Team Ai
Modelpublic

NS-Studio/Kernel-Logic1.5

sourceHugging Faceapache-2.0updated 21d agoView on Hugging Face
1likes82downloads
Model Card

Kernel Logic 1.5

Kernel Logic 1.5 is a fine-tuned large language model optimized for logical reasoning, mathematics, and code generation tasks. Provided in GGUF (Q4_K_M) format, it is engineered for efficient local inference across standard consumer hardware.


โš™๏ธ Training Hardware & Environment

Trained and quantized using Unsloth on Google Colab (Google Compute Engine backend):

ComponentSpecificationUsage During Run
GPUNVIDIA Tesla T4 (16 GB VRAM)6.1 GB / 15.0 GB
System RAM12.7 GB High-Memory6.0 GB / 12.7 GB
Storage / Disk112.6 GB Virtual Disk72.7 GB / 112.6 GB
EnvironmentUbuntu / Python 3 RuntimeGoogle Colab Cloud GPU
FrameworkUnsloth + Llama.cpp QuantizationQ4KM GGUF Export

๐Ÿ“Š Benchmark Results

Evaluated on September 17, 2026:

BenchmarkCapability / TaskScore / Accuracy
GSM8kMathematical Reasoning68.0%
HumanEvalPython Code Generation & Problem Solving57.5%
TruthfulQAFactual Accuracy & Hallucination Resistance48.0%
MMLU-ProMulti-discipline Academic Understanding38.0%
Average ScoreOverall Performance52.88%
  • โ€”Average Generation Speed: 14.68 tokens/second
json
{
  "Kernel Logic 1.5": {
    "evaluated_at": "2026-09-17 13:47:01.356204",
    "mmlu_pro_acc": 38.0,
    "gsm8k_acc": 68.0,
    "humaneval_acc": 57.5,
    "truthfulqa_acc": 48.0,
    "average_score_pct": 52.88,
    "avg_tokens_per_second": 14.68
  }
}