Update README.md
Browse files
README.md
CHANGED
|
@@ -3,15 +3,10 @@ license: apache-2.0
|
|
| 3 |
language:
|
| 4 |
- en
|
| 5 |
base_model:
|
| 6 |
-
- MultiverseComputingCAI/Pulsar-16B
|
| 7 |
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
|
| 8 |
-
tags:
|
| 9 |
-
- fp8
|
| 10 |
-
- quantized
|
| 11 |
-
- compactifai
|
| 12 |
---
|
| 13 |
|
| 14 |
-
#
|
| 15 |
<div align="center">
|
| 16 |
|
| 17 |
[](https://opensource.org/licenses/Apache-2.0)
|
|
@@ -20,7 +15,7 @@ tags:
|
|
| 20 |
|
| 21 |
Powered by CompactifAI
|
| 22 |
|
| 23 |
-
**Optimized for Fast and Efficient Inference** · **Reduced Memory Footprint**
|
| 24 |
|
| 25 |
</div>
|
| 26 |
|
|
@@ -30,7 +25,6 @@ Powered by CompactifAI
|
|
| 30 |
|
| 31 |
- [Model Overview](#model-overview)
|
| 32 |
- [Key Characteristics](#key-characteristics)
|
| 33 |
-
- [Quantization](#quantization)
|
| 34 |
- [Quick Start](#quick-start)
|
| 35 |
- [Reasoning Control](#thinking-reasoning-control)
|
| 36 |
- [Tool Calling](#tool-calling)
|
|
@@ -45,10 +39,14 @@ Powered by CompactifAI
|
|
| 45 |
|
| 46 |
## Model Overview
|
| 47 |
|
| 48 |
-
**
|
| 49 |
|
| 50 |
-
|
| 51 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
---
|
| 53 |
|
| 54 |
## Key Characteristics
|
|
@@ -56,22 +54,12 @@ The BF16 release applies **model compression techniques** to significantly reduc
|
|
| 56 |
| Characteristic | Description |
|
| 57 |
|-----------------------|-------------|
|
| 58 |
| Base model | [nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16). **31.6B** total parameters, **3.6B** activated per forward pass (11.34% activation ratio). [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). |
|
| 59 |
-
|
|
| 60 |
-
| Pulsar (this model) | **16.15B** parameters with weights quantized to **FP8**. |
|
| 61 |
| 📐 **Architecture** | Hybrid Mamba2-Transformer with MoE (same family as the base checkpoint). |
|
| 62 |
| 🛠️ **Tool calling** | Yes. Same tool-call structure and format as [Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16). See [Tool Calling](#tool-calling). |
|
| 63 |
| 🗜️ **Compression** | CompactifAI (proprietary compression technology) |
|
| 64 |
-
| 🔢 **Quantization** | FP8 (weights). |
|
| 65 |
| Primary language | English |
|
| 66 |
-
|
| 67 |
-
---
|
| 68 |
-
|
| 69 |
-
## Quantization
|
| 70 |
-
|
| 71 |
-
This checkpoint is a quantized version of [Pulsar-16B](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B): its weights have been quantized to **FP8**.
|
| 72 |
-
|
| 73 |
---
|
| 74 |
-
|
| 75 |
## Quick Start
|
| 76 |
This model can be loaded with the **Transformers** API. Use `trust_remote_code=True`. Recommended approach: `AutoModelForCausalLM` with `apply_chat_template`. This configuration has been tested with Transformers 4.57.6.
