Instructions to use continker/Qwen3.5-2B-metro-v23 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries
PEFT
How to use continker/Qwen3.5-2B-metro-v23 with PEFT:
```
Task type is invalid.
```

How to use continker/Qwen3.5-2B-metro-v23 with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="continker/Qwen3.5-2B-metro-v23",
	filename="Qwen3.5-2B-metro-v23-Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use continker/Qwen3.5-2B-metro-v23 with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Use Docker

docker model run hf.co/continker/Qwen3.5-2B-metro-v23:Q4_K_M

LM Studio
Jan

vLLM

How to use continker/Qwen3.5-2B-metro-v23 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "continker/Qwen3.5-2B-metro-v23"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "continker/Qwen3.5-2B-metro-v23",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/continker/Qwen3.5-2B-metro-v23:Q4_K_M

Ollama
How to use continker/Qwen3.5-2B-metro-v23 with Ollama:
```
ollama run hf.co/continker/Qwen3.5-2B-metro-v23:Q4_K_M
```

Unsloth Studio

How to use continker/Qwen3.5-2B-metro-v23 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for continker/Qwen3.5-2B-metro-v23 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for continker/Qwen3.5-2B-metro-v23 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for continker/Qwen3.5-2B-metro-v23 to start chatting

How to use continker/Qwen3.5-2B-metro-v23 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "continker/Qwen3.5-2B-metro-v23:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use continker/Qwen3.5-2B-metro-v23 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf continker/Qwen3.5-2B-metro-v23:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default continker/Qwen3.5-2B-metro-v23:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use continker/Qwen3.5-2B-metro-v23 with Docker Model Runner:
```
docker model run hf.co/continker/Qwen3.5-2B-metro-v23:Q4_K_M
```

Lemonade

How to use continker/Qwen3.5-2B-metro-v23 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull continker/Qwen3.5-2B-metro-v23:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-2B-metro-v23-Q4_K_M

List all available models

lemonade list

Qwen3.5-2B + metro-v23 LoRA

Domain-specialised tool-using agent for transit-kiosk tasks: routing, fare calculation, disruption advisories, accessibility, multilingual cultural notes, multi-turn context tracking, and policy adaptation across 6 metro systems (MARTA, BART, CTA, Doha, Taipei MRT, Beijing Subway).

QLoRA r=16 fine-tune of Qwen/Qwen3.5-2B on 790 distilled traces from Qwen3.5-27B and Qwen3.5-35B-A3B teachers (filtered to tier1 ≥ 90% per case, deduplicated by case_id, evaluated on the MetroLLM-Bench v23 harness).

Files

File	Purpose
`Qwen3.5-2B-metro-v23-Q4_K_M.gguf` (1.2 GB)	Runtime artifact for llama.cpp / Ollama
`adapter/`	Raw LoRA adapter (use with PEFT + base Qwen3.5-2B)
`training_summary.json`	Hyperparameters, seed, dataset version

Eval (v23, 6 systems, Haiku judge for Tier 2)

Cross-system average: Tier-1 81.5, Composite 80.1 (+6.2 T1 / +6.9 Comp vs base Qwen3.5-2B)

System	Tier-1 %
MARTA	84.0
BART	80.7
CTA	82.7
DOHA	81.0
TAIPEI	80.5
BEIJING	80.1

Quickstart (llama.cpp)

huggingface-cli download continker/Qwen3.5-2B-metro-v23 \
  Qwen3.5-2B-metro-v23-Q4_K_M.gguf --local-dir ./models

llama-server -m ./models/Qwen3.5-2B-metro-v23-Q4_K_M.gguf \
  --port 8080 --ctx-size 32768 --n-gpu-layers 999

Quickstart (PEFT adapter, Python)

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

base = AutoModelForCausalLM.from_pretrained("Qwen/Qwen3.5-2B", torch_dtype="bfloat16")
model = PeftModel.from_pretrained(base, "continker/Qwen3.5-2B-metro-v23", subfolder="adapter")
tokenizer = AutoTokenizer.from_pretrained("continker/Qwen3.5-2B-metro-v23", subfolder="adapter")

Training

Base: Qwen/Qwen3.5-2B
Method: QLoRA, rank=16, alpha=32, dropout=0.05
Targets: q/k/v/o + gate/up/down projections
Optimizer: AdamW, lr=2e-4, cosine, warmup 5%
Epochs: 3, effective batch 8 (per_device_train_batch_size=2 × grad_accum=4)
Max sequence length: 4096
Seed: 42 (default; multi-seed CI in progress for 27B)
Dataset: 790 distilled examples, see continker/metrollm-bench-train-data-v23

Limitations

Trained on 6 metro systems; generalisation to other systems untested.
Tool-use schema is specific to the MetroLLM-Bench mock server (route_planner, fare_calculator, station_info, disruption_feed, knowledge_base, submit_assistant_state).
Quantised to 4-bit (Q4_K_M); for full-precision behaviour use the adapter on bf16 base weights.

Citation

@misc{metrollm-bench-2026,
  title={MetroLLM-Bench: Evaluating LLMs as Prompt-Driven Transit Kiosk Agents},
  author={Hendriks, Remco and contributors},
  year={2026},
  publisher={HuggingFace},
  howpublished={\url{https://huggingface.co/continker}}
}

Downloads last month: 5

GGUF

Model size

2B params

Architecture

qwen35

Hardware compatibility

4-bit

Model tree for continker/Qwen3.5-2B-metro-v23

Base model

Qwen/Qwen3.5-2B-Base

Finetuned

Qwen/Qwen3.5-2B

Adapter

(92)

this model