Instructions to use YTan2000/Qwen3.6-27B-TQ3_4S with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use YTan2000/Qwen3.6-27B-TQ3_4S with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="YTan2000/Qwen3.6-27B-TQ3_4S",
	filename="Qwen3.6-27B-TQ3_4S.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": [
				{
					"type": "text",
					"text": "Describe this image in one sentence."
				},
				{
					"type": "image_url",
					"image_url": {
						"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
					}
				}
			]
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps

llama.cpp

How to use YTan2000/Qwen3.6-27B-TQ3_4S with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf YTan2000/Qwen3.6-27B-TQ3_4S
# Run inference directly in the terminal:
llama-cli -hf YTan2000/Qwen3.6-27B-TQ3_4S

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf YTan2000/Qwen3.6-27B-TQ3_4S
# Run inference directly in the terminal:
llama-cli -hf YTan2000/Qwen3.6-27B-TQ3_4S

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 YTan2000/Qwen3.6-27B-TQ3_4S
# Run inference directly in the terminal:
./llama-cli -hf YTan2000/Qwen3.6-27B-TQ3_4S

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 YTan2000/Qwen3.6-27B-TQ3_4S
# Run inference directly in the terminal:
./build/bin/llama-cli -hf YTan2000/Qwen3.6-27B-TQ3_4S

Use Docker

docker model run hf.co/YTan2000/Qwen3.6-27B-TQ3_4S

LM Studio
Jan

vLLM

How to use YTan2000/Qwen3.6-27B-TQ3_4S with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "YTan2000/Qwen3.6-27B-TQ3_4S"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "YTan2000/Qwen3.6-27B-TQ3_4S",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/YTan2000/Qwen3.6-27B-TQ3_4S

Ollama
How to use YTan2000/Qwen3.6-27B-TQ3_4S with Ollama:
```
ollama run hf.co/YTan2000/Qwen3.6-27B-TQ3_4S
```

Unsloth Studio new

How to use YTan2000/Qwen3.6-27B-TQ3_4S 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 YTan2000/Qwen3.6-27B-TQ3_4S 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 YTan2000/Qwen3.6-27B-TQ3_4S to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for YTan2000/Qwen3.6-27B-TQ3_4S to start chatting

Pi new

How to use YTan2000/Qwen3.6-27B-TQ3_4S with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf YTan2000/Qwen3.6-27B-TQ3_4S

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": "Qwen3.6-27B-TQ3_4S"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use YTan2000/Qwen3.6-27B-TQ3_4S with Docker Model Runner:
```
docker model run hf.co/YTan2000/Qwen3.6-27B-TQ3_4S
```

Lemonade

How to use YTan2000/Qwen3.6-27B-TQ3_4S with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull YTan2000/Qwen3.6-27B-TQ3_4S

Run and chat with the model

lemonade run user.Qwen3.6-27B-TQ3_4S-{{QUANT_TAG}}

List all available models

lemonade list

llama.cpp-tq3 crashes loading model in "model router" mode, works fine when run with llama-server directly.

by gdevenyi - opened 3 days ago

Discussion

gdevenyi

3 days ago

Sorry I'm putting this here, there's no issues on the github.

I can't tell if this is an upstream bug, or a bug in llama.cpp-tq3 because I can't test these GGUF in upstream, but the bug is this:

This launch command works fine:

llama-server \
  -m YTan2000/Qwen3.6-27B-TQ3_4S/Qwen3.6-27B-TQ3_4S.gguf \
  --mmproj YTan2000/Qwen3.6-27B-TQ3_4S/mmproj.gguf \
  --host 0.0.0.0 --port 8080 \
  --fit on --fit-ctx 16768  -np 1 \
  -ctk q8_0 -ctv tq3_0 -fa on \
  --jinja \
  --chat-template-file /home/gdevenyi/models/chat_template.jinja \
  --chat-template-kwargs '{"preserve_thinking": true}' \
  --no-mmproj-offload \
  --fit-target 512 \
  --cache-ram 32767 \
  --no-mmap --mlock \
  --props \
  --min-p 0.01 --temperature 0.6  --top-k 20 --top-p 0.95 \
  --reasoning-budget 8192 \
  --reasoning-budget-message "My reasoning budget is exhausted, but I have enough information to answer directly now."

However, if I dump this into my multi-model router ini file:

[*]
chat-template-kwargs = {"preserve_thinking": true}
models-dir = /home/gdevenyi/models

parallel = 1
models-max = 1

; memory
no-mmap = true
mlock = true
fit = on
fit-target = 512


cache-ram = 32768

; compute
flash-attn = on
cache-type-k = tq3_0
cache-type-v = q8_0

jinja = true
chat-template-file = /home/gdevenyi/models/chat_template.jinja

no-mmproj-offload = true

; sampling
temp = 0.6
top-p = 0.95
top-k = 20
min-p = 0.01


; allow remote control
props = true

; https://whamp.github.io/blog/qwen-3-6-27b-livecodebench-reasoning-budget/
reasoning-budget-message = "My reasoning budget is exhausted, but I have enough information to answer directly now."
reasoning-budget = 8192

[qwen3.6-27B]
model = /home/gdevenyi/models/YTan2000/Qwen3.6-27B-TQ3_4S/Qwen3.6-27B-TQ3_4S.gguf
mmproj = /home/gdevenyi/models/YTan2000/Qwen3.6-27B-TQ3_4S/mmproj.gguf
fit-ctx = 16768
cache-type-k = tq3_0
cache-type-v = q4_0

It fails with this error:

