Instructions to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with jev-style:

pip install jev-style
# GGUF builds score through llama.cpp: build the jev-score binary once
hf download chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF build_jev_score.sh jev_score.cpp --local-dir jev-score
export JEV_SCORE_BIN=$(sh jev-score/build_jev_score.sh /path/to/llama.cpp | tail -n 1)

from jev_style import JevStyle, noul, choice

js = JevStyle.from_pretrained("chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF")
out = js.decide("I was charged twice for one order.", {
    "billing": noul("This message is about billing."),
    "team": choice("Which team should handle it?", ["billing", "shipping", "tech"]),
})
print(out["answers"]["team"]["choice"])

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF: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 chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF: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 chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M

Use Docker

docker model run hf.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M

LM Studio
Jan

vLLM

How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M

Ollama
How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with Ollama:
```
ollama run hf.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
```
Unsloth Desktop
Docker Model Runner
How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with Docker Model Runner:
```
docker model run hf.co/chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M
```

Lemonade

How to use chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull chaoliangUNSW/Jev-Style-Qwen3.5-2B-Decision-v2-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Jev-Style-Qwen3.5-2B-Decision-v2-GGUF-Q4_K_M

List all available models

lemonade list

Atomic Chat

Jev-Style-Qwen3.5-2B-Decision-v2-GGUF / evaluation /quantization_summary.json

chaoliangUNSW

Map current GGUF filenames to validated build names in quantization_summary.json

e9a1bbe verified 10 days ago

Raw History Blame Contribute Delete

5.84 kB

	{
	"evaluation_decisions": 500,
	"comparison": "same frozen subset vs CUDA merged BF16",
	"metric": "real-label task-macro accuracy; teacher-reference decisions excluded from this macro",
	"variants": {
	"Q4_K_M": {
	"filename": "Jev-Style-v2-Calibrated-Q4_K_M.gguf",
	"build_filename": "Jev-Style-v2-Q4_K_M-Calibrated.gguf",
	"weight_bytes": 1274388384,
	"sha256": "c697d3b29d07fdd37b6ebeb5c98066f4c31632162adca6258db23f75184eb0c4",
	"argmax_agreement": 0.914,
	"real_label_macro_accuracy": 0.7818348025857907,
	"real_label_macro": {
	"accuracy": 0.7818348025857907,
	"macro_f1": 0.764930756852403,
	"nll": 0.6474576399658714,
	"brier": 0.3418843970562367,
	"ece": 0.17704328656854143
	},
	"calibration_n": 3100,
	"temperature_folded": true,
	"runtime_temperature": 1.0,
	"validation": {
	"format": "Q4_K_M",
	"n": 500,
	"argmax_agreement": 0.914,
	"cuda_same_subset": {
	"accuracy": 0.7910177949703642,
	"macro_f1": 0.7760857891780776,
	"nll": 0.6000106706289864,
	"brier": 0.317876961372947,
	"ece": 0.16696329399611096
	},
	"gguf_same_subset": {
	"accuracy": 0.7818348025857907,
	"macro_f1": 0.764930756852403,
	"nll": 0.6474576399658714,
	"brier": 0.3418843970562367,
	"ece": 0.17704328656854143
	},
	"accuracy_difference": -0.00918299238457343,
	"nll_difference": 0.04744696933688497,
	"temperature_folded": true,
	"runtime_temperature": 1.0,
	"calibration_n": 3100,
