Instructions to use migtissera/Tess-4-27B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use migtissera/Tess-4-27B with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="migtissera/Tess-4-27B")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("migtissera/Tess-4-27B")
model = AutoModelForMultimodalLM.from_pretrained("migtissera/Tess-4-27B", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use migtissera/Tess-4-27B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "migtissera/Tess-4-27B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "migtissera/Tess-4-27B",
		"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/migtissera/Tess-4-27B

SGLang

How to use migtissera/Tess-4-27B with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "migtissera/Tess-4-27B" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "migtissera/Tess-4-27B",
		"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 images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "migtissera/Tess-4-27B" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "migtissera/Tess-4-27B",
		"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"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use migtissera/Tess-4-27B with Docker Model Runner:
```
docker model run hf.co/migtissera/Tess-4-27B
```

Benchmarks?

by volodXYZ - opened Jul 8

Discussion

volodXYZ

Jul 8

Have you run any benchmarks compared to the base model?
Otherwise what is the point?

Dragonite9000

Jul 8

benchies should be a requirement for posting any finetunes

Iwaku-Real

Jul 8

benchies should be a requirement for posting any finetunes

Not in the HF community, you must https://tryitands.ee 😄

khronnuz

Jul 9

•

edited Jul 9

Running localbench full suite on a 48GB VRAM setup (RTX 5090+5060) via llama-server:

Tess-4-27B (Q8 thinking on)
81% (122/150) · 1h10m
Qwen3.6-27B (Q8_0 MTP thinking on)
80% (120/150) 47m
Qwen3.6-27B (Q8_0 MTP thinking off)
79% (119/150) · 37m49s
Qwen3.6-35B-A3B (UD-Q8_K_XL)
78% (117/150) · 20m24s
Gemma-4-31B (Q6 · 180k ctx)
78% (117/150) · 44m27s
Qwopus3.6-27B Coder-Compat (Q6_K)
77% (116/150) · 34m18s
Qwen3.6-27B pi-tune (Q8)
77% (115/150) · 27m42s
Gemma-4-31B (Q8)
77% (115/150) · 50m25s
Gemopus-4-31B (Q8)
76% (114/150) · 1h21m
Gemopus-4-31B (Q8 · half context)
76% (114/150) · 1h45m
Gemma-4-31B QAT (UD-Q4_K_XL)
75% (113/150) · 20m55s
Gemopus-4-31B (Q8)
75% (112/150) · 1h16m
Ornith-1.0-35B (Q8)
73% (110/150) · 31m45s
Qwopus3.6-35B-A3B Coder (Q8)
69% (103/150) · 14m23s
Qwen3.5-122B-A10B (UD-IQ4_XS · ik_llama)
69% (103/150) · 1h30m
Step-3.7-Flash (IQ4_XS)
65% (97/150) · 2h00m
Qwen3-Coder-Next (UD-Q4_K_M)
61% (91/150) · 14m23s
Gemopus-4-31B (Q8)
61% (91/150) · 1h26m

migtissera

Owner Jul 9

Thanks @khronnuz ! If you can provide something more detailed (like the 122/150 breakdown) I will add this to the README.

khronnuz

Jul 9

@migtissera you got it: https://gist.github.com/everson/261fdef8a3d35298b36a07f436e407f6

pick whatever parts you want.

migtissera

Owner Jul 9

@khronnuz Thanks my guy!

migtissera

Owner Jul 10

I have added the trained MTP head here: https://huggingface.co/migtissera/Tess-4-27B-EAGLE3

Will accelerate the speed 2x.

migtissera changed discussion status to closed Jul 10

TESTPOINTrxz

Jul 13

@khronnuz Was the Ornith-1.0-35B test done with thinking enabled or disabled? I swear it's better than some of the models listed above it.

khronnuz

Jul 15

@TESTPOINTrxz thinking disabled.

khronnuz

Jul 16

•

edited Jul 16

@TESTPOINTrxz I actually just ran Ornith with thinking enabled and it come out the same score.

llama-server --model ornith-1.0-35b-Q8_0.gguf --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0 \
--presence-penalty 1.5 --ctx-size 262144 -ngl auto --fit on --cache-type-k f16 --cache-type-v f16 \ 
--reasoning on -np 1 -fa on -b 4096 -ub 1024 --jinja --split-mode tensor --main-gpu 1 --tensor-split 3,4 \
--spec-type draft-mtp --spec-draft-n-max 4 --model-draft a4lg/Qwen3.5-35B-A3B-MTP-ONLY-GGUF/Qwen3.5-35B-A3B-MTP-ONLY-Q8_0.gguf \
--reasoning-preserve --reasoning-budget 4096

TESTPOINTrxz

Jul 20

Thanks! I thought the score was a little bit off compared to what i saw from the model.

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