[39955] /home/gdevenyi/projects/llama/llama.cpp-tq3/ggml/src/ggml-cuda/fattn.cu:304: fatal error
[39955] [New LWP 276111]
[39955] [New LWP 276110]
[39955] [New LWP 276109]
[39955] [New LWP 276108]
[39955] [New LWP 276107]
[39955] [New LWP 276106]
[39955] [New LWP 276105]
[39955] [New LWP 276104]
[39955] [New LWP 276102]
[39955] [New LWP 276101]
[39955] [New LWP 276100]
[39955] [New LWP 276099]
[39955] [New LWP 276098]
[39955] [New LWP 276097]
[39955] [New LWP 276096]
[39955] [New LWP 275984]
[39955] [New LWP 275983]
[39955] [New LWP 275982]
[39955] [New LWP 275981]
[39955] [New LWP 275980]
[39955] [New LWP 275979]
[39955] [New LWP 275978]
[39955] [New LWP 275977]
[39955] [New LWP 275976]
[39955] [New LWP 275975]
[39955] [New LWP 275974]
[39955] [New LWP 275973]
[39955] [New LWP 275972]
[39955] [New LWP 275971]
[39955] [New LWP 275970]
[39955] [New LWP 275969]
[39955] [New LWP 275968]
[39955] [New LWP 275967]
[39955] [New LWP 275966]
[39955] [New LWP 275965]
[39955] [New LWP 275964]
[39955] [New LWP 275963]
[39955] [New LWP 275962]
[39955] [New LWP 275961]
[39955] [New LWP 275960]
[39955] [New LWP 275959]
[39955] [New LWP 275958]
[39955] [New LWP 275957]
[39955] [New LWP 275956]
[39955] [New LWP 275955]
[39955] [New LWP 275954]
[39955] [New LWP 275953]
[39955] [New LWP 275952]
[39955] [New LWP 275951]
[39955] [New LWP 275950]
[39955] [New LWP 275949]
[39955] [New LWP 275948]
[39955] [New LWP 275947]
[39955] [New LWP 275945]
[39955] [New LWP 275936]
[39955]
[39955] This GDB supports auto-downloading debuginfo from the following URLs:
[39955]   <https://debuginfod.neon.kde.org/>
[39955]   <https://debuginfod.ubuntu.com/>
[39955] Enable debuginfod for this session? (y or [n]) [answered N; input not from terminal]
[39955] Debuginfod has been disabled.
[39955] To make this setting permanent, add 'set debuginfod enabled off' to .gdbinit.
[39955] [Thread debugging using libthread_db enabled]
[39955] Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
[39955] 0x00007faf1e110813 in __GI___wait4 (pid=276318, stat_loc=0x0, options=0, usage=0x0) at ../sysdeps/unix/sysv/linux/wait4.c:30
[39955] warning: 30     ../sysdeps/unix/sysv/linux/wait4.c: No such file or directory
[39955] #0  0x00007faf1e110813 in __GI___wait4 (pid=276318, stat_loc=0x0, options=0, usage=0x0) at ../sysdeps/unix/sysv/linux/wait4.c:30
[39955] 30      in ../sysdeps/unix/sysv/linux/wait4.c
[39955] #1  0x00007faf1ec4d213 in ggml_print_backtrace () from /opt/llama.cpp/lib/libggml-base.so.0
[39955] #2  0x00007faf1ec4d3bb in ggml_abort () from /opt/llama.cpp/lib/libggml-base.so.0
[39955] #3  0x00007faf1ac1940b in ggml_cuda_flash_attn_ext(ggml_backend_cuda_context&, ggml_tensor*) () from /opt/llama.cpp/lib/libggml-cuda.so.0
[39955] #4  0x00007faf1ac5dbc5 in ggml_cuda_compute_forward(ggml_backend_cuda_context&, ggml_tensor*) () from /opt/llama.cpp/lib/libggml-cuda.so.0
[39955] #5  0x00007faf1ac63051 in ggml_cuda_graph_evaluate_and_capture(ggml_backend_cuda_context*, ggml_cgraph*, bool, bool, void const*) () from /opt/llama.cpp/lib/libggml-cuda.so.0
[39955] #6  0x00007faf1ac64e1e in ggml_backend_cuda_graph_compute(ggml_backend*, ggml_cgraph*) () from /opt/llama.cpp/lib/libggml-cuda.so.0
[39955] #7  0x00007faf1ec6aef7 in ggml_backend_sched_graph_compute_async () from /opt/llama.cpp/lib/libggml-base.so.0
[39955] #8  0x00007faf1e8c2511 in llama_context::graph_compute(ggml_cgraph*, bool) () from /opt/llama.cpp/lib/libllama.so.0
[39955] #9  0x00007faf1e8c4ee2 in llama_context::process_ubatch(llama_ubatch const&, llm_graph_type, llama_memory_context_i*, ggml_status&) () from /opt/llama.cpp/lib/libllama.so.0
[39955] #10 0x00007faf1e8cbb5f in llama_context::decode(llama_batch const&) () from /opt/llama.cpp/lib/libllama.so.0
[39955] #11 0x00007faf1e8cd76f in llama_decode () from /opt/llama.cpp/lib/libllama.so.0
[39955] #12 0x000055be36967ce2 in common_init_from_params(common_params&) ()
[39955] #13 0x000055be3686784c in server_context_impl::load_model(common_params&) ()
[39955] #14 0x000055be367aa706 in main ()
[39955] [Inferior 1 (process 275932) detached]
^Csrv    operator(): operator(): cleaning up before exit...
srv    unload_all: stopping model instance name=qwen3.6-27B

I have confirmed that the router model invokes llama-server with the same command options, but for some reason it doesn't result in the same effect

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