arturo-fredes commited on
Commit
c06dca2
·
verified ·
1 Parent(s): f097522

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +28 -28
README.md CHANGED
@@ -6,7 +6,7 @@ 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
  [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
@@ -39,13 +39,13 @@ Powered by CompactifAI
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
 
@@ -54,7 +54,7 @@ This version applies **model compression techniques** to significantly reduce pa
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) |
@@ -67,7 +67,7 @@ This model can be loaded with the **Transformers** API. Use `trust_remote_code=T
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,
@@ -112,7 +112,7 @@ pip install -U "vllm>=0.12.0"
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
@@ -123,7 +123,7 @@ You can keep any local filename; the `vllm serve` flags below assume the file is
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 \
@@ -141,9 +141,9 @@ vllm serve MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16 \
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
  ---
@@ -199,7 +199,7 @@ Set the default for **all requests** at startup with `--default-chat-template-kw
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 \
@@ -210,7 +210,7 @@ vllm serve MultiverseComputingCAI/Nemotron-3-Nano-Slim-16B-BF16 \
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 \
@@ -245,7 +245,7 @@ response = requests.post("http://localhost:8000/v1/chat/completions", json={
245
 
246
  ## Tool Calling
247
 
248
- Nemotron 3 Nano Slim 16B emits tool calls in the following format:
249
 
250
  ```
251
  <tool_call>
@@ -282,7 +282,7 @@ CompactifAI was applied to produce a smaller, efficient model (16B parameters) w
282
 
283
  ![Combined benchmark chart](assets/benchmarks.png)
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 |
@@ -292,13 +292,13 @@ CompactifAI was applied to produce a smaller, efficient model (16B parameters) w
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
  ![Quantization results](assets/quantization_comparisons.png)
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 |
@@ -319,16 +319,16 @@ CompactifAI was applied to produce a smaller, efficient model (16B parameters) w
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
  ![Long-context benchmark results](assets/long_context_comparison.png)
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 |
@@ -389,7 +389,7 @@ Trained mainly on English with added Spanish. No systematic evaluation for langu
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 |
@@ -401,7 +401,7 @@ Trained mainly on English with added Spanish. No systematic evaluation for langu
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,
@@ -411,10 +411,10 @@ If you use this model, please cite the base model and Nemotron 3 Nano Slim 16B:
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
  ```
 
6
  - nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16
7
  ---
8
 
9
+ # Pulsar 16B
10
  <div align="center">
11
 
12
  [![License](https://img.shields.io/badge/License-Apache_2.0-blue.svg)](https://opensource.org/licenses/Apache-2.0)
 
39
 
40
  ## Model Overview
41
 
42
+ **Pulsar 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/Pulsar-16B-BF16)
47
+ - [FP8](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-FP8)
48
+ - [NVFP4](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-NVFP4)
49
 
50
  ---
51
 
 
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
+ | Pulsar-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) |
 
67
  import torch
68
  from transformers import AutoTokenizer, AutoModelForCausalLM
69
 
70
+ model_id = "MultiverseComputingCAI/Pulsar-16B-BF16"
71
 
72
  tokenizer = AutoTokenizer.from_pretrained(
73
  model_id,
 
112
 
113
  #### Reasoning parser (NVIDIA)
114
 
115
+ Pulsar 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/Pulsar-16B-BF16 \
127
  --served-model-name model \
128
  --max-num-seqs 8 \
129
  --tensor-parallel-size 1 \
 
141
 
142
  ## Thinking (Reasoning) Control
143
 
144
+ Pulsar 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 Pulsar 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/Pulsar-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/Pulsar-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
+ Pulsar 16B emits tool calls in the following format:
249
 
250
  ```
251
  <tool_call>
 
282
 
283
  ![Combined benchmark chart](assets/benchmarks.png)
284
 
285
+ | Benchmark | Nemotron 3 Nano 30B A3B | Pulsar 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 |
 
292
 
293
  ### Quantizations
294
 
295
+ - [BF16](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-BF16)
296
+ - [FP8](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-FP8)
297
+ - [NVFP4](https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-NVFP4)
298
 
299
  ![Quantization results](assets/quantization_comparisons.png)
300
 
301
+ | Benchmark | Nemotron 3 Nano 30B A3B | Pulsar 16B (BF16) | Pulsar 16B (fp8) | Pulsar 16B (nvfp4) |
302
  | --- | ---: | ---: | ---: | ---: |
303
  | AIME | 87.66 | 87.22 | 86.67 | 82.00 |
304
  | GPQA | 74.04 | 71.41 | 70.61 | 71.11 |
 
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
+ | Pulsar 16B (BF16) | 3,760 | 1.80 | 30 |
323
+ | Pulsar 16B (fp8) | 4,808 | 1.24 | 16 |
324
+ | Pulsar 16B (nvfp4) | 4,735 | 1.25 | 10 |
325
 
326
  ### Long Context
327
+ Pulsar 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
  ![Long-context benchmark results](assets/long_context_comparison.png)
330
 
331
+ | Benchmark | Nemotron 3 Nano 30B A3B | Pulsar 16B |
332
  | :--- | ---: | ---: |
333
  | Longbench | 31.84 | 29.84 |
334
  | AA-LCR | 33.67 | 29.33 |
 
389
 
390
  | Field | Value |
391
  |--------------|--------------------- |
392
+ | Model name | Pulsar 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 |
 
401
 
402
  ## Citation
403
 
404
+ If you use this model, please cite the base model and Pulsar 16B:
405
 
406
  ```bibtex
407
  @misc{nemotron3nanoTR,
 
411
  url = {https://research.nvidia.com/labs/nemotron/files/NVIDIA-Nemotron-3-Nano-Technical-Report.pdf}
412
  }
413
  @misc{nemotron3nanoslim16b,
414
+ title = {Pulsar 16B: Model developed from NVIDIA Nemotron-3-Nano-30B-A3B},
415
  author = {Multiverse Computing},
416
  year = {2026},
417
+ url = {https://huggingface.co/MultiverseComputingCAI/Pulsar-16B-BF16},
418
  note = {Model developed based on nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 using CompactifAI technology}
419
  }
420
  ```