NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16

Description

NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16 is a small language model (SLM) further post-trained from Gemma-4-E4B-it by NVIDIA. This is the full-precision (BF16) release. Nemotron-3.5-Bolt-E4B-4GB carries forward Gemma’s strong conversational capabilities and further improves multi-turn tool use, instruction following, and input grounding for fast local voice/device assistants, interactive personas, and embedded/Edge interfaces. Designed for low-memory, low-TTFT operation, the model allows agents to respond quickly while preserving headroom for applications or adjacent model pipelines.

This model is ready for commercial or non‑commercial use.

Model Summary

Model Overview

Model Developer: NVIDIA Corporation

Model Dates: December 2025 - May 2026

Data Freshness:

• The post-training data has a cutoff date of May 2026.

What is Nemotron?

NVIDIA Nemotron™ is a family of open models with open weights, training data, and recipes, delivering leading efficiency and accuracy for building specialized AI agents.

Quick Start

To get quickly started on a single H100 you can use the following command.

Run it with vLLM! (vLLM v0.31.0: vllm/vllm-openai:v0.31.0)

Without Speculative Decoding

export MODEL=nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16

vllm serve "$MODEL" \
  --served-model-name "nemotron-bolt" \
  --host 0.0.0.0 \
  --port 8000 \
  --dtype bfloat16 \
  --tensor-parallel-size 1 \
  --gpu-memory-utilization 0.85 \
  --kv-cache-dtype auto \
  --max-model-len 65536 \
  --max-num-seqs 64 \
  --max-num-batched-tokens 65536 \
  --chat-template "$MODEL/chat_template.jinja" \
  --enable-auto-tool-choice \
  --tool-call-parser gemma4 \
  --reasoning-parser gemma4 \
  --limit-mm-per-prompt '{"image":0,"video":0,"audio":0}'

With DFlash Speculative Decoding

Use the same downloaded BF16 checkpoint with its companion DFlash drafter. Run one serving configuration at a time. This example uses a single concurrent sequence and seven speculative tokens; tune these settings for your workload.

export MODEL=nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16
export DRAFTER_CKPT=nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-DFlash-BF16

vllm serve "$MODEL" \
  --served-model-name "nemotron-bolt" \
  --host 0.0.0.0 \
  --port 8000 \
  --dtype bfloat16 \
  --tensor-parallel-size 1 \
  --gpu-memory-utilization 0.85 \
  --kv-cache-dtype auto \
  --max-model-len 65536 \
  --max-num-seqs 1 \
  --max-num-batched-tokens 65536 \
  --chat-template "$MODEL/chat_template.jinja" \
  --enable-auto-tool-choice \
  --tool-call-parser gemma4 \
  --reasoning-parser gemma4 \
  --limit-mm-per-prompt '{"image":0,"video":0,"audio":0}' \
  --speculative-config "{
    \"method\": \"dflash\",
    \"model\": \"$DRAFTER_CKPT\",
    \"num_speculative_tokens\": 7
  }"

Sample Request

The same request works with either serving configuration.

curl -s "http://<host>:8000/v1/chat/completions" \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "nemotron-bolt",
    "messages": [
      {"role": "user", "content": "What is 17 multiplied by 23?"}
    ],
    "temperature": 0.0,
    "max_tokens": 512,
    "skip_special_tokens": false,
    "chat_template_kwargs": {
      "enable_thinking": true
    }
  }'

For more details on how to deploy and use the model — see the Quick Start Guide below!

License/Terms of Use

GOVERNING DOWNLOAD TERMS: Use of the model is governed by the OpenMDW License Agreement, version 1.1.

ADDITIONAL INFORMATION: Apache License, Version 2.0.

Benchmarks

Reasoning Benchmark Evaluations

We evaluated our model on the following benchmarks:

Accuracy numbers are measured by NVIDIA under a consistent harness (NeMo Gym / Nemo Evaluator SDK); they may differ from vendors' self-reported numbers.

TODO: Placeholder (Accuracy table)

Deployment Geography: Global

Use Case

NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16 is the full-precision release of a general purpose reasoning and chat model, and is intended primarily for customization and post-training rather than direct production inference. It is intended to be used by developers who want to: post-train the model on their own data (SFT, RL via NeMo RL and NeMo Gym, or distillation), adapt it to a domain or task, produce quantized variants (NVFP4, W4A16, GGUF) for their own deployment targets, or run full-precision research and evaluation. English and coding languages are the primary languages, with Spanish, French, German, Italian, and Japanese also supported.

