File size: 4,255 Bytes
7807849 33c8a2e 7807849 f3325f9 7807849 f3325f9 7807849 f3325f9 7807849 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 33c8a2e f3325f9 7807849 f3325f9 7807849 d3046db de5c185 f3325f9 d3046db 7807849 f3325f9 7807849 f3325f9 7807849 |
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 |
import gradio as gr
from backend.council import stage1_collect_responses, stage2_collect_rankings, stage3_synthesize_final_stream
from backend.config import COUNCIL_MODELS, CHAIRMAN_MODEL
async def ask_council(question: str, progress=gr.Progress()):
"""
Ask the LLM Council a question.
The council consists of multiple advanced LLMs (currently: {models}) that:
1. Individually answer the question
2. Rank each other's answers
3. Synthesize a final best answer (Chairman: {chairman})
Args:
question: The user's question to be discussed by the council.
progress: Gradio progress tracker.
Yields:
Status updates and finally the synthesized answer.
""".format(
models=", ".join([m.split("/")[-1] for m in COUNCIL_MODELS]), chairman=CHAIRMAN_MODEL.split("/")[-1]
)
try:
buffer = ""
# Stage 1: Collect individual responses
progress(0.1, desc="Stage 1: Collecting individual responses...")
buffer += "## 🟡 Stage 1: Collecting individual responses from council members...\n\n"
yield buffer
stage1_results = await stage1_collect_responses(question)
if not stage1_results:
buffer += "\n❌ The council failed to generate a response."
yield buffer
return
# Format Stage 1 results
buffer += f"### ✅ Received {len(stage1_results)} responses:\n"
for res in stage1_results:
model_name = res["model"].split("/")[-1]
preview = res["response"][:100].replace("\n", " ") + "..."
buffer += f"- **{model_name}**: {preview}\n"
buffer += "\n---\n\n"
yield buffer
# Stage 2: Collect rankings
progress(0.4, desc="Stage 2: Council members are ranking responses...")
buffer += "## 🟡 Stage 2: Council members are ranking each other's responses...\n\n"
yield buffer
stage2_results, _ = await stage2_collect_rankings(question, stage1_results)
# Format Stage 2 results
buffer += "### ✅ Rankings Collected:\n"
for res in stage2_results:
model_name = res["model"].split("/")[-1]
# Extract just the ranking part if possible, or just say "Ranked"
buffer += f"- **{model_name}** has submitted their rankings.\n"
buffer += "\n---\n\n"
yield buffer
# Stage 3: Synthesize final answer
progress(0.7, desc="Stage 3: Chairman is synthesizing the final answer...")
buffer += "## 🟡 Stage 3: Chairman is synthesizing the final answer...\n\n"
yield buffer
full_response = ""
async for chunk in stage3_synthesize_final_stream(question, stage1_results, stage2_results):
full_response += chunk
yield buffer + full_response
progress(1.0, desc="Complete!")
if not full_response:
buffer += "\n❌ The council failed to generate a final synthesis."
yield buffer
return
# Let's keep the history but mark Stage 3 as done
final_buffer = buffer.replace(
"## 🟡 Stage 3: Chairman is synthesizing the final answer...", "## 🟢 Stage 3: Final Answer"
)
yield final_buffer + full_response
except Exception as e:
yield f"❌ Error consulting the council: {str(e)}"
description = """
An MCP server that consults a council of LLMs to answer questions. [LLM Council](https://github.com/machine-theory/lm-council?tab=readme-ov-file) is a project by Machine Theory
and Andrej Karpathy. This space exposes it as an MCP server so you can use it in your own projchatects.
<img src="https://pbs.twimg.com/media/G6ZZO7ragAAtnCZ?format=jpg" alt="MCP Server" style="width: 300px; height: auto; text-align: center;">
⚠️ We're using 5 models in the council, so it takes a minute to answer.
"""
demo = gr.Interface(
fn=ask_council,
inputs=gr.Textbox(lines=2, placeholder="Ask the council..."),
outputs=gr.Markdown(height=200),
title="LLM Council MCP Server",
description=description,
)
if __name__ == "__main__":
# Launch with mcp_server=True to expose as MCP
demo.launch(mcp_server=True, show_error=True)
|