Qwen/Qwen2.5-1.5B-Instruct (Base Model)
└── AnkitNayak-dev/QJev-1.5B (Flagship Fused 16-bit, 3.08 GB)
├── AnkitNayak-dev/QJev-1.5B-LoRA (PEFT Adapter, 20 MB)
└── AnkitNayak-dev/QJev-1.5B-4bit (NF4 Quantized, 1.15 GB)
Overview • Benchmark Evaluation • Model Family • Quickstart • Serving and Deployment • Architecture • Repository Structure • Citation
Traditional large language models perform decision-making through sequential, autoregressive token generation. This process incurs substantial latency (often 300 to 600 ms) and introduces non-deterministic sampling variance when parsing formatted outputs.
QJev-1.5B implements a direct System 1 decision mechanism by coupling a calibrated Multi-Layer Perceptron (MLP) decision head to the final hidden representation of Qwen/Qwen2.5-1.5B-Instruct. Rather than generating text tokens, the model computes a calibrated probability distribution across up to 10 candidate actions in a single forward pass (~31 ms on consumer hardware).
QJev-1.5B is evaluated against the official JevBench metamorphic test harness (jevbench-mini, 240 evaluation cases, 50 scored relations, 1,468 trials) following standard BenchmarkHeaven submission criteria.
All models evaluated under official strict calibration criteria (tolerance = 0.05):
| Model | Parameters | Official Score (tol=0.05) | Category | Inference Latency | Memory Footprint |
|---|---|---|---|---|---|
EXAONE-1.2B |
1.2B | 47.7% | Sub-2B | ~25 ms | ~2.8 GB |
SargeDev/jev-gate-student-b |
0.49B | 50.2% | Sub-2B (Binary Only) | ~23 ms | ~1.5 GB |
decision-2b |
2.0B | 54.8% | 2B Tier | ~40 ms | ~4.5 GB |
pjev-2b |
2.0B | 56.5% | 2B Tier | ~38 ms | ~4.4 GB |
QJev-1.5B (Ours, Phase 3) |
1.54B | 62.53% | Rank 1 Sub-2B | ~31 ms | 3.4 GB (1.07 GB in 4-bit) |
jiwo-4b |
4.0B | 60.2% | 4B Tier | ~65 ms | ~8.8 GB |
decision-4b-v12 |
4.0B | 60.4% | 4B Tier | ~62 ms | ~8.5 GB |
autotrust/JEV-9B |
9.0B | 72.0% | 9B Tier | ~85 ms | ~18.5 GB |
autotrust/JEV-27B |
27.0B | 76.5% | Datacenter Tier | ~220 ms | ~55.0 GB |
Tolerance specification: The official BenchmarkHeaven evaluation suite strictly uses tolerance = 0.05 for headline Capability scores. Under exploratory relaxed evaluation (tolerance = 0.10), QJev-1.5B achieves 71.50%.
| Dimension | Baseline (Pre-Consistency) | QJev-1.5B (Phase 3 Consistency) | Delta | Formal Property Evaluated |
|---|---|---|---|---|
BAT.isolation |
100.0% | 100.0% | 0.0% | Invariance to batch composition and sample ordering |
REP.type |
77.1% | 83.3% | +6.2% | Invariance across prompt format representations (MCQ, boolean, ordinal) |
REP.invariance |
73.2% | 74.4% | +1.2% | Stability under option key renaming and syntax perturbations |
LOG.bounds |
62.5% | 67.9% | +5.4% | Adherence to probabilistic upper and lower bounds |
CHO.conditioning |
41.7% | 56.9% | +15.2% | Preservation of relative odds upon elimination of irrelevant options |
REP.equivariance |
48.6% | 55.6% | +7.0% | Consistent probability shifts under negation transformations |
MEA.partition |
30.7% | 41.1% | +10.4% | Probability mass conservation under option clustering |
| Overall (Official Strict 0.05) | 58.67% | 62.53% | +3.86% | Standard JevBench Capability Benchmark |
Machine-readable evaluation artifacts are archived under reports/jevbench/.