Release Date

Hugging Face 10/23/2026 via https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16

Model Architecture

• Architecture Type: Gemma 4
• Network Architecture: Gemma-4-E4B
• Number of model parameters: 8B (4B effective parameters)

Model Design

The model was further post-trained from Gemma-4-E4B-it.

Training Methodology

Reinforcement Learning

• The model underwent multi-environment reinforcement learning using GRPO (Group Relative Policy Optimization) across instruction following, multi-step tool use, multi-turn conversations, and structured output environments. It utilized an asynchronous RL architecture that decouples training from inference.
• Software used for reinforcement learning: NeMo RL, NeMo Gym

NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16 is a result of the above work.

Input

• Input Type(s): Text
• Input Format(s): String
• Input Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Input: Maximum context length up to 1M tokens. Supported languages include English, Spanish, French, German, Italian, and Japanese.

Output

• Output Type(s): Text
• Output Format: String
• Output Parameters: One-Dimensional (1D): Sequences
• Other Properties Related to Output: Maximum context length up to 128K tokens

Our AI models are designed and/or optimized to run on NVIDIA GPU-accelerated systems. By leveraging NVIDIA's hardware (e.g. GPU cores) and software frameworks (e.g., CUDA libraries), the model achieves faster training and inference times compared to CPU-only solutions.

Software Integration

• Runtime Engine(s): PyTorch
• Supported Hardware Microarchitecture Compatibility: NVIDIA Ampere - A100; NVIDIA Blackwell; NVIDIA Hopper - H100-80GB
• Supported Operating System(s): Linux

The integration of foundation and fine-tuned models into AI systems requires additional testing using use-case-specific data to ensure safe and effective deployment. Following the V-model methodology, iterative testing and validation at both unit and system levels are essential to mitigate risks, meet technical and functional requirements, and ensure compliance with safety and ethical standards before deployment.

Model Version(s)

• GA (10/23/2026)

Quick Start Guide

All deployment snippets below assume:

export MODEL_CKPT=nvidia/Gemma-Nemotron-4B-BF16

vLLM

• vLLM: vllm/vllm-openai:v0.31.0

1x H100 / A100 / GB200

For max throughput deployments, use the following configuration, no speculative decoding strategy is best for this serving configuration, and due to memory constraints the Mamba cache dtype is set as FP16:

vllm serve "$MODEL" \
  --served-model-name "nemotron-bolt" \
  --host 0.0.0.0 \
  --port 8000 \
  --dtype bfloat16 \
  --tensor-parallel-size 1 \
  --gpu-memory-utilization 0.85 \
  --kv-cache-dtype auto \
  --max-model-len 65536 \
  --max-num-seqs 64 \
  --max-num-batched-tokens 65536 \
  --chat-template "$MODEL/chat_template.jinja" \
  --enable-auto-tool-choice \
  --tool-call-parser gemma4 \
  --reasoning-parser gemma4 \
  --limit-mm-per-prompt '{"image":0,"video":0,"audio":0}'

API Client

The examples below use the OpenAI-compatible client and work with the serving backend above. Recommended sampling settings are Temperature 1.0 and Top_P 0.95 in both thinking enabled and disabled modes.

from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
MODEL = "nemotron-bolt"

Gemma-Nemotron exposes reasoning control through chat-template kwargs: thinking enabled (the default), thinking disabled for direct answers, and a runtime thinking budget.

Reasoning On

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "Write a haiku about GPUs"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={
        "skip_special_tokens": False,
        "chat_template_kwargs": {"enable_thinking": True},
    }
)
print(response.choices[0].message.content)

Reasoning Off

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "What is the capital of Japan?"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={
        "skip_special_tokens": False,
        "chat_template_kwargs": {"enable_thinking": False},
    }
)
print(response.choices[0].message.content)

Streaming

stream = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "Explain speculative decoding in two sentences"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    stream=True,
    extra_body={"skip_special_tokens": False},
)
for chunk in stream:
    print(chunk.choices[0].delta.content or "", end="", flush=True)

Tool Calling

For vLLM, add the following to any serve command above:

--enable-auto-tool-choice \
--tool-call-parser gemma4 \
--reasoning-parser gemma4 \

Sample request with tool-calling enabled:

tools = [{
    "type": "function",
    "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "parameters": {
            "type": "object",
            "properties": {"city": {"type": "string"}},
            "required": ["city"],
        },
    },
}]