|
| 77 |
|
|
@@ -79,7 +67,7 @@ This model can be loaded with the **Transformers** API. Use `trust_remote_code=T
|
|
| 79 |
import torch
|
| 80 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 81 |
|
| 82 |
-
model_id = "MultiverseComputingCAI/
|
| 83 |
|
| 84 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 85 |
model_id,
|
|
@@ -89,6 +77,7 @@ tokenizer = AutoTokenizer.from_pretrained(
|
|
| 89 |
model = AutoModelForCausalLM.from_pretrained(
|
| 90 |
model_id,
|
| 91 |
device_map="cuda" if torch.cuda.is_available() else "auto",
|
|
|
|
| 92 |
trust_remote_code=True,
|
| 93 |
)
|
| 94 |
messages = [
|
|
@@ -123,7 +112,7 @@ pip install -U "vllm>=0.12.0"
|
|
| 123 |
|
| 124 |
#### Reasoning parser (NVIDIA)
|
| 125 |
|
| 126 |
-
|
| 127 |
|
| 128 |
```bash
|
| 129 |
wget https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/resolve/main/nano_v3_reasoning_parser.py
|
|
@@ -134,7 +123,7 @@ You can keep any local filename; the `vllm serve` flags below assume the file is
|
|
| 134 |
#### Serve
|
| 135 |
|
| 136 |
```bash
|
| 137 |
-
vllm serve MultiverseComputingCAI/
|
| 138 |
--served-model-name model \
|
| 139 |
--max-num-seqs 8 \
|
| 140 |
--tensor-parallel-size 1 \
|
|
@@ -152,9 +141,9 @@ vllm serve MultiverseComputingCAI/Pulsar-16B-FP8 \
|
|
| 152 |
|
| 153 |
## Thinking (Reasoning) Control
|
| 154 |
|
| 155 |
-
|
| 156 |
|
| 157 |
-
> This section provides a brief overview of reasoning control in
|
| 158 |
|
| 159 |
|
| 160 |
---
|
|
@@ -210,7 +199,7 @@ Set the default for **all requests** at startup with `--default-chat-template-kw
|
|
| 210 |
|
| 211 |
**Thinking OFF for all requests**
|
| 212 |
```bash
|
| 213 |
-
vllm serve MultiverseComputingCAI/
|
| 214 |
--served-model-name model \
|
| 215 |
--reasoning-parser-plugin nano_v3_reasoning_parser.py \
|
| 216 |
--reasoning-parser nano_v3 \
|
|
@@ -221,7 +210,7 @@ vllm serve MultiverseComputingCAI/Pulsar-16B-FP8 \
|
|
| 221 |
|
| 222 |
**Thinking ON for all requests (default if flag is omitted)**
|
| 223 |
```bash
|
| 224 |
-
vllm serve MultiverseComputingCAI/
|
| 225 |
--served-model-name model \
|
| 226 |
--reasoning-parser-plugin nano_v3_reasoning_parser.py \
|
| 227 |
--reasoning-parser nano_v3 \
|
|
@@ -256,7 +245,7 @@ response = requests.post("http://localhost:8000/v1/chat/completions", json={
|
|
| 256 |
|
| 257 |
## Tool Calling
|
| 258 |
|
| 259 |
-
|
| 260 |
|
| 261 |
```
|
| 262 |
<tool_call>
|
|
@@ -287,61 +276,87 @@ The base model [nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.
|
|
| 287 |
|
| 288 |
CompactifAI was applied to produce a smaller, efficient model (16B parameters) while aiming to preserve reasoning and tool-use capabilities. Supervised Fine Tuning was applied for improving cabapilities.
|
| 289 |
|
| 290 |
-
### FP8 Quantization
|
| 291 |
-
|
| 292 |
-
After compression and fine-tuning, the BF16 checkpoint was quantized to FP8.
|
| 293 |
-
|
| 294 |
---
|
| 295 |
|
| 296 |
## Evaluation & Benchmarks
|
| 297 |
|
| 298 |

|
| 299 |
|
| 300 |
-
| Benchmark
|
| 301 |
-
|---------
|
| 302 |
-
|
|
| 303 |
-
|
|
| 304 |
-
| IFBench
|
| 305 |
-
|
|
| 306 |
-
|
|
| 307 |
-
|
| 308 |
-
|
| 309 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 310 |
|
| 311 |
### Performance
|
| 312 |

|
| 313 |
- **Framework:** [guidellm](https://github.com/vllm-project/guidellm)
|
| 314 |
- **Inference:** vLLM 0.18.0
|
| 315 |
-
- **
|
| 316 |
-
- **
|
| 317 |
-
- **
|
| 318 |
-
- **
|
| 319 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 320 |
### Evaluation Methodology
|
| 321 |
|
| 322 |
Benchmark scores were obtained with the following setups. Methodology varies by benchmark family.
