Image-Text-to-Text
Transformers
Safetensors
qwen3_5
efficient
qwen
qwen3.5
nomi
lazyloopstudio
unsloth
nomi2
conversational
Instructions to use JallyAI/Nomi-2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use JallyAI/Nomi-2 with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="JallyAI/Nomi-2") 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("JallyAI/Nomi-2") model = AutoModelForMultimodalLM.from_pretrained("JallyAI/Nomi-2", 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 JallyAI/Nomi-2 with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "JallyAI/Nomi-2" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "JallyAI/Nomi-2", "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/JallyAI/Nomi-2
- SGLang
How to use JallyAI/Nomi-2 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 "JallyAI/Nomi-2" \ --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": "JallyAI/Nomi-2", "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 "JallyAI/Nomi-2" \ --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": "JallyAI/Nomi-2", "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" } } ] } ] }' - Unsloth Studio
How to use JallyAI/Nomi-2 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 JallyAI/Nomi-2 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 JallyAI/Nomi-2 to start chatting
Using HuggingFace Spaces for Unsloth
# No setup required # Open https://huggingface.co/spaces/unsloth/studio in your browser # Search for JallyAI/Nomi-2 to start chatting
Load model with FastModel
pip install unsloth from unsloth import FastModel model, tokenizer = FastModel.from_pretrained( model_name="JallyAI/Nomi-2", max_seq_length=2048, ) - Docker Model Runner
How to use JallyAI/Nomi-2 with Docker Model Runner:
docker model run hf.co/JallyAI/Nomi-2
File size: 11,185 Bytes
a2ee815 0664ff1 202075c f2e9275 65958a3 202075c 65958a3 0664ff1 0097c74 202075c 12839aa 202075c 2b02981 202075c 65958a3 202075c 9e1d4eb 65958a3 cb0d509 65958a3 cb0d509 65958a3 cb0d509 39b898f 188d55e cb0d509 65958a3 202075c 0097c74 65958a3 0097c74 65958a3 0097c74 dd7a4bc | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 | ---
license: apache-2.0
base_model:
- Qwen/Qwen3.5-4B
pipeline_tag: image-text-to-text
library_name: transformers
tags:
- efficient
- qwen
- qwen3.5
- nomi
- lazyloopstudio
- unsloth
- nomi2
---
<p align="center">
<img src="https://cdn-uploads.huggingface.co/production/uploads/6921fa6332f7fb129563d495/aR36SrpWzksbcGbcp84pE.png" width="128">
</p>
# Nomi 2.0
## Introduction
Introducing **Nomi 2**, an 4B LLM based on **Qwen 3.5 4B**, that thinks more efficient while keeping great performance.
In this training, we used another base and improved it's reasoning. It's base, Qwen 3.5 4B was great, but overthought most of the requests und looped.
**Nomi 2** fixes that.
Even at more complicated prompts it only thinks for about 10 sec on consumer hardware (~500 Reasoning tokens).
Nomi 2 works by using one specific reasoning style:
Instead of human like CoT, Nomi uses what I like to call **RASV**.
In it's reasoning, it got 4 main parts:
- Restatement
- Approach
- Step-by-step derivation
- Verification
These steps help the LLM to **keep the reasoning short, while keeping it effective, and avoiding loops.**
## π Key Features & Improvements
* **Architecture:** Qwen-3.5-4B (runs on 8 GB VRAM GPUs like the RTX 4060).
* **Multilingual Support:** Can understand and generate text in German and English and many other languages.
* **Efficiency:** Runs with 60+ Tokens/s at Q4 with an Context Window of almost 100k tokens on consumer hardware.
## π Evals and Showcase
### Evals
| Model Name | GSM8K* | GPQA Diamond (0-Shot)*
| --- | --- | --- |
| Nomi 2.0 | 41/100 | 32/50
<details>
<summary>*</summary>
Benchmarks were done without reasoning (didn't got it to work) and only the first 100/50 Questions of the Benchmark were used.
</details>
### Reasoning Style Showcase
Q: What is 8x8x8+43+(432+432x3)-39?
<details>
<summary>Nomi</summary>
<div style="background-color: #616161; padding: 12px; border-radius: 8px;">
## Restatement
We need to evaluate the arithmetic expression 8 Γ 8 Γ 8 + 43 + (432 + 432 Γ 3) β 39. The operations include multiplication, addition, and subtraction, and we must apply the standard order of operations (PEMDAS/BODMAS).