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "What's the weather in Santa Clara?"}],
    tools=tools,
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={
        "skip_special_tokens": False,
        "chat_template_kwargs": {"enable_thinking": True},
    },
)
print(response.choices[0].message.tool_calls)

Conversational Mode

Gemma-Nemotron supports a conversational mode. If enabled, Gemma-Nemotron responds concisely and without as much textual formatting. This mode is suitable for conversational and spoken interaction-based applications. Conversational mode can be toggled as follows:

response = client.chat.completions.create(
    model=MODEL,
    messages=[{"role": "user", "content": "What is the capital of Japan?"}],
    max_tokens=16000,
    temperature=1.0,
    top_p=0.95,
    extra_body={
        "skip_special_tokens": False,
        "chat_template_kwargs": {"conversational_mode": True},
    }
)
print(response.choices[0].message.content)

Conversational mode can be combined with thinking enabled and disabled mode, as well as tool-calling.

Runtime Thinking Budget Control

import copy
import requests

class GemmaNemotronThinkingBudgetClient:
    def __init__(self, base_url):
        self.base_url = base_url.rstrip("/")

    def _post(self, payload):
        response = requests.post(
            f"{self.base_url}/chat/completions",
            json=payload,
            timeout=300,
        )
        response.raise_for_status()
        return response.json()

    def chat_completion(
        self,
        model,
        messages,
        max_thinking_budget=512,
        max_tokens=1024,
        tools=None,
        temperature=1.0,
        top_p=0.95,
        seed=None,
    ):
        if max_tokens <= max_thinking_budget:
            raise ValueError(
                "max_tokens must be larger than max_thinking_budget"
            )

        original_messages = copy.deepcopy(messages)
        sampling = {
            "temperature": temperature,
            "top_p": top_p,
        }
        if seed is not None:
            sampling["seed"] = seed

        first_payload = {
            "model": model,
            "skip_special_tokens": False,
            "messages": original_messages,
            "max_tokens": max_thinking_budget,
            "chat_template_kwargs": {"enable_thinking": True},
            **sampling,
        }
        if tools is not None:
            first_payload["tools"] = tools
            first_payload["tool_choice"] = "auto"

        first = self._post(first_payload)
        choice = first["choices"][0]
        message = choice["message"]
        reasoning = (
            message.get("reasoning_content")
            or message.get("reasoning")
            or ""
        )
        partial_answer = message.get("content") or ""
        tool_calls = message.get("tool_calls") or []

        if choice.get("finish_reason") == "stop" and (
            partial_answer or tool_calls
        ):
            return {
                "reasoning_content": reasoning,
                "content": partial_answer,
                "tool_calls": tool_calls,
                "finish_reason": choice.get("finish_reason"),
                "budget_forced": False,
            }

        used_tokens = first.get("usage", {}).get(
            "completion_tokens",
            max_thinking_budget,
        )
        remaining_tokens = max_tokens - used_tokens
        if remaining_tokens <= 0:
            raise ValueError("Increase max_tokens or lower the budget")

        continuation_messages = copy.deepcopy(original_messages)
        continuation_messages.append(
            {
                "role": "assistant",
                "reasoning_content": reasoning,
                "content": partial_answer,
            }
        )

        second_payload = {
            "model": model,
            "skip_special_tokens": False,
            "messages": continuation_messages,
            "max_tokens": remaining_tokens,
            "add_generation_prompt": False,
            "continue_final_message": True,
            "chat_template_kwargs": {
                "enable_thinking": True,
                "preserve_thinking": True,
            },
            **sampling,
        }
        if tools is not None:
            second_payload["tools"] = tools
            second_payload["tool_choice"] = "auto"

        second = self._post(second_payload)
        second_choice = second["choices"][0]
        second_message = second_choice["message"]
        return {
            "reasoning_content": reasoning,
            "content": partial_answer + (
                second_message.get("content") or ""
            ),
            "tool_calls": second_message.get("tool_calls") or [],
            "finish_reason": second_choice.get("finish_reason"),
            "budget_forced": True,
        }

client = GemmaNemotronThinkingBudgetClient(
    base_url="http://localhost:8000/v1", api_key="EMPTY"
)
MODEL = "nemotron-bolt"

