AGENTS.md to change behavior.
Run locally:
export HF_TOKEN=hf_... # token with Inference access
pip install -r requirements.txt
python app.py # opens http://127.0.0.1:7860
"""
import json
import os
import gradio as gr
from huggingface_hub import HfApi, InferenceClient
HERE = os.path.dirname(os.path.abspath(__file__))
MODEL_ID = os.environ.get("AGENT_MODEL", "Qwen/Qwen2.5-72B-Instruct")
HF_TOKEN = os.environ.get("HF_TOKEN")
with open(os.path.join(HERE, "AGENTS.md"), encoding="utf-8") as f:
AGENTS_MD = f.read()
assert AGENTS_MD.strip(), "AGENTS.md is empty — the agent would have no instructions"
client = InferenceClient(model=MODEL_ID, provider="auto", token=HF_TOKEN)
hf_api = HfApi(token=HF_TOKEN)
TOOLS = [
{
"type": "function",
"function": {
"name": "search_hf_models",
"description": (
"Search the Hugging Face Hub for model repos matching keywords, "
"ranked by likes. Returns repo ids with likes and downloads."
),
"parameters": {
"type": "object",
"properties": {
"query": {
"type": "string",
"description": "Search keywords, e.g. 'multilingual embeddings'",
},
"limit": {
"type": "integer",
"description": "How many results to return (default 5, max 10)",
},
"task": {
"type": "string",
"description": (
"Optional pipeline_tag filter, e.g. 'text-generation', "
"'feature-extraction', 'text-classification'"
),
},
},
"required": ["query"],
},
},
},
{
"type": "function",
"function": {
"name": "get_model_stats",
"description": (
"Get current stats for one model repo: downloads, likes, "
"pipeline tag and last-modified date."
),
"parameters": {
"type": "object",
"properties": {
"repo_id": {
"type": "string",
"description": "Model repo id, e.g. 'Qwen/Qwen2.5-72B-Instruct'",
}
},
"required": ["repo_id"],
},
},
},
]
def _search_hf_models(query, limit=5, task=None):
limit = max(1, min(int(limit or 5), 10))
kwargs = {"search": query, "limit": limit, "sort": "likes", "direction": -1}
if task:
kwargs["pipeline_tag"] = task
lines = [
f"- {m.id} (likes={getattr(m, 'likes', '?')}, downloads={getattr(m, 'downloads', '?')})"
for m in hf_api.list_models(**kwargs)
]
return "\n".join(lines) if lines else "No models matched that query."
def _get_model_stats(repo_id):
info = hf_api.model_info(repo_id)
return (
f"{info.id}: pipeline_tag={getattr(info, 'pipeline_tag', None)}, "
f"downloads={getattr(info, 'downloads', None)}, likes={getattr(info, 'likes', None)}, "
f"lastModified={getattr(info, 'lastModified', None)}"
)
def run_tool(name, arguments):
try:
args = json.loads(arguments or "{}")
except json.JSONDecodeError:
return f"Invalid JSON arguments: {arguments!r}"
try:
if name == "search_hf_models":
return _search_hf_models(args["query"], args.get("limit", 5), args.get("task"))
if name == "get_model_stats":
return _get_model_stats(args["repo_id"])
return f"Unknown tool: {name}"
except Exception as exc:
return f"Tool '{name}' failed: {type(exc).__name__}: {exc}"
MAX_TOOL_ROUNDS = 3
def run_agent(message, history):
"""history is OpenAI-style dicts: {"role": "user"|"assistant", "content": str}."""
messages = [{"role": "system", "content": AGENTS_MD}]
messages.extend(history)
messages.append({"role": "user", "content": message})
reply = ""
msg = None
for _ in range(MAX_TOOL_ROUNDS):
resp = client.chat_completion(
messages=messages, tools=TOOLS, tool_choice="auto", max_tokens=1024
)
msg = resp.choices[0].message
calls = getattr(msg, "tool_calls", None) or []
if not calls:
reply = msg.content or ""
break
# Record the assistant's tool-call turn, run each tool, feed results back.
messages.append(
{
"role": "assistant",
"content": msg.content or "",
"tool_calls": [
{
"id": tc.id,
"type": "function",
"function": {
"name": tc.function.name,
"arguments": tc.function.arguments,
},
}
for tc in calls
],
}
)
for tc in calls:
result = run_tool(tc.function.name, tc.function.arguments)
messages.append({"role": "tool", "tool_call_id": tc.id, "content": result})
else:
reply = msg.content or "(I stopped after 3 tool rounds.)"
return reply
demo = gr.ChatInterface(
fn=run_agent,
title="Hub Scout — an AGENTS.md agent",
description=(
f"Instructions live in <code>AGENTS.md</code> and are loaded verbatim at startup. "
f"Model: <code>{MODEL_ID}</code>. Tools: search_hf_models, get_model_stats."
),
examples=[
"Find me a small chat model that runs on a laptop CPU.",
"What's the most downloaded sentence-embedding model right now?",
"How actively used is Qwen/Qwen2.5-72B-Instruct?",
],
)
if __name__ == "__main__":
demo.launch() |