|
| 323 |
|
| 324 |
-
###
|
| 325 |
-
|
| 326 |
-
- **
|
| 327 |
-
- **
|
| 328 |
-
- **
|
| 329 |
-
|
| 330 |
-
|
| 331 |
-
|
| 332 |
-
-
|
| 333 |
-
-
|
| 334 |
-
|
| 335 |
-
|
| 336 |
-
|
| 337 |
-
|
| 338 |
-
|
| 339 |
-
|
| 340 |
-
|
| 341 |
-
|
| 342 |
-
- **Inference**: vLLM 0.18.0
|
| 343 |
-
- **Decode**: temperature 1.0, top_p 1.0
|
| 344 |
-
- **Decode (judge)**: judge top_p 0.8, top_k 20, min_p 0.0; pattern scoring.
|
| 345 |
|
| 346 |
---
|
| 347 |
|
|
@@ -360,7 +375,7 @@ Trained mainly on English with added Spanish. No systematic evaluation for langu
|
|
| 360 |
|
| 361 |
- English-centric training data (inherited from base model).
|
| 362 |
- Tool calling depends on correct schema and tool design; exact parity with the original model is not guaranteed.
|
| 363 |
-
- Compression
|
| 364 |
|
| 365 |
### Recommendations
|
| 366 |
|
|
@@ -374,9 +389,8 @@ Trained mainly on English with added Spanish. No systematic evaluation for langu
|
|
| 374 |
|
| 375 |
| Field | Value |
|
| 376 |
|--------------|--------------------- |
|
| 377 |
-
| Model name |
|
| 378 |
-
| Based on | [
|
| 379 |
-
| Precision | FP8 weights |
|
| 380 |
| Version | v1.5.0 |
|
| 381 |
| Release date | TBD |
|
| 382 |
| Developed by | Multiverse Computing |
|
|
@@ -387,7 +401,7 @@ Trained mainly on English with added Spanish. No systematic evaluation for langu
|
|
| 387 |
|
| 388 |
## Citation
|
| 389 |
|
| 390 |
-
If you use this model, please cite the base model and
|
| 391 |
|
| 392 |
```bibtex
|
| 393 |
@misc{nemotron3nanoTR,
|
|
@@ -396,20 +410,13 @@ If you use this model, please cite the base model and pulsar:
|
|
| 396 |
year = {2025},
|
| 397 |
url = {https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Nano-Technical-Report.pdf}
|
| 398 |
}
|
| 399 |
-
@misc{
|
| 400 |
-
title = {
|
| 401 |
author = {Multiverse Computing},
|
| 402 |
year = {2026},
|
| 403 |
-
url = {https://huggingface.co/MultiverseComputingCAI/
|
| 404 |
note = {Model developed based on nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 using CompactifAI technology}
|
| 405 |
}
|
| 406 |
-
@misc{pulsar16bfp8,
|
| 407 |
-
title = {Pulsar 16B FP8: FP8-quantized release of Pulsar-16B},
|
| 408 |
-
author = {Multiverse Computing},
|
| 409 |
-
year = {2026},
|
| 410 |
-
url = {https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-FP8},
|
| 411 |
-
note = {FP8 weight quantization of Pulsar-16B}
|
| 412 |
-
}
|
| 413 |
```
|
| 414 |
|
| 415 |
**Built by [Multiverse Computing](https://www.multiversecomputing.com)** · [Report an issue](TODO_PULSAR_HF_URL/discussions) · [Discord](https://discord.gg/cGas9uStqp)
|
|
|
|
| 3 |
language:
|
| 4 |
- en
|
| 5 |
base_model:
|
|
|
|
| 6 |
- nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
---
|
| 8 |
|
| 9 |
+
# Nemotron 3 Nano Slim 16B
|
| 10 |
<div align="center">
|
| 11 |
|
| 12 |
[](https://opensource.org/licenses/Apache-2.0)
|
|
|
|
| 15 |
|
| 16 |
Powered by CompactifAI
|
| 17 |
|
| 18 |
+
**Optimized for Fast and Efficient Inference** · **Reduced Memory Footprint**
|
| 19 |
|
| 20 |
</div>
|
| 21 |
|
|
|
|
| 25 |
|
| 26 |
- [Model Overview](#model-overview)
|
| 27 |
- [Key Characteristics](#key-characteristics)
|
|
|
|
| 28 |
- [Quick Start](#quick-start)
|
| 29 |
- [Reasoning Control](#thinking-reasoning-control)
|
| 30 |
- [Tool Calling](#tool-calling)
|
|
|
|
| 39 |
|
| 40 |
## Model Overview
|
| 41 |
|
| 42 |
+
**Nemotron 3 Nano Slim 16B** is a **model based on [NVIDIA-Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16)**, developed by **Multiverse Computing**. The original model is a **~31.6B parameter**, part of the Nemotron model family. It supports **long-context inference up to 1M tokens** and is designed for general-purpose language modeling tasks.
