nimishchaudhari
/

Magistral-Small-2509-IQ4_KS-GGUF

Eval Results (legacy)

Model card Files Files and versions

Magistral-Small-2509-IQ4_KS GGUF

A high-quality quantized version of mistralai/Magistral-Small-2509 using ik_llama.cpp with importance matrix optimization and custom tensor-specific quantization.

Model Details

Base Model: mistralai/Magistral-Small-2509
Quantization Method: Custom IQ4_KS recipe with importance matrix
Target VRAM: 12GB for full offload
Quantization Tool: ik_llama.cpp with imatrix calibration

Benchmark Results

Perplexity (wiki.test.raw, ctx=512)

Metric	Value
Final PPL	5.6245 ± 0.03312
Evaluation Chunks	594
Evaluation Tokens	304,128
Context Size	512

Downloads last month: 25

GGUF

Model size

24B params

Architecture

llama

Hardware compatibility

Log In to add your hardware

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for nimishchaudhari/Magistral-Small-2509-IQ4_KS-GGUF

Base model

mistralai/Mistral-Small-3.1-24B-Base-2503

Finetuned

mistralai/Mistral-Small-3.2-24B-Instruct-2506

Finetuned

mistralai/Magistral-Small-2509

Quantized

(27)

this model

Evaluation results

perplexity on wiki.test.raw
self-reported

5.625