result = client.chat_completion(
    model=MODEL,
    messages=[
        {"role": "system", "content": ""},
        {"role": "user", "content": "What is 2+2?"},
    ],
    max_thinking_budget=32,
    max_tokens=512,
    temperature=1.0,
    top_p=0.95,
    seed=17,
)

print(result["content"])
# Tool-Calling Example

calculate_tip = {
    "type": "function",
    "function": {
        "name": "calculate_tip",
        "description": "Calculate the tip for a bill.",
        "parameters": {
            "type": "object",
            "properties": {
                "bill_total": {"type": "number"},
                "tip_percentage": {"type": "number"},
            },
            "required": ["bill_total", "tip_percentage"],
            "additionalProperties": False,
        },
    },
}

tool_result = client.chat_completion(
    model=MODEL,
    messages=[
        {"role": "system", "content": ""},
        {
            "role": "user",
            "content": (
                "My bill is $100. Use calculate_tip to calculate an 18% tip."
            ),
        },
    ],
    tools=[calculate_tip],
    max_thinking_budget=32,
    max_tokens=512,
    temperature=1.0,
    top_p=0.95,
    seed=23,
)

if not tool_result["tool_calls"]:
    raise RuntimeError("The model did not return a parsed tool call.")

tool_call = tool_result["tool_calls"][0]
print(tool_call["function"]["name"])
print(tool_call["function"]["arguments"])

# Example output:
# calculate_tip
# {"bill_total":100,"tip_percentage":18}

Asynchronous Tool Calling

• Transmit asynchronous functions via the standard OpenAI tools structure.
• Retrieve assistant function executions directly from message.tool_calls.
• Restrict generation to a single tool invocation per assistant turn.
• Assign a persistent task identifier to each decoupled job execution.
• Provide subsequent execution outputs as a user role turn prefixed with something like "TOOL RESULT:".
• Avoid submitting secondary tool turns for requests with an active pending response.
• Maintain tracked status sets for unresolved and finalized job identifiers.
• Synthesize the final response only after all background processes finish and zero calls remain outstanding.

Example:

Assistant tool-call

{
  "role": "assistant",
  "tool_calls": [{
    "id": "call_1",
    "type": "function",
    "function": {
      "name": "build_report",
      "arguments": "{\"_task_id\":\"task_1\",\"scope\":\"weekly\"}"
    }
  }]
}

Immediate acknowledgement with pending result

{
  "role": "tool",
  "tool_call_id": "call_1",
  "content": "{\"status\":\"pending\",\"task_id\":\"task_1\",\"function\":\"build_report\"}"
}

Later tool result completion result

{
  "role": "user",
  "content": "ASYNC TOOL UPDATE: {\"status\":\"completed\",\"task_id\":\"task_1\",\"function\":\"build_report\",\"response\":\"Report is ready.\"}"
}

The model processes the finalized payload to resume generation.

Prompting and Usage Guidelines

General guidelines for using Gemma-Nemotron in conversational and agentic applications.

Prompt Structure and Content

• Store persistent behavior constraints and operational guidelines in system prompts.
• Use explicit headers to organize roles, tasks, tool configs, and formatting safety rules.
• Keep the primary objective clear and concise.
• Write unambiguous, verifiable instructions and resolve conflicting guidance.
• Remove redundant guidance and repetitive few-shot examples.
• Use explicit constraints (e.g., "limit responses to 1-2 sentences") instead of vague requests (e.g., "be brief").
• Avoid synthetic or non-model outputs in "assistant" turns.
• Adhere strictly to standard system/user/assistant/tool turn sequences.

Conveying Environment State in Stateful Applications

• Prefix context blocks with explicit structural markers (e.g., "# CURRENT STATE").
• Clearly specify what part of the context belongs to the model and what part refers to information for other entities like the user.
• When state updates end a message, restate the core goal alongside them.

Instructing the Model

• Define required behaviors with precise language and mandate necessary tool calls.
• Provide fallback protocols for requests exceeding available tool capabilities.
• Separate output formatting rules from task directives, and use the official chat template.

Conversational Responses

• Set conversational_mode=True for concise, low-formatting outputs suited for audio or spoken interaction.
• Set or leave the flag to False to retain standard generation style.