|
| 43 |
|
| 44 |
+
This version applies **model compression techniques** to significantly reduce parameter count and deployment requirements while maintaining compatibility with the Nemotron Hybrid Mamba2-Transformer with MoE architecture. The resulting model achieves **50% compression**, reducing the parameter count to **16.15B parameters** and lowering memory requirements.
|
| 45 |
|
| 46 |
+
- [BF16](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16)
|
| 47 |
+
- [FP8](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-FP8)
|
| 48 |
+
- [NVFP4](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-NVFP4)
|
| 49 |
+
|
| 50 |
---
|
| 51 |
|
| 52 |
## Key Characteristics
|
|
|
|
| 54 |
| Characteristic | Description |
|
| 55 |
|-----------------------|-------------|
|
| 56 |
| Base model | [nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16). **31.6B** total parameters, **3.6B** activated per forward pass (11.34% activation ratio). [NVIDIA Open Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-open-model-license/). |
|
| 57 |
+
| Nemotron-3-Nano-Slim-16B-BF16 (this model) | **16.15B** total parameters, **3.1B** activated per forward pass (19.28% activation ratio) after CompactifAI compression. |
|
|
|
|
| 58 |
| 📐 **Architecture** | Hybrid Mamba2-Transformer with MoE (same family as the base checkpoint). |
|
| 59 |
| 🛠️ **Tool calling** | Yes. Same tool-call structure and format as [Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16). See [Tool Calling](#tool-calling). |
|
| 60 |
| 🗜️ **Compression** | CompactifAI (proprietary compression technology) |
|
|
|
|
| 61 |
| Primary language | English |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 62 |
---
|
|
|
|
| 63 |
## Quick Start
|
| 64 |
This model can be loaded with the **Transformers** API. Use `trust_remote_code=True`. Recommended approach: `AutoModelForCausalLM` with `apply_chat_template`. This configuration has been tested with Transformers 4.57.6.
