CoolFace
Apppublic

build-small-hackathon/MiniCPM5-1B-Agent

sourceHugging Faceupdated 3mo agoView on Hugging Face
3likes
App README

🛠️ MiniCPM5-1B-Agent

A tiny agentic coding agent that runs the whole write → run → read → debug → verify loop on a free CPU.

Social media post link; demo video: <table><tr> <td><img src="minicpm5-1b-agent-demo.gif" alt="MiniCPM5-1B-Agent demo GIF" width="480"></td> <td><video src="https://huggingface.co/spaces/build-small-hackathon/MiniCPM5-1B-Agent/resolve/main/longdemobuild-small-hackathonMiniCPM5-1B-Agent16x.mp4" controls width="480"></video></td> </tr></table>

A full fine-tune of `openbmb/MiniCPM5-1B` (1B params), served as a Q80 GGUF on llama.cpp, no GPU. Give it a task; it reasons in `<think>`, then uses `bash` / `write` / `read` / `edit` in a sandbox to build, run, and fix code, and renders the result (charts, images, live HTML) inline in the chat. Multi-turn: files and history persist across messages. It is also exposed as an **MCP tool** (`runcodingtask` at `/gradioapi/mcp/`).

What it is

Most coding agents are 70B+ behind a cloud API. This is the opposite: a 1B model doing the real agentic loop on a 2-vCPU CPU Space, no GPU. It writes a file, runs it, reads the output, debugs, and shows you the artifact, the same loop a big agent runs, shrunk to something you could host in your own backyard.

How it was built

  • —Modal platform was used to evaluate the model via inference.
  • —Data (`train_v4`, 45,762 rows): the proven v2 backbone (retail-filtered teacher mixes + real-usage agent traces) kept whole, plus ~3,538 curated additions, gated to a small served tool vocab and solution-aware MinHash-deduped. Bundled on the model repo under dataset/.
  • —SFT: full fine-tune (not LoRA, the long agentic mix needs the capacity) of the abliterated base, 1 epoch, 24k context, fits in ~15-18 GB VRAM (direct Liger fused cross-entropy + mem-efficient SDPA).
  • —DPO (on-policy): run the SFT model over the training prompts and capture its OWN behaviour. chosen = a valid <function> tool call; rejected = its real miss (rambling in <think> / answering with no call). ~649 pairs. This rewards ACTING over stalling.
  • —Serving: Q8_0 GGUF on llama.cpp; a two-phase decode bounds the <think> separately from the action so the model acts instead of looping; produced files render inline (charts, images, sandboxed live-HTML iframes).

Try it

  • —"Write a Python script that makes a bar chart of 30, 45, 25 (A, B, C), save chart.png, then run it." → writes + runs it; the PNG renders inline.
  • —"Write an HTML page quote.html with a button that shows a random quote each click (hard-coded, no internet)." → writes the file; renders live in a sandboxed iframe.

It is a tiny 1B on a free CPU: expect ~4 min per simple turn, longer for multi-step tasks (the demo video shows it working end-to-end, so it can be judged even if a live run is slow).

<details><summary><b>Output examples</b></summary>

Real single-turn outputs from the Q8_0 GGUF, generated on Modal (CPU):

Prompt: Write a Python function fib(n) that returns the nth Fibonacci number iteratively. Print fib(10).

python
def fib(n):
    if n == 0:
        return 0
    elif n == 1:
        return 1
    a, b = 0, 1
    for _ in range(2, n + 1):
        a, b = b, a + b
    return b

print(fib(10))

Prompt: Convert an integer (1-3999) to a Roman numeral; demonstrate on 1994.

python
def int_to_roman(num):
    if not 1 <= num <= 3999:
        raise ValueError("Number must be between 1 and 3999")
    roman_map = [(1000,'M'),(900,'CM'),(500,'D'),(400,'CD'),(100,'C'),(90,'XC'),
                 (50,'L'),(40,'XL'),(10,'X'),(9,'IX'),(5,'V'),(4,'IV'),(1,'I')]
    result = []
    for value, symbol in roman_map:
        while num >= value:
            result.append(symbol); num -= value
    return ''.join(result)

print(int_to_roman(1994))  # -> MCMXCIV

Prompt: How many $40 games can I buy with $200, and how much is left over?

You can buy 5 games with $200, and have $0 left over. Number of games = 200 / 40 = 5; remaining = 200 - (5 x 40) = $0.

</details>

Model, dataset & full reproduction

→ [Luminia/MiniCPM5-1B-Agent-GGUF](https://huggingface.co/Luminia/MiniCPM5-1B-Agent-GGUF) (model card = the full data mix, SFT/DPO recipe, eval, and exact reproduce commands; v4 dataset bundled under dataset/).

💻 Code on GitHub: Katehuuh/MiniCPM5-1B-Agent (the Space + the full training pipeline; code reviewed with OpenAI Codex).

Built for the Build Small Hackathon · OpenBMB + OpenAI Codex tracks.