	"criteria": {
	"agreement_min": 0.9,
	"accuracy_loss_max": 0.03,
	"nll_increase_max": 0.1
	},
	"calibration_backend": "llama.cpp batched, checked against native readout",
	"evaluation_backend": "llama.cpp native single-sequence exact option logits",
	"weight_bytes": 1274388384,
	"passed": true
	}
	},
	"Q8_0": {
	"filename": "Jev-Style-v2-Calibrated-Q8_0.gguf",
	"build_filename": "Jev-Style-v2-Q8_0-Calibrated.gguf",
	"weight_bytes": 2012004256,
	"sha256": "5c2aa0d35b24a27f03228b2c62ebaaebd9b5b785844634d4217278d822751494",
	"argmax_agreement": 0.992,
	"real_label_macro_accuracy": 0.7868855635654054,
	"real_label_macro": {
	"accuracy": 0.7868855635654054,
	"macro_f1": 0.773615445189482,
	"nll": 0.6015417570028033,
	"brier": 0.318930434743987,
	"ece": 0.16630794459650552
	},
	"calibration_n": 3100,
	"temperature_folded": true,
	"runtime_temperature": 1.0,
	"validation": {
	"n": 500,
	"argmax_agreement": 0.992,
	"cuda_same_subset": {
	"accuracy": 0.7910177949703642,
	"macro_f1": 0.7760857891780776,
	"nll": 0.6000106706289864,
	"brier": 0.317876961372947,
	"ece": 0.16696329399611096
	},
	"q8_same_subset": {
	"accuracy": 0.7868855635654054,
	"macro_f1": 0.773615445189482,
	"nll": 0.6015417570028033,
	"brier": 0.318930434743987,
	"ece": 0.16630794459650552
	},
	"accuracy_difference": -0.004132231404958775,
	"nll_difference": 0.0015310863738169367,
	"temperature_folded": true,
	"passed": true,
	"practical_deployment_gate_passed": true,
	"initial_strict_accuracy_target_met_on_subset": false,
	"initial_accuracy_loss_target": 0.003,
	"note": "500-example subset check: 99.2% agreement. Observed task-macro loss is 0.413 pp, so the initial 0.3 pp accuracy goal is not met on this subset. Prefer BF16/MLX when accuracy is the priority; this is not a bound on population degradation."
	}
	},
	"BF16": {
	"filename": "Jev-Style-v2-Calibrated-BF16.gguf",
	"build_filename": "Jev-Style-v2-BF16-Calibrated.gguf",
	"weight_bytes": 3775700896,
	"sha256": "8baa111eec6e30a5c9e97d559b53127a19ef30171e8aed26673153b4f77adcbb",
	"argmax_agreement": 0.996,
	"real_label_macro_accuracy": 0.7936151975677668,
	"real_label_macro": {
	"accuracy": 0.7936151975677668,
	"macro_f1": 0.77850284511292,
	"nll": 0.6007551256850553,
	"brier": 0.31813150017828207,
	"ece": 0.16639447171288832
	},
	"calibration_n": 3100,
	"temperature_folded": true,
	"runtime_temperature": 1.0,
	"validation": {
	"format": "BF16",
	"n": 500,
	"argmax_agreement": 0.996,
	"cuda_same_subset": {
	"accuracy": 0.7910177949703642,
	"macro_f1": 0.7760857891780776,
	"nll": 0.6000106706289864,
	"brier": 0.317876961372947,
	"ece": 0.16696329399611096
	},
	"gguf_same_subset": {
	"accuracy": 0.7936151975677668,
	"macro_f1": 0.77850284511292,
	"nll": 0.6007551256850553,
	"brier": 0.31813150017828207,
	"ece": 0.16639447171288832
	},
	"accuracy_difference": 0.0025974025974025983,
	"nll_difference": 0.0007444550560689045,
	"temperature_folded": true,
	"runtime_temperature": 1.0,
	"calibration_n": 3100,
	"criteria": {
	"agreement_min": 0.99,
	"accuracy_loss_max": 0.005,
	"nll_increase_max": 0.015
	},
	"calibration_backend": "llama.cpp batched, checked against native readout",
	"evaluation_backend": "llama.cpp native single-sequence exact option logits",
	"weight_bytes": 3775700896,
	"passed": true
	}
	}
	},
	"filename_note": "filename is the current published name; build_filename is the name used when the file was validated (renamed 2026-09-24 for Ollama quant tags; bytes and sha256 unchanged)."
	}