|
| 65 |
|
|
|
|
| 67 |
import torch
|
| 68 |
from transformers import AutoTokenizer, AutoModelForCausalLM
|
| 69 |
|
| 70 |
+
model_id = "MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16"
|
| 71 |
|
| 72 |
tokenizer = AutoTokenizer.from_pretrained(
|
| 73 |
model_id,
|
|
|
|
| 77 |
model = AutoModelForCausalLM.from_pretrained(
|
| 78 |
model_id,
|
| 79 |
device_map="cuda" if torch.cuda.is_available() else "auto",
|
| 80 |
+
torch_dtype=torch.bfloat16,
|
| 81 |
trust_remote_code=True,
|
| 82 |
)
|
| 83 |
messages = [
|
|
|
|
| 112 |
|
| 113 |
#### Reasoning parser (NVIDIA)
|
| 114 |
|
| 115 |
+
Nemotron 3 Nano Slim 16B uses the same Nemotron v3 reasoning tags as the base model. NVIDIA provides the vLLM plugin as [`nano_v3_reasoning_parser.py`](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/blob/main/nano_v3_reasoning_parser.py) on the base Hugging Face repo (not specific to Pulsar). Direct download:
|
| 116 |
|
| 117 |
```bash
|
| 118 |
wget https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16/resolve/main/nano_v3_reasoning_parser.py
|
|
|
|
| 123 |
#### Serve
|
| 124 |
|
| 125 |
```bash
|
| 126 |
+
vllm serve MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16 \
|
| 127 |
--served-model-name model \
|
| 128 |
--max-num-seqs 8 \
|
| 129 |
--tensor-parallel-size 1 \
|
|
|
|
| 141 |
|
| 142 |
## Thinking (Reasoning) Control
|
| 143 |
|
| 144 |
+
Nemotron 3 Nano Slim 16B supports a **hybrid reasoning mode**: the model can either think step-by-step before answering (reasoning mode) or reply directly (non-reasoning mode). The behaviour is controlled via the `enable_thinking` flag in the chat template.
|
| 145 |
|
| 146 |
+
> This section provides a brief overview of reasoning control in Nemotron 3 Nano Slim 16B. For comprehensive details please see the official Nemotron-3 Nano-30B model card at: https://build.nvidia.com/nvidia/nemotron-3-nano-30b-a3b/modelcard
|
| 147 |
|
| 148 |
|
| 149 |
---
|
|
|
|
| 199 |
|
| 200 |
**Thinking OFF for all requests**
|
| 201 |
```bash
|
| 202 |
+
vllm serve MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16 \
|
| 203 |
--served-model-name model \
|
| 204 |
--reasoning-parser-plugin nano_v3_reasoning_parser.py \
|
| 205 |
--reasoning-parser nano_v3 \
|
|
|
|
| 210 |
|
| 211 |
**Thinking ON for all requests (default if flag is omitted)**
|
| 212 |
```bash
|
| 213 |
+
vllm serve MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16 \
|
| 214 |
--served-model-name model \
|
| 215 |
--reasoning-parser-plugin nano_v3_reasoning_parser.py \
|
| 216 |
--reasoning-parser nano_v3 \
|
|
|
|
| 245 |
|
| 246 |
## Tool Calling
|
| 247 |
|
| 248 |
+
Nemotron 3 Nano Slim 16B emits tool calls in the following format:
|
| 249 |
|
| 250 |
```
|
| 251 |
<tool_call>
|
|
|
|
| 276 |
|
| 277 |
CompactifAI was applied to produce a smaller, efficient model (16B parameters) while aiming to preserve reasoning and tool-use capabilities. Supervised Fine Tuning was applied for improving cabapilities.
|
| 278 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 279 |
---
|
| 280 |
|
| 281 |
## Evaluation & Benchmarks
|
