---

base_model: IFM/K2-Horizon-0.9B

language:

- en

license: apache-2.0

tags:

- gguf

- quantized

- llama.cpp

- k2-horizon

- causal-lm

- text-generation

pipeline_tag: text-generation

---

K2-Horizon-0.9B — GGUF Quantizations

GGUF quantizations of IFM/K2-Horizon-0.9B,

a dense 0.9B-parameter causal decoder (K2HorizonForCausalLM, model_type: k2_horizon).

Quantized by NANI-Nithin using a custom pipeline built on the

MBZUAI-IFM llama.cpp fork (branch model/K2Horizon),

which adds native support for the k2-horizon architecture not yet merged into upstream llama.cpp.

---

Model Details

| Property | Value |

|---|---|

| Base model | IFM/K2-Horizon-0.9B |

| Architecture | K2HorizonForCausalLM (k2_horizon) |

| Parameters | ~0.9B (dense decoder, no MoE) |

| Original dtype | BF16 |

| Source GGUF | IFM/K2-Horizon-0.9B-GGUF |

| llama.cpp fork | MBZUAI-IFM/llama.cpp @ model/K2Horizon |

> Note: These GGUFs carry the k2-horizon architecture token and require

> the MBZUAI-IFM fork (or upstream llama.cpp once support is merged) to run.

> Vanilla upstream llama.cpp (as of September 2026) does not support

> K2HorizonForCausalLM.

---

Included Files

Standard Quantizations

| File | Bits/Weight | Notes |

|---|---|---|

| K2-Horizon-0.9B-BF16.gguf | 16 bpw | Source quant, full precision |

| K2-Horizon-0.9B-Q8_0.gguf | 8 bpw | Near-lossless, recommended reference |

| K2-Horizon-0.9B-Q6_K.gguf | 6 bpw | Near-lossless K-quant |

| K2-Horizon-0.9B-Q5_K_M.gguf | 5 bpw | Best quality/size in the 5-bit range |

| K2-Horizon-0.9B-Q5_K_S.gguf | 5 bpw | Smaller 5-bit variant |

| K2-Horizon-0.9B-Q5_1.gguf | 5 bpw | Legacy 5-bit |

| K2-Horizon-0.9B-Q5_0.gguf | 5 bpw | Legacy 5-bit |

| K2-Horizon-0.9B-Q4_K_M.gguf | 4 bpw | Recommended general use |

| K2-Horizon-0.9B-Q4_K_S.gguf | 4 bpw | Smaller 4-bit K-quant |

| K2-Horizon-0.9B-Q4_1.gguf | 4 bpw | Legacy 4-bit |

| K2-Horizon-0.9B-Q4_0.gguf | 4 bpw | Legacy 4-bit |

| K2-Horizon-0.9B-Q3_K_L.gguf | 3 bpw | Large 3-bit K-quant |

| K2-Horizon-0.9B-Q3_K_M.gguf | 3 bpw | Medium 3-bit K-quant |

| K2-Horizon-0.9B-Q3_K_S.gguf | 3 bpw | Small 3-bit K-quant |

| K2-Horizon-0.9B-Q2_K.gguf | 2 bpw | Aggressive compression |

| K2-Horizon-0.9B-Q2_K_S.gguf | 2 bpw | Smaller 2-bit K-quant (imatrix-guided) |

| K2-Horizon-0.9B-Q2_0.gguf | 2.25 bpw | Group-64 2-bit |

| K2-Horizon-0.9B-Q1_0.gguf | 1.125 bpw | Maximum compression |

IQ (Importance-Matrix) Quantizations

| File | Bits/Weight |

|---|---|

| K2-Horizon-0.9B-IQ4_NL.gguf | ~4 bpw |

| K2-Horizon-0.9B-IQ4_XS.gguf | ~4 bpw |

| K2-Horizon-0.9B-IQ3_M.gguf | ~3 bpw |

| K2-Horizon-0.9B-IQ3_S.gguf | ~3 bpw |

| K2-Horizon-0.9B-IQ3_XS.gguf | ~3 bpw |

| K2-Horizon-0.9B-IQ3_XXS.gguf | ~3 bpw |

| K2-Horizon-0.9B-IQ2_M.gguf | ~2 bpw |

| K2-Horizon-0.9B-IQ2_S.gguf | ~2 bpw |

| K2-Horizon-0.9B-IQ2_XS.gguf | ~2 bpw |

| K2-Horizon-0.9B-IQ2_XXS.gguf | ~2 bpw |

| K2-Horizon-0.9B-IQ1_M.gguf | 1.75 bpw |

| K2-Horizon-0.9B-IQ1_S.gguf | 1.56 bpw |

---

Quantization Method

• Source: IFM's official BF16 GGUF is downloaded directly (no safetensors conversion needed).
• imatrix: Computed from the Salesforce/wikitext

(wikitext-2-raw-v1) calibration dataset (500 rows). Applied to all K-quants below Q6

and all IQ quants for guided, higher-quality compression.

• Fork: MBZUAI-IFM/llama.cpp, branch model/K2Horizon.

---

Usage

> Requires the MBZUAI-IFM llama.cpp fork (model/K2Horizon branch).

# Clone and build the fork
git clone -b model/K2Horizon https://github.com/MBZUAI-IFM/llama.cpp
cd llama.cpp && cmake -B build -DGGML_CUDA=ON && cmake --build build --config Release

# Run inference
./build/bin/llama-cli \
  -m K2-Horizon-0.9B-Q4_K_M.gguf \
  -p "Hello, I am" \
  -n 128

---

License

Weights are released under the same license as the original

IFM/K2-Horizon-0.9B model.

Please refer to the original repository for full license terms.

---

Credits

• Original model: MBZUAI IFM
• Quantized by: NANI-Nithin
• Quantization tool: MBZUAI-IFM/llama.cpp