AesSedai/Kimi-K2.7-Code-GGUF overview

---

base_model:

• moonshotai/Kimi-K2.7-Code

---

Model

This is a text-and-image-only GGUF quantization of moonshotai/Kimi-K2.7-Code. This means that video input is not present in this GGUF, and will not be available until support is added upstream in llama.cpp. MMPROJ files for image vision input have been provided.

This Q4_X quant is the "full quality" equivalent since the conditional experts are natively INT4 quantized directly from the original model, and the rest of the model is Q8_0.

| Quant | Size | Mixture | PPL | 1-(Mean PPL(Q)/PPL(base)) | KLD |

| :----- | :--------------------- | :------------------------ | :------------------ | :------------------------ | :------------------ |

| Q4_X | 543.62 GiB (4.55 BPW) | Q8_0 / Q4_0 / Q4_0 / Q4_0 | 2.009155 ± 0.008931 | +0% | 0 |

| Q3_K_L | 459.94 GiB (3.85 BPW) | Q8_0 / Q3_K / Q3_K / Q4_0 | 2.085207 ± 0.009398 | +3.9167% | 0.063007 ± 0.000544 |

| IQ3_S | 377.54 GiB (3.16 BPW) | Q8_0 / varies | 2.310321 ± 0.010823 | +15.1353% | 0.178673 ± 0.001326 |

| IQ2_S | 311.80 GiB (2.61 BPW) | Q8_0 / varies | 2.690751 ± 0.013248 | +34.0941% | 0.343187 ± 0.002100 |

| IQ2_XXS | 262.78 GiB (2.20 BPW) | Q8_0 / varies | 3.420305 ± 0.018189 | +70.4515% | 0.587342 ± 0.002980 |

The mixtures with varies in them were produced by @eaddario 's llama.cpp branches for imatrix and quantize respectively:

• imatrix: https://github.com/ggml-org/llama.cpp/pull/14891
• quantize: https://github.com/ggml-org/llama.cpp/pull/15550

!kld_graph

!ppl_graph