QJev-1.5B is available in three deployment formats:
| Format | Hugging Face Repository | Disk Size | VRAM Footprint | Recommended Use Case |
|---|---|---|---|---|
| Full Fused (FP16) | AnkitNayak-dev/QJev-1.5B | 3.08 GB | 3.4 GB | Production servers, vLLM, standalone microservices |
| PEFT Adapter | AnkitNayak-dev/QJev-1.5B-LoRA | 20.7 MB | Shared | Multi-tenant deployments caching Qwen2.5-1.5B-Instruct |
| NF4 Quantized | AnkitNayak-dev/QJev-1.5B-4bit | 1.15 GB | 1.07 GB | Resource-constrained edge instances, local workstations |
git clone https://github.com/AnkitNayak-dev/QJev-1.5B.git
cd QJev-1.5B
pip install -r requirements.txt
from qjev import QJev
# Load any model variant directly from Hugging Face Hub
model = QJev.from_pretrained("AnkitNayak-dev/QJev-1.5B")
# Or 4-bit quantized: model = QJev.from_pretrained("AnkitNayak-dev/QJev-1.5B-4bit")
# Or PEFT adapter: model = QJev.from_pretrained("AnkitNayak-dev/QJev-1.5B-LoRA")
# Execute decision inference
result = model.decide(
context="Production database primary replica CPU utilization reached 100% and connection pool is exhausted.",
choices=[
"Trigger immediate automated failover to standby replica",
"Restart the primary database daemon immediately",
"Double connection pool limits without scaling compute",
"Silence alerts and monitor for next 15 minutes"
]
)
print(f"Recommended Action: {result['best_choice']} ({result['confidence']}%)")
for choice, prob in result['probabilities'].items():
print(f" [{prob:>5.2f}%] {choice}")
Recommended Action: Trigger immediate automated failover to standby replica (88.28%)
[88.28%] Trigger immediate automated failover to standby replica
[ 6.54%] Restart the primary database daemon immediately
[ 4.79%] Double connection pool limits without scaling compute
[ 0.45%] Silence alerts and monitor for next 15 minutes
python qjev.py --context "Patient presents with acute chest pain and elevated troponin levels." --choices "Emergency PCI" "Aspirin" "Observation"
serve.py)A production FastAPI service providing REST, Jev protocol, and OpenAI-compatible endpoints:
python serve.py --model AnkitNayak-dev/QJev-1.5B --port 8000
POST /v1/decide - Direct decision endpoint returning structured probabilities.POST /v1/systemone - Official BenchmarkHeaven JEV evaluation protocol.POST /v1/chat/completions - Drop-in OpenAI-compatible wrapper.curl -X POST http://localhost:8000/v1/decide \
-H "Content-Type: application/json" \
-d '{
"context": "Customer requests immediate refund after policy window has expired.",
"options": ["Approve refund", "Issue store credit", "Deny request"]
}'
A preconfigured Modelfile is provided in the repository root:
ollama create qjev-1.5b -f Modelfile
ollama run qjev-1.5b
Querying via the Python SDK:
from qjev import QJev
client = QJev.from_ollama("http://localhost:11434", model="qjev-1.5b")
result = client.decide("Route network traffic to cluster Alpha or Beta?", ["Alpha", "Beta"])
Convert weights to GGUF format for edge CPU and local deployment:
python scripts/export_gguf.py --model-dir export/QJev-1.5B-Phase3 --out-file qjev-1.5b-f16.gguf
./llama-quantize qjev-1.5b-f16.gguf qjev-1.5b-q4_k_m.gguf Q4_K_M
vllm serve AnkitNayak-dev/QJev-1.5B --task classify --port 8000
[Prompt Context + Formatted Candidate Choices]
│
▼
┌────────────────────────┐
│ Qwen2.5-1.5B Backbone │ (QLoRA adapted: q, k, v, o projections)
└───────────┬────────────┘
│ (Final Hidden State: 1536 dim)
▼
┌────────────────────────┐
│ LayerNorm(1536) │
└───────────┬────────────┘
│
▼
┌────────────────────────┐
│ Linear(1536 -> 512) │ (GELU Activation, Dropout 0.1)
└───────────┬────────────┘
│
▼
┌────────────────────────┐
│ Linear(512 -> 10) │ (Dynamic Logit Masking for N Choices)
└───────────┬────────────┘
│
▼
[Calibrated Softmax Probabilities]
Qwen/Qwen2.5-1.5B-InstructSargeDev/jev-distill-corpus-v3 (655,806 distillation examples)├── README.md # Project documentation and benchmarks
├── LICENSE # Apache 2.0 open-source license
├── requirements.txt # Python dependencies
├── Modelfile # Ollama configuration
├── qjev.py # Unified Python SDK and CLI client
├── serve.py # Production FastAPI service
├── predict.py # Forward pass evaluation script
├── handler.py # Hugging Face inference handler
│
├── benchmark/ # Benchmark evaluation suites
│ ├── run_jevbench.py # JevBench test harness runner
│ ├── self_test_benchmarkheaven.py # Official BenchmarkHeaven evaluation runner
│ ├── self_test_phase3.py # Phase 3 verification runner
│ └── self_test_step1.py # Calibration sweep script
│
├── reports/ # Empirical evaluation logs
│ └── jevbench/ # Benchmark records
│ ├── self_test_phase3_consistency.json # Official 62.53% report
│ ├── jevbench_phase1_report.json
│ └── jevbench_phase2_report.json
│
├── src/ # Model definitions and training routines
│ ├── model.py # MLP Decision Head and Qwen integration
│ ├── dataset.py # Streaming ChatML dataset loader
│ ├── train.py # Phase 1 initialization
│ ├── train_phase2.py # Phase 2 QLoRA training
│ ├── train_phase3_consistency.py # Phase 3 consistency alignment
│ └── evaluate.py # Metric evaluation routines
│
└── scripts/ # Utilities and deployment tools
├── build_consistency_dataset.py # Consistency dataset generator
├── merge_phase3.py # Weight fusing and export
├── merge_and_export.py # Model export routine
├── export_gguf.py # GGUF conversion utility
├── publish_all_hf.py # Hugging Face hub publishing script
├── download_data.py # Dataset download helper
└── download_model.py # Base model download helper
@misc{nayak2026qjev15b,
author = {Ankit Nayak},
title = {QJev-1.5B: High-Speed System 1 JEV Decision Model based on Qwen2.5},
year = {2026},
publisher = {GitHub and Hugging Face},
howpublished = {\url{https://github.com/AnkitNayak-dev/QJev-1.5B}}
}
This project is licensed under the Apache 2.0 License.
Base model weights are governed by the Alibaba Cloud Qwen License. Training data is provided by SargeDev (jev-distill-corpus-v3).
model.safetensors 3.1 GBlora_adapter/adapter_model.bin 18 MBlora_adapter/adapter_model.safetensors 17 MBdecision_head.pt 3 MBdecision_head.safetensors 3 MB