Training, Testing, and Evaluation Datasets

Training

Dataset Overview

Data Modality: Text

Training Data Size: More than 20 Trillion Tokens

Dataset partition: Training [100%], testing [0%], validation [0%]

Time period for training data collection: 2013 to December 2025

Time period for testing data collection: 2013 to December 2025

Time period for validation data collection: 2013 to December 2025

Data Collection Method by dataset: Hybrid: Automated, Manually-Collected, Synthetic

Labeling Method by dataset: Hybrid: Automated, Manually-Labeled, Synthetic

The post-training corpus for NVIDIA-Nemotron-3.5-Bolt-E4B-4GB-BF16 consists of high-quality curated and synthetically-generated data. Primary languages used for post-training include English, French, German, Italian, Japanese, Spanish, and Chinese.

These datasets, such as FinePDFs, EssentialWeb, HotpotQA, SQuAD, and HelpSteer3, do not collectively or exhaustively represent all demographic groups (and proportionally therein). For instance, these datasets do not contain explicit mentions of demographic classes such as age, gender, or ethnicity in 64-99% of samples, depending on the source. In the subset where such terms are present, document-based datasets (FinePDFs and EssentialWeb) contain representational skews, such as references to "male" outnumbering those to "female", and mentions of "White" as the most frequent among ethnic identifiers (comprising 43-44% of ethnicity mentions). To mitigate these imbalances, we recommend considering evaluation techniques such as bias audits, fine-tuning with demographically balanced datasets, and mitigation strategies like counterfactual data augmentation to align with the desired model behavior. This evaluation used a 3,000-sample subset per dataset, identified as the optimal threshold for maximizing embedder accuracy.

During post-training, we generate synthetic data by distilling trajectories, solutions, and translations from strong teacher models and agent systems, often grounded in real tasks or documents and aggressively filtered for quality. For math, code, and science, we start from curated problem sets and use open source permissive models such as GPT-OSS-120B to produce step-by-step reasoning traces, candidate solutions, best-of-n selection traces, and verified CUDA kernels. For long-context and science, we build synthetic QA and reasoning data by retrieving passages from long documents, generating MCQ/OpenQA questions and answers, and paraphrasing them into multiple prompt/response formats to ensure diversity. Across all pipelines we stack automated verification—compilers, numerical checks, language identification—to ensure our data is high quality.

For all domains, we apply a unified data filtering pipeline to ensure that only high-quality, license-compliant, and verifiable samples are used for post-training. We first discard malformed examples using structural checks (e.g., missing tool definitions when tool calls are present). We then aggressively filter reasoning traces exhibiting pathological repetition, such as repeated n-grams within a sliding window or across the entire trajectory, which we found to be a strong indicator of malformed or low-quality reasoning. Finally, based on internal audits of synthetically generated datasets, we observed that some teacher models occasionally produce reasoning traces and final responses that implicitly align with specific political entities or promote nationalistic narratives. To mitigate this, we apply targeted keyword- and regex-based filters and remove all trajectories matching such behavior.

Alongside the model, we release our final pre-training and post-training data, as outlined in this section. For ease of analysis, there is a sample set that is ungated. For all remaining code, math and multilingual data, gating and approval is required, and the dataset is permissively licensed for model training purposes.

Testing Datasets

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic

Labeling Method by dataset

• Hybrid: Automated, Manually-Labeled, Synthetic

Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Evaluation Datasets

Data Collection Method by dataset

• Hybrid: Automated, Manually-Collected, Synthetic

Labeling Method by dataset

• Hybrid: Automated, Manually-Labeled, Synthetic

Properties: This corpus comprises a mix of high-quality standard benchmarks and test suites for modern agentic AI. These benchmarks test model capabilities on tasks such as tool-calling and instruction following.

Inference

• Acceleration Engine: PyTorch
• Test Hardware:
  • NVIDIA Hopper
    • 1-8x H100
    • 1-8x H200
  • NVIDIA Blackwell
    • GB200
    • GeForce RTX 5090

Ethical Considerations

NVIDIA believes Trustworthy AI is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. Developers should work with their internal model team to ensure this model meets requirements for the relevant industry and use case and addresses unforeseen product misuse.

We advise against circumvention of any provided safety guardrails contained in the Model without a substantially similar guardrail appropriate for your use case. For more details: Safety and Explainability Subcards.

For more detailed information on ethical considerations for this model, please see the Model Card++ Bias, Explainability, Safety & Security, and Privacy Subcards.

Please report model quality, risk, security vulnerabilities or NVIDIA AI Concerns here.

Get Help

NVIDIA Developer Community Forum

Get access to community knowledge base articles and support cases (https://forums.developer.nvidia.com/)

Enterprise Support

Get access to knowledge base articles and support cases or submit a ticket.