| 282 |
|
| 283 |

|
| 284 |
|
| 285 |
+
| Benchmark | Nemotron 3 Nano 30B A3B | Nemotron 3 Nano Slim 16B | gpt-oss-20b | Qwen3-14B | Ministral-3-14B-Instruct-2512 |
|
| 286 |
+
| --- | ---: | ---: | ---: | ---: | ---: |
|
| 287 |
+
| AIME | 87.66 | 87.22 | 87.66 | 76.00 | 33.00 |
|
| 288 |
+
| GPQA | 74.04 | 71.41 | 68.99 | 63.63 | 56.45 |
|
| 289 |
+
| IFBench | 72.31 | 70.79 | 68.46 | 39.20 | 32.80 |
|
| 290 |
+
| MMLU-Pro | 78.90 | 74.78 | 76.65 | 85.01 | 70.09 |
|
| 291 |
+
| LiveCodeBench | 71.11 | 68.04 | 64.65 | 66.35 | 29.84 |
|
| 292 |
+
|
| 293 |
+
### Quantizations
|
| 294 |
+
|
| 295 |
+
- [BF16](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16)
|
| 296 |
+
- [FP8](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-FP8)
|
| 297 |
+
- [NVFP4](https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-NVFP4)
|
| 298 |
+
|
| 299 |
+

|
| 300 |
+
|
| 301 |
+
| Benchmark | Nemotron 3 Nano 30B A3B | Nemotron 3 Nano Slim 16B (BF16) | Nemotron 3 Nano Slim 16B (fp8) | Nemotron 3 Nano Slim 16B (nvfp4) |
|
| 302 |
+
| --- | ---: | ---: | ---: | ---: |
|
| 303 |
+
| AIME | 87.66 | 87.22 | 86.67 | 82.00 |
|
| 304 |
+
| GPQA | 74.04 | 71.41 | 70.61 | 71.11 |
|
| 305 |
+
| IFBench | 72.31 | 70.79 | 69.60 | 69.90 |
|
| 306 |
+
| MMLU-Pro | 78.90 | 74.78 | 74.76 | 74.19 |
|
| 307 |
+
| LiveCodeBench | 71.11 | 68.04 | 68.68 | 65.60 |
|
| 308 |
+
|
| 309 |
|
| 310 |
### Performance
|
| 311 |

|
| 312 |
- **Framework:** [guidellm](https://github.com/vllm-project/guidellm)
|
| 313 |
- **Inference:** vLLM 0.18.0
|
| 314 |
+
- **GPU:** NVIDIA B200
|
| 315 |
+
- **Decode:** `temperature: 0.0`, `top_p: 1.0`
|
| 316 |
+
- **Measure Window:** Each phase lasts 3 minutes (excluding ramp-up and cool-down periods).
|
| 317 |
+
- **Workload shape:** 8k/16k workload as in the original model's card.
|
| 318 |
+
|
| 319 |
+
| Model | System throughput (tok/s, p50) | TTFT (s, p50) | Model weights (GB) |
|
| 320 |
+
| :--- | ---: | ---: | ---: |
|
| 321 |
+
| Nemotron 3 Nano 30B A3B | 3,363 | 2.18 | 59 |
|
| 322 |
+
| Nemotron 3 Nano Slim 16B (BF16) | 3,760 | 1.80 | 30 |
|
| 323 |
+
| Nemotron 3 Nano Slim 16B (fp8) | 4,808 | 1.24 | 16 |
|
| 324 |
+
| Nemotron 3 Nano Slim 16B (nvfp4) | 4,735 | 1.25 | 10 |
|
| 325 |
+
|
| 326 |
+
### Long Context
|
| 327 |
+
Nemotron 3 Nano Slim 16B preserves strong long-context behavior after compression, tracking the Nemotron-3-Nano-30B-A3B baseline closely across retrieval-heavy and full-suite long-context evaluations. Results are reported for LongBench v1, AA-LCR, NIAH, and RULER groupings up to 256k context.
|
| 328 |
+
|
| 329 |
+

|
| 330 |
+
|
| 331 |
+
| Benchmark | Nemotron 3 Nano 30B A3B | Nemotron 3 Nano Slim 16B |
|
| 332 |
+
| :--- | ---: | ---: |
|
| 333 |
+
| Longbench | 31.84 | 29.84 |
|
| 334 |
+
| AA-LCR | 33.67 | 29.33 |
|
| 335 |
+
| NIAH (@100K) | 100.00 | 100.00 |
|
| 336 |
+
| RULER (@128K) | 95.02 | 94.20 |
|
| 337 |
+
| RULER (@256K) | 92.02 | 87.74 |
|
| 338 |
### Evaluation Methodology
|
| 339 |
|
| 340 |
Benchmark scores were obtained with the following setups. Methodology varies by benchmark family.