## Approach
Multiplication has higher precedence than addition and subtraction, so we compute all products first, then handle the parentheses grouping, then add/subtract from left to right. Specifically:
- Inside the parentheses: 432 Γ 3 first, then add 432
- Outside: 8 Γ 8 Γ 8, then combine everything
## Step-by-step derivation
**Step 1 β Inside parentheses:** 432 Γ 3 = 1296. Then 432 + 1296 = 1728. So (432 + 432 Γ 3) = 1728.
**Step 2 β Outside:** 8 Γ 8 Γ 8 = 64 Γ 8 = 512.
**Step 3 β Combine:** 512 + 43 + 1728 β 39.
- 512 + 43 = 555
- 555 + 1728 = 2283
- 2283 β 39 = 2244
**Answer: 2244.**
## Verification
Check by recombining: 512 + 43 + 1728 = 2283, and 2283 β 39 = 2244. Alternatively, note 512 + 1728 = 2240, then 2240 + 43 β 39 = 2240 + 4 = 2244. Both methods agree.
</div>
**2244**
To evaluate 8 Γ 8 Γ 8 + 43 + (432 + 432 Γ 3) β 39, we follow the order of operations: multiplication first, then parentheses, then addition and subtraction from left to right.
Inside the parentheses, we compute 432 Γ 3 = 1296, then 432 + 1296 = 1728. Outside, we compute 8 Γ 8 Γ 8 = 64 Γ 8 = 512. The expression now reads 512 + 43 + 1728 β 39.
Adding from left to right: 512 + 43 = 555, then 555 + 1728 = 2283, and finally 2283 β 39 = 2244.
As a sanity check, I can reorder the additions: 512 + 1728 = 2240, and 2240 + 43 β 39 = 2240 + 4 = 2244. The two independent paths give the same result, confirming the answer.
</details>
<details>
<summary>Qwen</summary>
<div style="background-color: #616161; padding: 12px; border-radius: 8px;">
Here's my thought process for solving this math problem:
1. **Analyze the Request:** The user wants to solve a specific arithmetic expression: $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$.
2. **Identify the Operations:** I need to perform multiplication, addition, and subtraction.
3. **Apply Order of Operations (PEMDAS/BODMAS):**
* **P/B:** Parentheses/Brackets. I need to solve the expression inside the parentheses first.
* **E/O:** Exponents/Orders. There are none here (unless $8 \times 8 \times 8$ is interpreted as a power, but standard notation implies multiplication).
* **MD/DM:** Multiplication and Division (left to right).
* **AS:** Addition and Subtraction (left to right).
4. **Step-by-Step Breakdown:**
* **Expression:** $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$
* **Step 1: Parentheses**
* Inside the parentheses: $(432 + 432 \times 3)$
* Within the parentheses, multiplication comes before addition.
* Calculate $432 \times 3$.
* $400 \times 3 = 1200$
* $30 \times 3 = 90$
* $2 \times 3 = 6$
* $1200 + 90 + 6 = 1296$
* Now add 432 to the result.
* $432 + 1296$
* $2 + 6 = 8$
* $3 + 9 = 12$ (carry 1)
* $4 + 2 + 1 = 7$
* $1$
* Result: $1728$
* So, the expression becomes: $8 \times 8 \times 8 + 43 + 1728 - 39$
* **Step 2: Multiplication (outside the parentheses)**
* Calculate $8 \times 8 \times 8$.
* $8 \times 8 = 64$
* $64 \times 8 = 512$
* (Note: $8^3 = 512$)
* The expression becomes: $512 + 43 + 1728 - 39$
* **Step 3: Addition and Subtraction (Left to Right)**
* Current expression: $512 + 43 + 1728 - 39$
* First addition: $512 + 43$
* $512 + 40 = 552$
* $552 + 3 = 555$
* Current expression: $555 + 1728 - 39$
* Second addition: $555 + 1728$
* $500 + 1700 = 2200$
* $55 + 28 = 83$
* $2200 + 83 = 2283$
* Current expression: $2283 - 39$
* Final subtraction: $2283 - 39$
* $2283 - 30 = 2253$
* $2253 - 9 = 2244$
* Alternatively: $2283 - 40 + 1 = 2243 + 1 = 2244$
5. **Double Check:**
* Original: $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$
* $8^3 = 512$
* $432 \times 3 = 1296$
* $432 + 1296 = 1728$
* Sum so far: $512 + 43 + 1728 - 39$
* $512 + 43 = 555$
* $555 + 1728 = 2283$
* $2283 - 39 = 2244$
* Wait, let me re-read the expression carefully.
* $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$
* Is it possible the user meant $432 + (432 \times 3)$? Yes, that's what I did.
* Is it possible there is a typo in my reading? "8x8x8+43+(432+432x3)-39"
* Let's check the numbers again.
* $512 + 43 = 555$
* $432 \times 3 = 1296$
* $432 + 1296 = 1728$
* $555 + 1728 = 2283$
* $2283 - 39 = 2244$
* Let's check if there's any ambiguity. "432+432x3" is inside parentheses.
