CoolFace
Modelpublic

Alcoft/dnotitia_Smoothie-Qwen3-4B-GGUF

sourceHugging Faceapache-2.0updated 1y agoView on Hugging Face
0likes1.6kdownloads
Model Card
QuantSizeDescription
Q2_K_XXS1.38 GBNot recommended for most people. Extremelly low quality.
Q2_K_XS1.38 GBNot recommended for most people. Very low quality.
Q2_K1.38 GBNot recommended for most people. Very low quality.
Q2_K_L1.64 GBNot recommended for most people. Uses Q80 for output and embedding, and Q2K for everything else. Very low quality.
Q2_K_XL1.98 GBNot recommended for most people. Uses F16 for output and embedding, and Q2_K for everything else. Very low quality.
Q3_K_XXS1.62 GBNot recommended for most people. Prefer any bigger Q3_K quantization. Very low quality.
Q3_K_XS1.62 GBNot recommended for most people. Prefer any bigger Q3_K quantization. Very low quality.
Q3_K_S1.62 GBNot recommended for most people. Prefer any bigger Q3_K quantization. Low quality.
Q3_K_M1.79 GBNot recommended for most people. Low quality.
Q3_K_L1.94 GBNot recommended for most people. Low quality.
Q3_K_XL2.17 GBNot recommended for most people. Uses Q80 for output and embedding, and Q3K_L for everything else. Low quality.
Q3_K_XXL2.51 GBNot recommended for most people. Uses F16 for output and embedding, and Q3KL for everything else. Low quality.
Q4_K_XS2.13 GBLower quality than Q4KS.
Q4_K_S2.13 GBRecommended. Slightly low quality.
Q4_K_M2.23 GBRecommended. Decent quality for most use cases.
Q4_K_L2.41 GBRecommended. Uses Q80 for output and embedding, and Q4K_M for everything else. Decent quality.
Q4_K_XL2.75 GBRecommended. Uses F16 for output and embedding, and Q4KM for everything else. Decent quality.
Q5_K_XXS2.58 GBLower quality than Q5KS.
Q5_K_XS2.58 GBLower quality than Q5KS.
Q5_K_S2.58 GBRecommended. High quality.
Q5_K_M2.64 GBRecommended. High quality.
Q5_K_L2.78 GBRecommended. Uses Q80 for output and embedding, and Q5K_M for everything else. High quality.
Q5_K_XL3.12 GBRecommended. Uses F16 for output and embedding, and Q5KM for everything else. High quality.
Q6_K_S3.08 GBLower quality than Q6_K.
Q6_K3.08 GBRecommended. Very high quality.
Q6_K_L3.17 GBRecommended. Uses Q80 for output and embedding, and Q6K for everything else. Very high quality.
Q6_K_XL3.51 GBRecommended. Uses F16 for output and embedding, and Q6_K for everything else. Very high quality.
Q8_K_XS3.99 GBLower quality than Q8_0.
Q8_K_S3.99 GBLower quality than Q8_0.
Q8_03.99 GBRecommended. Quality almost like F16.
Q8_K_XL4.33 GBRecommended. Uses F16 for output and embedding, and Q8_0 for everything else. Quality almost like F16.
F167.5 GBNot recommended. Overkill. Prefer Q8_0.
ORIGINAL (BF16)7.5 GBNot recommended. Overkill. Prefer Q8_0.

Quantized using TAO71-AI AutoQuantizer. You can check out the original model card here.