|
| 341 |
|
| 342 |
+
### Inference:
|
| 343 |
+
- **Backend:** VLLM 0.18.0
|
| 344 |
+
- **Nemotron models:** `temp 1.0`, `top_p 1.0`
|
| 345 |
+
- **GPT-OSS-20B:** `temp: 1.0`, `top_p: 1.0`, `reasoning_effort: high`
|
| 346 |
+
- **Qwen3-14B:** `temp: 0.6`, `top_p: 0.95`, `top_k: 20`, `min_p: 0.0`
|
| 347 |
+
- **Ministral-3-14B-Instruct-2512:** `temp: 0.15`
|
| 348 |
+
|
| 349 |
+
| Benchmark | Framework | Repeats | Other |
|
| 350 |
+
|-----------|-----------|--------:|-------|
|
| 351 |
+
| MMLU-Pro | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) | 1 | |
|
| 352 |
+
| AIME25 | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) | 10 | |
|
| 353 |
+
| GPQA:d | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) | 5 | |
|
| 354 |
+
| LiveCodeBench | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) | 3 | |
|
| 355 |
+
| IFBench | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) | 5 | |
|
| 356 |
+
| LongBench v1 | [lm-evaluation-harness](https://github.com/EleutherAI/lm-evaluation-harness) | 1 | |
|
| 357 |
+
| AA-LCR | [EvalScope](https://github.com/modelscope/evalscope) 1.4.1 | 3 | Judge: **`Qwen/Qwen3-235B-A22B-Instruct-2507`**. **`judge_score_type`:** `pattern`. **`judge_args` → `generation_config`:** `top_p` 0.8, `top_k` 20, `min_p` 0.0, `temperature` 0.7. |
|
| 358 |
+
| NIAH | [EvalScope](https://github.com/modelscope/evalscope) 1.4.1 | 1 | Judge: **`qwen/qwen3-235b-a22b-2507`** . **`judge_model_args`:** `{}` (no extra judge settings in YAML). |
|
| 359 |
+
| RULER | [NeMo-Skills](https://github.com/NVIDIA-NeMo/Skills) (+ [RULER](https://github.com/NVIDIA/RULER)) | 1 | |
|
|
|
|
|
|
|
|
|
|
| 360 |
|
| 361 |
---
|
| 362 |
|
|
|
|
| 375 |
|
| 376 |
- English-centric training data (inherited from base model).
|
| 377 |
- Tool calling depends on correct schema and tool design; exact parity with the original model is not guaranteed.
|
| 378 |
+
- Compression may affect some behaviors; evaluate for your use case.
|
| 379 |
|
| 380 |
### Recommendations
|
| 381 |
|
|
|
|
| 389 |
|
| 390 |
| Field | Value |
|
| 391 |
|--------------|--------------------- |
|
| 392 |
+
| Model name | Nemotron 3 Nano Slim 16B |
|
| 393 |
+
| Based on | [NVIDIA-Nemotron-3-Nano-30B-A3B-BF16](https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16) |
|
|
|
|
| 394 |
| Version | v1.5.0 |
|
| 395 |
| Release date | TBD |
|
| 396 |
| Developed by | Multiverse Computing |
|
|
|
|
| 401 |
|
| 402 |
## Citation
|
| 403 |
|
| 404 |
+
If you use this model, please cite the base model and Nemotron 3 Nano Slim 16B:
|
| 405 |
|
| 406 |
```bibtex
|
| 407 |
@misc{nemotron3nanoTR,
|
|
|
|
| 410 |
year = {2025},
|
| 411 |
url = {https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Nano-Technical-Report.pdf}
|
| 412 |
}
|
| 413 |
+
@misc{nemotron3nanoslim16b,
|
| 414 |
+
title = {Nemotron 3 Nano Slim 16B: Model developed from NVIDIA Nemotron-3-Nano-30B-A3B},
|
| 415 |
author = {Multiverse Computing},
|
| 416 |
year = {2026},
|
| 417 |
+
url = {https://huggingface.co/MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16},
|
| 418 |
note = {Model developed based on nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 using CompactifAI technology}
|
| 419 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 420 |
```
|
| 421 |
|
| 422 |
**Built by [Multiverse Computing](https://www.multiversecomputing.com)** · [Report an issue](TODO_PULSAR_HF_URL/discussions) · [Discord](https://discord.gg/cGas9uStqp)
|