Downloads · 30 days
15
14% of all-time downloads
yammi/frago-router-qwen3-1.7b
frago-router-qwen3-1.7b is a text generation model from yammi. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
A narrow-purpose model that routes Chinese natural-language instructions to frago CLI commands. LoRA supervised fine-tune (SFT) of Qwen3-1.7B.
Downloads · 30 days
15
14% of all-time downloads
All-time downloads
106
Public
Parameters
1.7B
5.3 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors3.4 GB · 65%
From the Hugging Face model README
A narrow-purpose model that routes Chinese natural-language instructions to frago CLI commands. LoRA supervised fine-tune (SFT) of Qwen3-1.7B.
frago is an agent runtime / automation infrastructure invoked as uv run frago <command> (browser automation, unified timeline, task board, recipes, knowledge base, etc.). A general LLM, asked to "search something" or "show what happened recently", defaults to WebSearch or writing ad-hoc curl/python — not frago. This model internalizes the instinct to reach for frago commands into the weights, so the agent runtime no longer needs to inject routing rules via hooks on every session.
Input: one Chinese user instruction.
Output: a ```bash code block containing the matching uv run frago ... command, with arguments filled in from the instruction.
It covers ~22 frago commands, e.g.:
frago chrome navigate "https://www.google.com/search?q=<query>"frago timeline view --recent 24hfrago task list / frago task activefrago recipe run|list|infofrago todo list / frago todo add "..."frago book <topic>frago def list / frago run insights --query "..."The decisive test is without any system prompt — if frago routing only worked when a "use frago" system prompt is present, it would not be internalized. Below are the model's actual greedy-decoding outputs on six held-out instructions.
Without system prompt (the real test):
| Instruction (zh) | Output | Verdict |
|---|---|---|
| 查一下今天 BTC 的价格 (check BTC price) | uv run frago chrome navigate "https://www.google.com/search?q=BTC价格" | ✅ correct + real value filled |
| 看看最近 24 小时发生了什么 (what happened in 24h) | uv run frago timeline view --recent 24h | ✅ |
| 帮我搜一下 Qwen3 的发布说明 (search Qwen3 release notes) | uv run frago chrome navigate "https://www.google.com/search?q=Qwen3发布说明" | ✅ value filled |
| 现在有哪些任务在进行 (what tasks are running) | uv run frago task list | ✅ |
| 把这条经验记到知识库:T4 不支持 bf16 (save this lesson) | uv run frago run insights --add "T4 不支持 bf16" | ⚠️ right command + content, imprecise flag (--add vs --save --payload) |
| 我的待办还有哪些 (my todos) | uv run frago todo list | ✅ |
All six reach for frago (no curl/python fallback), and arguments are filled with real values copied from the instruction rather than placeholders. Behavior with a frago-oriented system prompt is equivalent.
An earlier iteration of this dataset got these same prompts wrong, and the two failures pinpoint exactly which design choices matter:
| Instruction | Earlier iteration (no fixes) | This model |
|---|---|---|
| 帮我搜一下 Qwen3 的发布说明 | chrome navigate "...q=<关键词>" (literal placeholder, never filled) | chrome navigate "...q=Qwen3发布说明" (real value) |
| 现在有哪些任务在进行 (no system prompt) | chrome navigate "<网址>" — wrong command, literal placeholder | task list |
| 查一下今天 BTC 的价格 (no system prompt) | python3 .../eth-price-checker.py — abandons frago entirely | chrome navigate "...q=BTC价格" |
Two fixes turned the failures into the passes above:
<placeholder> tokens). Training on placeholders teaches the model to echo the symbol; training on filled values teaches it to copy the relevant span from the instruction.thread group was removed and folded into timeline/task). This model reflects the command surface at one point in time.transformers:
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
m = "yammi/frago-router-qwen3-1.7b"
tok = AutoTokenizer.from_pretrained(m)
model = AutoModelForCausalLM.from_pretrained(m, torch_dtype=torch.float16, device_map="auto")
msgs = [{"role": "user", "content": "帮我搜一下 Qwen3 的发布说明"}]
inp = tok.apply_chat_template(msgs, add_generation_prompt=True, return_tensors="pt",
return_dict=True, enable_thinking=False).to(model.device)
out = model.generate(**inp, max_new_tokens=128, do_sample=False)
print(tok.decode(out[0][inp["input_ids"].shape[1]:], skip_special_tokens=True))
# -> ```bash\nuv run frago chrome navigate "https://www.google.com/search?q=Qwen3发布说明"\n```
A companion Q8_0 GGUF (~1.8GB) is included for llama.cpp / Ollama:
ollama create frago-router -f Modelfile # Modelfile: FROM ./frago-qwen3-1.7b-v4-q8_0.gguf
ollama run frago-router "我的待办还有哪些"
Qwen/Qwen3-1.7Brun insights --add instead of --save --payload).Qwen3-1.7B is Apache-2.0; this derivative is released under Apache-2.0.