* Multiplication takes precedence inside the parentheses.
* $432 + 1296 = 1728$. Correct.
* $8^3 = 512$. Correct.
* $512 + 43 + 1728 - 39$.
* $512 + 43 = 555$.
* $555 + 1728 = 2283$.
* $2283 - 39 = 2244$.
* Wait, is there a trick?
* $432 + 432 \times 3$...
* $432(1+3) = 432 \times 4 = 1728$.
* Ah, that's a nice shortcut. $432 \times 4$.
* $432 \times 2 = 864$.
* $864 \times 2 = 1728$.
* Yes, calculation holds up.
* $8 \times 8 \times 8 = 512$.
* $512 + 43 + 1728 - 39$.
* $512 + 1728 = 2240$.
* $2240 + 43 = 2283$.
* $2283 - 39 = 2244$.
* Wait, $2240 + 43 - 39$.
* $43 - 39 = 4$.
* $2240 + 4 = 2244$.
* This confirms the previous calculation ($2283 - 39 = 2244$) but is slightly faster to verify.
6. **Construct Final Output:**
* State the expression.
* Show the order of operations steps clearly.
* Calculate the power ($8^3$).
* Calculate the multiplication inside parentheses.
* Add inside parentheses.
* Perform the final additions and subtractions.
* State the final answer clearly.
7. **Final Polish:**
* Expression: $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$
* Step 1: Parentheses. $432 \times 3 = 1296$. Then $432 + 1296 = 1728$.
* Step 2: Multiplication. $8 \times 8 \times 8 = 512$.
* Step 3: Remaining additions/subtractions. $512 + 43 + 1728 - 39$.
* $555 + 1728 - 39$.
* $2283 - 39 = 2244$.
Looks solid. Ready to write.
Wait, just a quick sanity check on the numbers.
$432 + 432 \times 3$.
$432 \times (1+3) = 432 \times 4$.
$400 \times 4 = 1600$.
$32 \times 4 = 128$.
$1600 + 128 = 1728$.
Correct.
$8^3 = 512$.
Correct.
$512 + 43 = 555$.
Correct.
$555 + 1728 = 2283$.
Correct.
$2283 - 39 = 2244$.
Correct.
All checks pass.
Wait, could the input be interpreted differently?
"8x8x8+43+(432+432x3)-39"
Is it possible "432" is a variable? No, context implies arithmetic.
Is it possible "8x8x8" is $8 \times 8 \times 8$? Yes.
Is it possible the user meant a specific math puzzle with a trick?
Usually, these are just straight arithmetic.
The number $432$ is specific.
$432 \times 3 = 1296$.
$432 + 1296 = 1728$.
$8^3 = 512$.
$512 + 43 + 1728 - 39$.
Maybe there is a pattern?
$1728 = 12^3$.
$512 = 8^3$.
$43 = 43$.
$-39$.
So $8^3 + 43 + 12^3 - 39$.
$8^3 + 12^3 + (43 - 39)$.
$512 + 1728 + 4$.
$2240 + 4 = 2244$.
Interesting, but not a "trick", just observation. The calculation stands.
Okay, proceed with generating the response.
Structure:
1. State order of operations (PEMDAS).
2. Solve parentheses.
3. Solve multiplication.
4. Solve addition/subtraction.
5. Final result.
Keep it clear and concise.
Wait, re-read the input string one more time to ensure no typos in my transcription.
"8x8x8+43+(432+432x3)-39"
My transcription: $8 \times 8 \times 8 + 43 + (432 + 432 \times 3) - 39$
Matches perfectly.
Ready.
</div>
To solve the expression **8x8x8+43+(432+432x3)-39**, we follow the order of operations (PEMDAS/BODMAS): Parentheses, Exponents/Multiplication, then Addition and Subtraction.
**Step 1: Simplify the parentheses**
Inside the parentheses, we have: `(432 + 432x3)`
Multiplication comes before addition.
* $432 \times 3 = 1296$
* $432 + 1296 = 1728$
Now the expression looks like this:
$8 \times 8 \times 8 + 43 + 1728 - 39$
**Step 2: Perform the multiplication**
* $8 \times 8 \times 8 = 512$
Now the expression looks like this:
$512 + 43 + 1728 - 39$
**Step 3: Perform addition and subtraction (left to right)**
* $512 + 43 = 555$
* $555 + 1728 = 2283$
* $2283 - 39 = 2244$
**Answer:**
**2244**
</details>
## π§ Training Details
* **Base Model:** `Qwen/Qwen3.5-4B`
* **Fine-tuning:** SFT (Supervised Fine-Tuning).
* **Training Tool:** **Unsloth** (for 4-bit optimized training).
## π Cool License
**Feel free to use or improve Nomi! Benchmark results are always welcome.**
---
|