Instructions to use ubergarm/Kimi-K2.6-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ubergarm/Kimi-K2.6-GGUF with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="ubergarm/Kimi-K2.6-GGUF",
	filename="IQ3_K/Kimi-K2.6-IQ3_K-00001-of-00012.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use ubergarm/Kimi-K2.6-GGUF with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K
# Run inference directly in the terminal:
llama-cli -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K
# Run inference directly in the terminal:
llama-cli -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K
# Run inference directly in the terminal:
./llama-cli -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K
# Run inference directly in the terminal:
./build/bin/llama-cli -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Use Docker

docker model run hf.co/ubergarm/Kimi-K2.6-GGUF:Q2_K

LM Studio
Jan

vLLM

How to use ubergarm/Kimi-K2.6-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ubergarm/Kimi-K2.6-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ubergarm/Kimi-K2.6-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ubergarm/Kimi-K2.6-GGUF:Q2_K

Ollama
How to use ubergarm/Kimi-K2.6-GGUF with Ollama:
```
ollama run hf.co/ubergarm/Kimi-K2.6-GGUF:Q2_K
```

Unsloth Studio

How to use ubergarm/Kimi-K2.6-GGUF with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for ubergarm/Kimi-K2.6-GGUF to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for ubergarm/Kimi-K2.6-GGUF to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for ubergarm/Kimi-K2.6-GGUF to start chatting

How to use ubergarm/Kimi-K2.6-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ubergarm/Kimi-K2.6-GGUF:Q2_K"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use ubergarm/Kimi-K2.6-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf ubergarm/Kimi-K2.6-GGUF:Q2_K

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default ubergarm/Kimi-K2.6-GGUF:Q2_K

Run Hermes

hermes

Docker Model Runner
How to use ubergarm/Kimi-K2.6-GGUF with Docker Model Runner:
```
docker model run hf.co/ubergarm/Kimi-K2.6-GGUF:Q2_K
```

Lemonade

How to use ubergarm/Kimi-K2.6-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ubergarm/Kimi-K2.6-GGUF:Q2_K

Run and chat with the model

lemonade run user.Kimi-K2.6-GGUF-Q2_K

List all available models

lemonade list

Testing smol-IQ3_KS

by shewin - opened Apr 22

Discussion

shewin

Apr 22

Computed blk.60.attn_kv_b.weight as 512 x 16384 of type q8_0 and stored in buffer CUDA0
llama_init_from_model: n_ctx = 161280
llama_init_from_model: n_batch = 8192
llama_init_from_model: n_ubatch = 8192
llama_init_from_model: flash_attn = 1
llama_init_from_model: mla_attn = 3
llama_init_from_model: attn_max_b = 4096
llama_init_from_model: fused_moe = 1
llama_init_from_model: grouped er = 1
llama_init_from_model: fused_up_gate = 1
llama_init_from_model: fused_mmad = 1
llama_init_from_model: rope_cache = 0
llama_init_from_model: graph_reuse = 1
llama_init_from_model: k_cache_hadam = 0
llama_init_from_model: v_cache_hadam = 0
llama_init_from_model: split_mode_graph_scheduling = 0
llama_init_from_model: reduce_type = f16
llama_init_from_model: sched_async = 0
llama_init_from_model: ser = -1, 0
llama_init_from_model: freq_base = 50000.0
llama_init_from_model: freq_scale = 0.015625
llama_kv_cache_init: CUDA0 KV buffer size = 5742.01 MiB
llama_init_from_model: KV self size = 5741.98 MiB, c^KV (q8_0): 5741.98 MiB, kv^T: not used
llama_init_from_model: CUDA_Host output buffer size = 0.62 MiB
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 1
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 2
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 3
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 4
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 5
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 6
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 7
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 8
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 9
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 10
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 11
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 12
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 13
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 14
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 15
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 16
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 17
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 18
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 19
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 20
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 21
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 22
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 23
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 24
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 25
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 26
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 27
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 28
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 29
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 30
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 31
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 32
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 33
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 34
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 35
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 36
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 37
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 38
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 39
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 40
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 41
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 42
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 43
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 44
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 45
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 46
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 47
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 48
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 49
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 50
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 51
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 52
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 53
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 54
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 55
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 56
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 57
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 58
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 59
llama_repack_up_gate_exps: repacking up/gate experts weight in layer 60
llama_init_from_model: CUDA0 compute buffer size = 8023.03 MiB
llama_init_from_model: CUDA_Host compute buffer size = 2744.09 MiB
llama_init_from_model: graph nodes = 5417
llama_init_from_model: graph splits = 122
llama_init_from_model: enabling only_active_experts scheduling

main: n_kv_max = 161280, n_batch = 8192, n_ubatch = 8192, flash_attn = 1, n_gpu_layers = 99, n_threads = 101, n_threads_batch = 101

PP	TG	N_KV	T_PP s	S_PP t/s	T_TG s	S_TG t/s
8192	2048	0	12.100	677.04	109.233	18.75
8192	2048	8192	13.790	594.04	177.558	11.53
8192	2048	16384	15.557	526.57	118.881	17.23
8192	2048	24576	17.367	471.71	123.425	16.59
8192	2048	32768	19.337	423.64	133.373	15.36

shewin

Apr 22

better and faster than IQ3_K

ubergarm

Owner Apr 22

@shewin

fascinating.. the smol-IQ3_KS uses imatrix for everything. it should definitely be faster given less overall active weights size, so more TG throughput for sure.

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment