60:I[510592,["/_next/static/chunks/1zn7-pkefn5mp.js","/_next/static/chunks/2416xwpy5rxui.js","/_next/static/chunks/12dsosltjyg1u.js","/_next/static/chunks/1pko35n022sc4.js","/_next/static/chunks/1la1pzs9j8idc.js","/_next/static/chunks/0a3-qckivp0du.js","/_next/static/chunks/18kgcphq1ij4p.js","/_next/static/chunks/3r5yd4k4rms8-.js","/_next/static/chunks/3xjunri0_zphs.js","/_next/static/chunks/3_35p5sequswa.js","/_next/static/chunks/1m-9t_u1hp8rn.js","/_next/static/chunks/16velfb8dkiep.js","/_next/static/chunks/297op6qeay682.js","/_next/static/chunks/0w0yyo8vyhsf6.js","/_next/static/chunks/0bvim9_g0hu68.js","/_next/static/chunks/3w0ugpudrtlcg.js","/_next/static/chunks/20pz4v29zsnjc.js","/_next/static/chunks/19lkn0qr55tn4.js","/_next/static/chunks/3u8aw78q_egjz.js","/_next/static/chunks/2c-_npfgeu3fm.js","/_next/static/chunks/1501jye_fy898.js","/_next/static/chunks/3qiw0xdgp5feu.js","/_next/static/chunks/2rog_m594n8nz.js","/_next/static/chunks/029ws04y34uan.js","/_next/static/chunks/0czpc-f8wjri8.js","/_next/static/chunks/3cn9umrvnflw9.js","/_next/static/chunks/1ed1ts1_2o5t4.js","/_next/static/chunks/1o2tx5hwmx3ny.js","/_next/static/chunks/3ep4akbi_k_oe.js","/_next/static/chunks/24_gg9g_4g22i.js","/_next/static/chunks/26-5cr0izarn5.js","/_next/static/chunks/1md1s092f0wra.js","/_next/static/chunks/3-u20_a3nl5lp.js","/_next/static/chunks/20d9b23nl375u.js","/_next/static/chunks/345i_dhsjw0l-.js","/_next/static/chunks/2pqhypz2vo8mu.js","/_next/static/chunks/0cno27aove9yt.js","/_next/static/chunks/3hz5y9ooldh1p.js","/_next/static/chunks/0-4d8cut1k8ad.js"],"PlaygroundContextProvider"] 61:T5674,# Gemma 3n e4b-it Overview ## Description Gemma is a family of lightweight, state-of-the-art open models from Google, built from the same research and technology used to create the Gemini models. Gemma 3n models are designed for efficient execution on low-resource devices. They are capable of multimodal input, handling text, image, video, and audio input, and generating text outputs, with open weights for pre-trained and instruction-tuned variants. These models were trained with data in over 140 spoken languages. This model is ready for commercial/non-commercial use. ## Third-Party Community Consideration This model is not owned or developed by NVIDIA. It has been produced to a third-party's requirements for this application and use-case. See the external card: [Gemma 3n e4b-it Model Card](https://huggingface.co/google/gemma-3n-e4b-it). ### License and Terms of Use:
GOVERNING TERMS: This trial service is governed by the [NVIDIA API Trial Terms of Service](https://assets.ngc.nvidia.com/products/api-catalog/legal/NVIDIA%20API%20Trial%20Terms%20of%20Service.pdf). Use of this model is governed by the [NVIDIA Community Model License](https://www.nvidia.com/en-us/agreements/enterprise-software/nvidia-community-models-license/). Additional Information: [Gemma Terms of Use](https://ai.google.dev/gemma/terms) ### Deployment Geography:
Global ### Use Case:
Content Creation and Communication (Text Generation, Chatbots, Summarization, Image/Audio Data Extraction), Research and Education (NLP Research, Language Learning, Knowledge Exploration) ### Intended Usage Open generative models have a wide range of applications across various industries and domains. The following list of potential uses is not comprehensive. The purpose of this list is to provide contextual information about the possible use-cases that the model creators considered as part of model training and development. - Content Creation and Communication - **Text Generation**: Generate creative text formats such as poems, scripts, code, marketing copy, and email drafts. - **Chatbots and Conversational AI**: Power conversational interfaces for customer service, virtual assistants, or interactive applications. - **Text Summarization**: Generate concise summaries of a text corpus, research papers, or reports. - **Image Data Extraction**: Extract, interpret, and summarize visual data for text communications. - **Audio Data Extraction**: Transcribe spoken language, translate speech to text in other languages, and analyze sound-based data. - Research and Education - **Natural Language Processing (NLP) and generative model Research**: These models can serve as a foundation for researchers to experiment with generative models and NLP techniques, develop algorithms, and contribute to the advancement of the field. - **Language Learning Tools**: Support interactive language learning experiences, aiding in grammar correction or providing writing practice. - **Knowledge Exploration**: Assist researchers in exploring large bodies of data by generating summaries or answering questions about specific topics. ### Release Date:
**Build.NVIDIA.com:** 06/26/2025 via ([link](https://build.nvidia.com/google/gemma-3n-e4b-it))
**Hugging Face:** 06/26/2025 via ([link](https://huggingface.co/google/gemma-3n-e4b-it))
### References:
- [Gemma 3n model overview](https://ai.google.dev/gemma/docs/gemma-3n) - [Gemma 3n's efficient parameter management technolog](https://ai.google.dev/gemma/docs/gemma-3n#parameters) - [Responsible Generative AI Toolkit](https://ai.google.dev/responsible) - [Gemma on Kaggle](https://www.kaggle.com/models/google/gemma-3n) - [Gemma on HuggingFace](https://huggingface.co/collections/google/gemma-3n-685065323f5984ef315c93f4) - [Gemma on Vertex Model Garden](https://console.cloud.google.com/vertex-ai/publishers/google/model-garden/gemma3n) ## Model Architecture:
* Architecture Type: Matryoshka Transformer
* Network Architecture: Matryoshka Transformer (MatFormer)
* Parameter Count: 4B (base model), ~6.8B (with Per-Layer Embeddings)
* Number of Layers: 35
* Notable Architectural Features: Selective parameter activation technology
* Base Model: google/gemma-3n-e4b
* Additional Notes: The model's full parameter count is higher than its base model size due to Per-Layer Embeddings (PLE). Standard implementations will load all parameters, including PLE, into VRAM.
### Input
* **Input Type(s):** Text, Image, Audio
* **Input Formats:** Text string, Images (normalized to 256x256, 512x512, or 768x768), Audio data (single channel)
* **Input Parameters:** 1D, 2D, 3D
* **Other Properties Related to Input:** Total input context of 32K tokens. Images are encoded to 256 tokens each. Audio data is encoded to 6.25 tokens per second from a single channel.
### Output
* **Output Type(s):** Text
* **Output Formats:** Text
* **Output Parameters:** 1D
* **Other Properties Related to Output:** Total output length up to 32K tokens, subtracting the request input tokens.
Our Al 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
**Supported Hardware Microarchitecture Compatibility:**
#### NVIDIA GPU Micro-architectures Suitable for Serving **Gemma 3n** in Production *(≥ 16 GB VRAM + Tensor-core /mixed-precision support)* | µArch | First Public Release | Example SKUs (≥ 16 GB) | Tensor-core Gen / Precision | Production Suitability | |-------|---------------------|-------------------------|----------------------------|------------------------| | **Blackwell** | 2024 | B100 (192 GB HBM3e) · B200 (192 GB HBM3e) · RTX 5090 (24 GB GDDR7) | 5-th-gen, FP4 / FP8 | Best-in-class throughput & memory for high-QPS clusters | | **Hopper** | 2022 | H100 (80 / 94 GB) · H200 (141 GB) | 4-th-gen, FP8 (Transformer Engine) | Datacenter standard for LLM inference & training | | **Ada Lovelace** | 2022 | RTX 6000 Ada (48 GB) · L40/L40 S (48 GB) | 4-th-gen, FP8 | Cost-effective edge / on-prem deployments with strong media blocks | | **Ampere** | 2020 | A100 (40 / 80 GB) · A30 (24 GB) · RTX 3090 (24 GB) | 3-rd-gen, BF16 / TF32 | Proven, widely available choice for medium-to-large scale serving | | **Turing** | 2018 | Quadro/RTX 6000 (24 GB) · RTX 8000 (48 GB) | 2-nd-gen, FP16 / INT8 | Viable for latency-tolerant or dev/test replicas | | **Volta** | 2017 | Tesla V100 (16 / 32 GB) | 1-st-gen, FP16 | Legacy datacenter GPUs still supported by CUDA 12 drivers | | **Pascal** *(edge case)* | 2016 | Tesla P100 (16 GB) · P40 (24 GB) | *No* Tensor Cores | Only for low-QPS single-replica use; still covered by R570/575 drivers | > **Recommendation:** Start with **Ampere or newer** for production workloads that demand real-time multimodal responses or higher concurrency. Turing/Volta can host smaller replica pools; Pascal is generally **not** advised for new deployments. ## Model Version:
gemma-3n-e4b-it v1.0 ## Training, Testing, and Evaluation Datasets:
### Training Dataset:
* Data Collection Method: Hybrid: Automated, Synthetic, Human
* Labeling Method: Undisclosed
* Properties: A diverse collection of web text in over 140 languages, code, mathematics, images, and audio, totalling approximately 11 trillion tokens. Knowledge cutoff date is June 2024.
#### Data Preprocessing Here are the key data cleaning and filtering methods applied to the training data: - **CSAM Filtering**: Rigorous CSAM (Child Sexual Abuse Material) filtering was applied at multiple stages in the data preparation process to ensure the exclusion of harmful and illegal content. - **Sensitive Data Filtering**: As part of making Gemma pre-trained models safe and reliable, automated techniques were used to filter out certain personal information and other sensitive data from training sets. - **Additional methods**: Filtering based on content quality and safety in line with [our policies](https://ai.google/static/documents/ai-responsibility-update-published-february-2025.pdf). ### Testing Dataset:
* Data Collection Method by dataset: Hybrid: Automated, Synthetic, Human
* Labeling Method by dataset: Undisclosed
* Properties: Undisclosed
### Evaluation Benchmark Results:
* Data Collection Method: Undisclosed
* Labeling Method: Undisclosed
* Properties: Benchmarks include: HellaSwag, BoolQ, PIQA, SocialIQA, TriviaQA, Natural Questions, ARC-c, ARC-e, WinoGrande, BIG-Bench Hard, DROP, MGSM, WMT24++, Include, MMLU, GPQA Diamond, LiveCodeBench, Codegolf, AIME 2025, MBPP, HumanEval, HiddenMath, Global-MMLU-Lite.
Model evaluation metrics and results. #### Benchmark Results These models were evaluated at full precision (float32) against a large collection of different datasets and metrics to cover different aspects of content generation. Evaluation results marked with **IT** are for instruction-tuned models. Evaluation results marked with **PT** are for pre-trained models. ##### Reasoning and factuality | Benchmark | Metric | n-shot | E2B PT | E4B PT | | ------------------------------ |----------------|----------|:--------:|:--------:| | [HellaSwag][hellaswag] | Accuracy | 10-shot | 72.2 | 78.6 | | [BoolQ][boolq] | Accuracy | 0-shot | 76.4 | 81.6 | | [PIQA][piqa] | Accuracy | 0-shot | 78.9 | 81.0 | | [SocialIQA][socialiqa] | Accuracy | 0-shot | 48.8 | 50.0 | | [TriviaQA][triviaqa] | Accuracy | 5-shot | 60.8 | 70.2 | | [Natural Questions][naturalq] | Accuracy | 5-shot | 15.5 | 20.9 | | [ARC-c][arc] | Accuracy | 25-shot | 51.7 | 61.6 | | [ARC-e][arc] | Accuracy | 0-shot | 75.8 | 81.6 | | [WinoGrande][winogrande] | Accuracy | 5-shot | 66.8 | 71.7 | | [BIG-Bench Hard][bbh] | Accuracy | few-shot | 44.3 | 52.9 | | [DROP][drop] | Token F1 score | 1-shot | 53.9 | 60.8 | [hellaswag]: https://arxiv.org/abs/1905.07830 [boolq]: https://arxiv.org/abs/1905.10044 [piqa]: https://arxiv.org/abs/1911.11641 [socialiqa]: https://arxiv.org/abs/1904.09728 [triviaqa]: https://arxiv.org/abs/1705.03551 [naturalq]: https://github.com/google-research-datasets/natural-questions [arc]: https://arxiv.org/abs/1911.01547 [winogrande]: https://arxiv.org/abs/1907.10641 [bbh]: https://paperswithcode.com/dataset/bbh [drop]: https://arxiv.org/abs/1903.00161 #### Multilingual | Benchmark | Metric | n-shot | E2B IT | E4B IT | | ------------------------------------|-------------------------|----------|:--------:|:--------:| | [MGSM][mgsm] | Accuracy | 0-shot | 53.1 | 60.7 | | [WMT24++][wmt24pp] (ChrF) | Character-level F-score | 0-shot | 42.7 | 50.1 | | [Include][include] | Accuracy | 0-shot | 38.6 | 57.2 | | [MMLU][mmlu] (ProX) | Accuracy | 0-shot | 8.1 | 19.9 | | [OpenAI MMLU][openai-mmlu] | Accuracy | 0-shot | 22.3 | 35.6 | | [Global-MMLU][global-mmlu] | Accuracy | 0-shot | 55.1 | 60.3 | | [ECLeKTic][eclektic] | ECLeKTic score | 0-shot | 2.5 | 1.9 | [mgsm]: https://arxiv.org/abs/2210.03057 [wmt24pp]: https://arxiv.org/abs/2502.12404v1 [include]:https://arxiv.org/abs/2411.19799 [mmlu]: https://arxiv.org/abs/2009.03300 [openai-mmlu]: https://huggingface.co/datasets/openai/MMMLU [global-mmlu]: https://huggingface.co/datasets/CohereLabs/Global-MMLU [eclektic]: https://arxiv.org/abs/2502.21228 #### STEM and code | Benchmark | Metric | n-shot | E2B IT | E4B IT | | ------------------------------------|--------------------------|----------|:--------:|:--------:| | [GPQA][gpqa] Diamond | RelaxedAccuracy/accuracy | 0-shot | 24.8 | 23.7 | | [LiveCodeBench][lcb] v5 | pass@1 | 0-shot | 18.6 | 25.7 | | Codegolf v2.2 | pass@1 | 0-shot | 11.0 | 16.8 | | [AIME 2025][aime-2025] | Accuracy | 0-shot | 6.7 | 11.6 | [gpqa]: https://arxiv.org/abs/2311.12022 [lcb]: https://arxiv.org/abs/2403.07974 [aime-2025]: https://huggingface.co/datasets/opencompass/AIME2025 #### Additional benchmarks | Benchmark | Metric | n-shot | E2B IT | E4B IT | | ------------------------------------ |------------|----------|:--------:|:--------:| | [MMLU][mmlu] | Accuracy | 0-shot | 60.1 | 64.9 | | [MBPP][mbpp] | pass@1 | 3-shot | 56.6 | 63.6 | | [HumanEval][humaneval] | pass@1 | 0-shot | 66.5 | 75.0 | | [LiveCodeBench][lcb] | pass@1 | 0-shot | 13.2 | 13.2 | | HiddenMath | Accuracy | 0-shot | 27.7 | 37.7 | | [Global-MMLU-Lite][global-mmlu-lite] | Accuracy | 0-shot | 59.0 | 64.5 | | [MMLU][mmlu] (Pro) | Accuracy | 0-shot | 40.5 | 50.6 | [gpqa]: https://arxiv.org/abs/2311.12022 [mbpp]: https://arxiv.org/abs/2108.07732 [humaneval]: https://arxiv.org/abs/2107.03374 [lcb]: https://arxiv.org/abs/2403.07974 [global-mmlu-lite]: https://huggingface.co/datasets/CohereForAI/Global-MMLU-Lite ### Evaluation Approach Our evaluation methods include structured evaluations and internal red-teaming testing of relevant content policies. Red-teaming was conducted by a number of different teams, each with different goals and human evaluation metrics. These models were evaluated against a number of different categories relevant to ethics and safety, including: - **Child Safety**: Evaluation of text-to-text and image to text prompts covering child safety policies, including child sexual abuse and exploitation. - **Content Safety:** Evaluation of text-to-text and image to text prompts covering safety policies including, harassment, violence and gore, and hate speech. - **Representational Harms**: Evaluation of text-to-text and image to text prompts covering safety policies including bias, stereotyping, and harmful associations or inaccuracies. In addition to development level evaluations, we conduct "assurance evaluations" which are our 'arms-length' internal evaluations for responsibility governance decision making. They are conducted separately from the model development team, to inform decision making about release. High level findings are fed back to the model team, but prompt sets are held-out to prevent overfitting and preserve the results' ability to inform decision making. Notable assurance evaluation results are reported to our Responsibility & Safety Council as part of release review. ### Evaluation Results For all areas of safety testing, we saw safe levels of performance across the categories of child safety, content safety, and representational harms relative to previous Gemma models. All testing was conducted without safety filters to evaluate the model capabilities and behaviors. For text-to-text, image-to-text, and audio-to-text, and across all model sizes, the model produced minimal policy violations, and showed significant improvements over previous Gemma models' performance with respect to high severity violations. A limitation of our evaluations was they included primarily English language prompts. ## Inference:
**Acceleration Engine:** vLLM
**Test Hardware:** L40s
### Usage Below, there are some code snippets on how to get quickly started with running the model. First, install the Transformers library. Gemma 3n is supported starting from transformers 4.53.0. ```sh $ pip install -U transformers ``` Then, copy the snippet from the section that is relevant for your use case. #### Running with the `pipeline` API You can initialize the model and processor for inference with `pipeline` as follows. ```python from transformers import pipeline import torch pipe = pipeline( "image-text-to-text", model="google/gemma-3n-e4b-it", device="cuda", torch_dtype=torch.bfloat16, ) output = pipe( "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg", text=" in this image, there is" ) print(output) # [{'input_text': ' in this image, there is', # 'generated_text': ' in this image, there is a beautiful flower and a bee is sucking nectar and pollen from the flower.'}] ``` #### Running the model on a single GPU ```python from transformers import AutoProcessor, Gemma3nForConditionalGeneration from PIL import Image import requests import torch model_id = "google/gemma-3n-e4b-it" model = Gemma3nForConditionalGeneration.from_pretrained(model_id, device="cuda", torch_dtype=torch.bfloat16,).eval() processor = AutoProcessor.from_pretrained(model_id) url = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg" image = Image.open(requests.get(url, stream=True).raw) prompt = " in this image, there is" model_inputs = processor(text=prompt, images=image, return_tensors="pt").to(model.device) input_len = model_inputs["input_ids"].shape[-1] with torch.inference_mode(): generation = model.generate(**model_inputs, max_new_tokens=10) generation = generation[0][input_len:] decoded = processor.decode(generation, skip_special_tokens=True) print(decoded) # one picture of flowers which shows that the flower is ``` ## Additional Details:
### Usage and Limitations These models have certain limitations that users should be aware of. #### Limitations - Training Data - The quality and diversity of the training data significantly influence the model's capabilities. Biases or gaps in the training data can lead to limitations in the model's responses. - The scope of the training dataset determines the subject areas the model can handle effectively. - Context and Task Complexity - Models are better at tasks that can be framed with clear prompts and instructions. Open-ended or highly complex tasks might be challenging. - A model's performance can be influenced by the amount of context provided (longer context generally leads to better outputs, up to a certain point). - Language Ambiguity and Nuance - Natural language is inherently complex. Models might struggle to grasp subtle nuances, sarcasm, or figurative language. - Factual Accuracy - Models generate responses based on information they learned from their training datasets, but they are not knowledge bases. They may generate incorrect or outdated factual statements. - Common Sense - Models rely on statistical patterns in language. They might lack the ability to apply common sense reasoning in certain situations. Risks identified and mitigations: - **Perpetuation of biases**: It's encouraged to perform continuous monitoring (using evaluation metrics, human review) and the exploration of de-biasing techniques during model training, fine-tuning, and other use cases. - **Generation of harmful content**: Mechanisms and guidelines for content safety are essential. Developers are encouraged to exercise caution and implement appropriate content safety safeguards based on their specific product policies and application use cases. - **Misuse for malicious purposes**: Technical limitations and developer and end-user education can help mitigate against malicious applications of generative models. Educational resources and reporting mechanisms for users to flag misuse are provided. Prohibited uses of Gemma models are outlined in the [Gemma Prohibited Use Policy](https://ai.google.dev/gemma/prohibited_use_policy). - **Privacy violations**: Models were trained on data filtered for removal of certain personal information and other sensitive data. Developers are encouraged to adhere to privacy regulations with privacy-preserving techniques. #### Benefits At the time of release, this family of models provides high-performance open generative model implementations designed from the ground up for responsible AI development compared to similarly sized models. Using the benchmark evaluation metrics described in this document, these models have shown to provide superior performance to other, comparably-sized open model alternatives. ## Ethical Considerations:
NVIDIA believes Trustworthy Al is a shared responsibility and we have established policies and practices to enable development for a wide array of AI applications. When downloaded or used in accordance with our terms of service, 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. Please report security vulnerabilities or NVIDIA AI Concerns [here](https://www.nvidia.com/en-us/support/submit-security-vulnerability/). ## Citation ``` @article{gemma_3n_2025, title={Gemma 3n}, url={https://ai.google.dev/gemma/docs/gemma-3n}, publisher={Google DeepMind}, author={Gemma Team}, year={2025} } ```62:T1daa,<% // Catalog ejs locals replace JSON with {stringify} only (spread skips non-enumerable parse). const deepClone = (value) => { if (Array.isArray(value)) return value.map(deepClone); if (value && typeof value === "object") { const out = {}; for (const key of Object.keys(value)) out[key] = deepClone(value[key]); return out; } return value; }; const mediaConfig = {"image": {"mimeTypes": {".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".webp": "image/webp"}}, "video": {"mimeTypes": {}}, "audio": {"mimeTypes": {".wav": "audio/wav", ".mp3": "audio/mpeg"}, "defaultFormat": "wav"}}; const stripNulls = (value) => { if (Array.isArray(value)) return value.map(stripNulls); if (value && typeof value === "object") { const out = {}; for (const [key, item] of Object.entries(value)) { if (item === null || item === undefined) continue; out[key] = stripNulls(item); } return out; } return value; }; const payloadForSnippet = stripNulls(deepClone(request || {})); const hasStream = payloadForSnippet.stream !== undefined && payloadForSnippet.stream !== null; const pyStream = payloadForSnippet.stream ? "True" : "False"; if (hasStream) payloadForSnippet.stream = "__STREAM__"; const hasMediaKind = (kind) => Object.keys(mediaConfig[kind].mimeTypes || {}).length > 0; const isInlineDataUrl = (url, kind) => ( typeof url === "string" && new RegExp(`^data:${kind}/[^;]+;base64,`, "i").test(url) ); const mediaUrl = (part, key) => { const slot = part?.[key]; return typeof slot === "string" ? slot : slot?.url; }; const setMediaUrl = (part, key, url) => { if (typeof part[key] === "string") part[key] = url; else { if (!part[key] || typeof part[key] !== "object") part[key] = {}; part[key].url = url; } }; let imageCount = 0; let videoCount = 0; let audioCount = 0; (payloadForSnippet.messages || []).forEach((message) => { if (!Array.isArray(message?.content)) return; message.content.forEach((part) => { const imageUrl = mediaUrl(part, "image_url"); const videoUrl = mediaUrl(part, "video_url"); const audioUrl = mediaUrl(part, "audio_url"); // File helpers only for inline base64; http(s) URLs stay as-is in the payload. if (hasMediaKind("image") && part?.type === "image_url" && isInlineDataUrl(imageUrl, "image")) { setMediaUrl(part, "image_url", `__IMAGE_DATA_URL_${imageCount}__`); imageCount += 1; } else if (hasMediaKind("video") && part?.type === "video_url" && isInlineDataUrl(videoUrl, "video")) { setMediaUrl(part, "video_url", `__VIDEO_DATA_URL_${videoCount}__`); videoCount += 1; } else if (hasMediaKind("audio") && part?.type === "audio_url" && isInlineDataUrl(audioUrl, "audio")) { setMediaUrl(part, "audio_url", `__AUDIO_DATA_URL_${audioCount}__`); audioCount += 1; } else if (hasMediaKind("audio") && part?.type === "input_audio" && part.input_audio?.data) { part.input_audio.data = `__AUDIO_B64_${audioCount}__`; part.input_audio.format = part.input_audio.format || mediaConfig.audio.defaultFormat; audioCount += 1; } }); }); const pyRepr = (value, level = 0) => { if (value === null) return "None"; if (typeof value === "boolean") return value ? "True" : "False"; if (typeof value === "number") return String(value); if (typeof value === "string") return JSON.stringify(value); const pad = " ".repeat(level); const innerPad = " ".repeat(level + 2); if (Array.isArray(value)) { if (value.length === 0) return "[]"; return "[\n" + value.map((item) => `${innerPad}${pyRepr(item, level + 2)}`).join(",\n") + "\n" + pad + "]"; } if (typeof value === "object") { const entries = Object.entries(value); if (entries.length === 0) return "{}"; return "{\n" + entries.map(([key, item]) => `${innerPad}${JSON.stringify(key)}: ${pyRepr(item, level + 2)}`).join(",\n") + "\n" + pad + "}"; } return "None"; }; const replaceAll = (s, a, b) => s.split(a).join(b); let renderedPayload = pyRepr(payloadForSnippet); if (hasStream) renderedPayload = replaceAll(renderedPayload, '"__STREAM__"', "stream"); for (let i = 0; i < imageCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__IMAGE_DATA_URL_${i}__"`, `image_data_urls[${i}]`); } for (let i = 0; i < videoCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__VIDEO_DATA_URL_${i}__"`, `video_data_urls[${i}]`); } for (let i = 0; i < audioCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__AUDIO_DATA_URL_${i}__"`, `audio_data_urls[${i}]`); renderedPayload = replaceAll(renderedPayload, `"__AUDIO_B64_${i}__"`, `audio_b64s[${i}]`); } // Either file helpers XOR inline base64 - never emit unused helpers. const usesFileHelpers = imageCount + videoCount + audioCount > 0 && [...Array(imageCount).keys()].every((i) => renderedPayload.includes(`image_data_urls[${i}]`)) && [...Array(videoCount).keys()].every((i) => renderedPayload.includes(`video_data_urls[${i}]`)) && [...Array(audioCount).keys()].every( (i) => renderedPayload.includes(`audio_data_urls[${i}]`) || renderedPayload.includes(`audio_b64s[${i}]`) ); if (!usesFileHelpers) { imageCount = 0; videoCount = 0; audioCount = 0; const restored = stripNulls(deepClone(request || {})); const restoredStream = restored.stream !== undefined && restored.stream !== null; if (restoredStream) restored.stream = "__STREAM__"; renderedPayload = pyRepr(restored); if (restoredStream) renderedPayload = replaceAll(renderedPayload, '"__STREAM__"', "stream"); } const hasMedia = usesFileHelpers; const mimePy = (mimes) => "{" + Object.entries(mimes).map(([k, v]) => `${JSON.stringify(k)}: ${JSON.stringify(v)}`).join(", ") + "}"; %> <% if (hasMedia) { %>import base64 from pathlib import Path <% } %>import requests invoke_url = "https://integrate.api.nvidia.com/v1/chat/completions" stream = <%- pyStream %> <% if (hasMedia) { %> def media_data_url(kind, index, mime_types): for suffix, mime in mime_types.items(): path = Path(f"{kind}_{index}{suffix}") if path.exists(): return f"data:{mime};base64,{base64.b64encode(path.read_bytes()).decode()}" raise FileNotFoundError(f"Expected {kind}_{index} with suffix in {list(mime_types)}") <% if (imageCount > 0) { %> image_data_urls = [media_data_url("image", i, <%- mimePy(mediaConfig.image.mimeTypes) %>) for i in <%- imageCount === 1 ? "[1]" : ("range(1, " + (imageCount + 1) + ")") %>] <% } %><% if (videoCount > 0) { %> video_data_urls = [media_data_url("video", i, <%- mimePy(mediaConfig.video.mimeTypes) %>) for i in <%- videoCount === 1 ? "[1]" : ("range(1, " + (videoCount + 1) + ")") %>] <% } %><% if (audioCount > 0) { %> audio_data_urls = [media_data_url("audio", i, <%- mimePy(mediaConfig.audio.mimeTypes) %>) for i in <%- audioCount === 1 ? "[1]" : ("range(1, " + (audioCount + 1) + ")") %>] audio_b64s = [url.split(",", 1)[1] for url in audio_data_urls] <% } %><% } %> headers = { "Authorization": "Bearer $NVIDIA_API_KEY", "Accept": "text/event-stream" if stream else "application/json", } payload = <%- renderedPayload %> response = requests.post(invoke_url, headers=headers, json=payload, stream=stream) if stream: for line in response.iter_lines(): if line: print(line.decode("utf-8")) else: print(response.json()) 63:T16c7,<% const deepClone = (value) => { if (Array.isArray(value)) return value.map(deepClone); if (value && typeof value === "object") { const out = {}; for (const key of Object.keys(value)) out[key] = deepClone(value[key]); return out; } return value; }; const mediaConfig = {"image": {"mimeTypes": {".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".webp": "image/webp"}}, "video": {"mimeTypes": {}}, "audio": {"mimeTypes": {".wav": "audio/wav", ".mp3": "audio/mpeg"}, "defaultFormat": "wav"}}; const payloadForSnippet = deepClone(request || {}); const messagesForSnippet = deepClone(payloadForSnippet.messages || []); const hasMediaKind = (kind) => Object.keys(mediaConfig[kind].mimeTypes || {}).length > 0; const isInlineDataUrl = (url, kind) => ( typeof url === "string" && new RegExp(`^data:${kind}/[^;]+;base64,`, "i").test(url) ); const mediaUrl = (part, key) => { const slot = part?.[key]; return typeof slot === "string" ? slot : slot?.url; }; const setMediaUrl = (part, key, url) => { if (typeof part[key] === "string") part[key] = url; else { if (!part[key] || typeof part[key] !== "object") part[key] = {}; part[key].url = url; } }; let imageCount = 0; let hasUnsupportedMedia = false; messagesForSnippet.forEach((message) => { if (!Array.isArray(message?.content)) return; message.content.forEach((part) => { const imageUrl = mediaUrl(part, "image_url"); if (hasMediaKind("image") && part?.type === "image_url" && isInlineDataUrl(imageUrl, "image")) { setMediaUrl(part, "image_url", `__IMAGE_DATA_URL_${imageCount}__`); imageCount += 1; } else if (part?.type === "video_url" && isInlineDataUrl(mediaUrl(part, "video_url"), "video")) { // LangChain cannot ship local file helpers for video; reject inline only. hasUnsupportedMedia = true; setMediaUrl(part, "video_url", "__UNSUPPORTED_MEDIA_DATA__"); } else if (part?.type === "audio_url" && isInlineDataUrl(mediaUrl(part, "audio_url"), "audio")) { hasUnsupportedMedia = true; setMediaUrl(part, "audio_url", "__UNSUPPORTED_MEDIA_DATA__"); } else if (part?.type === "input_audio" && part.input_audio?.data) { hasUnsupportedMedia = true; part.input_audio.data = "__UNSUPPORTED_MEDIA_DATA__"; } }); }); const pyRepr = (value, level = 0) => { if (value === null) return "None"; if (typeof value === "boolean") return value ? "True" : "False"; if (typeof value === "number") return String(value); if (typeof value === "string") return JSON.stringify(value); const pad = " ".repeat(level); const innerPad = " ".repeat(level + 2); if (Array.isArray(value)) { if (value.length === 0) return "[]"; return "[\n" + value.map((item) => `${innerPad}${pyRepr(item, level + 2)},`).join("\n") + "\n" + pad + "]"; } if (typeof value === "object") { const entries = Object.entries(value); if (entries.length === 0) return "{}"; return "{\n" + entries.map(([key, item]) => `${innerPad}${JSON.stringify(key)}: ${pyRepr(item, level + 2)},`).join("\n") + "\n" + pad + "}"; } return "None"; }; const replaceAll = (s, a, b) => s.split(a).join(b); let renderedMessages = pyRepr(messagesForSnippet); for (let i = 0; i < imageCount; i++) { renderedMessages = replaceAll(renderedMessages, `"__IMAGE_DATA_URL_${i}__"`, `image_data_urls[${i}]`); } // Either file helpers XOR inline base64 - never emit unused helpers. const usesFileHelpers = imageCount > 0 && [...Array(imageCount).keys()].every((i) => renderedMessages.includes(`image_data_urls[${i}]`)); if (!usesFileHelpers) { imageCount = 0; renderedMessages = pyRepr(deepClone((request || {}).messages || [])); } %> from langchain_nvidia_ai_endpoints import ChatNVIDIA <% if (imageCount > 0) { %>from pathlib import Path import base64 <% } %> client = ChatNVIDIA( model="<%- request.model %>", api_key="$NVIDIA_API_KEY", temperature=<%- request.temperature %>, top_p=<%- request.top_p %>, max_completion_tokens=<%- request.max_tokens %>, ) <% if (imageCount > 0) { %> def find_image_path(index): image_mime_types = <%- pyRepr(mediaConfig.image.mimeTypes) %> for suffix in image_mime_types: path = Path(f"image_{index}{suffix}") if path.exists(): return path expected = "/".join(f".{suffix.lstrip('.')}" for suffix in image_mime_types) raise FileNotFoundError(f"Expected image_{index}{expected}") def read_image_data_url(path): image_mime_types = <%- pyRepr(mediaConfig.image.mimeTypes) %> with open(path, "rb") as f: image_b64 = base64.b64encode(f.read()).decode() return f"data:{image_mime_types[path.suffix.lower()]};base64,{image_b64}" image_data_urls = [read_image_data_url(find_image_path(i)) for i in range(1, <%- imageCount + 1 %>)] <% } %><% if (hasUnsupportedMedia) { %> raise ValueError("LangChain snippet currently supports text and image input. Use Python, Node.js, or Shell for video or audio input.") <% } %> lc_messages = <%- renderedMessages %> <% if (request.stream) { %> for chunk in client.stream(lc_messages<%= request.chat_template_kwargs ? ', chat_template_kwargs=' + pyRepr(request.chat_template_kwargs) : '' %>): if chunk.additional_kwargs and "reasoning_content" in chunk.additional_kwargs: print(chunk.additional_kwargs["reasoning_content"], end="") print(chunk.content, end="") <% } else { %> response = client.invoke(lc_messages<%= request.chat_template_kwargs ? ', chat_template_kwargs=' + pyRepr(request.chat_template_kwargs) : '' %>) if response.additional_kwargs and "reasoning_content" in response.additional_kwargs: print(response.additional_kwargs["reasoning_content"]) print(response.content) <% } %> 64:T1b06,<% const deepClone = (value) => { if (Array.isArray(value)) return value.map(deepClone); if (value && typeof value === "object") { const out = {}; for (const key of Object.keys(value)) out[key] = deepClone(value[key]); return out; } return value; }; const mediaConfig = {"image": {"mimeTypes": {".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".webp": "image/webp"}}, "video": {"mimeTypes": {}}, "audio": {"mimeTypes": {".wav": "audio/wav", ".mp3": "audio/mpeg"}, "defaultFormat": "wav"}}; const payloadForSnippet = deepClone(request || {}); const hasStream = payloadForSnippet.stream !== undefined && payloadForSnippet.stream !== null; if (hasStream) payloadForSnippet.stream = "__STREAM__"; const hasMediaKind = (kind) => Object.keys(mediaConfig[kind].mimeTypes || {}).length > 0; const isInlineDataUrl = (url, kind) => ( typeof url === "string" && new RegExp(`^data:${kind}/[^;]+;base64,`, "i").test(url) ); let imageCount = 0; let videoCount = 0; let audioCount = 0; (payloadForSnippet.messages || []).forEach((message) => { if (!Array.isArray(message?.content)) return; message.content.forEach((part) => { if (hasMediaKind("image") && part?.type === "image_url" && isInlineDataUrl(part.image_url?.url, "image")) { part.image_url.url = `__IMAGE_DATA_URL_${imageCount}__`; imageCount += 1; } else if (hasMediaKind("video") && part?.type === "video_url" && isInlineDataUrl(part.video_url?.url, "video")) { part.video_url.url = `__VIDEO_DATA_URL_${videoCount}__`; videoCount += 1; } else if (hasMediaKind("audio") && part?.type === "audio_url" && isInlineDataUrl(part.audio_url?.url, "audio")) { part.audio_url.url = `__AUDIO_DATA_URL_${audioCount}__`; audioCount += 1; } else if (hasMediaKind("audio") && part?.type === "input_audio" && part.input_audio?.data) { part.input_audio.data = `__AUDIO_B64_${audioCount}__`; part.input_audio.format = part.input_audio.format || mediaConfig.audio.defaultFormat; audioCount += 1; } }); }); const replaceAll = (s, a, b) => s.split(a).join(b); let renderedPayload = JSON.stringify(payloadForSnippet, null, 2); if (hasStream) renderedPayload = replaceAll(renderedPayload, '"__STREAM__"', "stream"); for (let i = 0; i < imageCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__IMAGE_DATA_URL_${i}__"`, `imageDataUrls[${i}]`); } for (let i = 0; i < videoCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__VIDEO_DATA_URL_${i}__"`, `videoDataUrls[${i}]`); } for (let i = 0; i < audioCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__AUDIO_DATA_URL_${i}__"`, `audioDataUrls[${i}]`); renderedPayload = replaceAll(renderedPayload, `"__AUDIO_B64_${i}__"`, `audioB64s[${i}]`); } // Either file helpers XOR inline base64 - never emit unused helpers. const usesFileHelpers = imageCount + videoCount + audioCount > 0 && [...Array(imageCount).keys()].every((i) => renderedPayload.includes(`imageDataUrls[${i}]`)) && [...Array(videoCount).keys()].every((i) => renderedPayload.includes(`videoDataUrls[${i}]`)) && [...Array(audioCount).keys()].every( (i) => renderedPayload.includes(`audioDataUrls[${i}]`) || renderedPayload.includes(`audioB64s[${i}]`) ); if (!usesFileHelpers) { imageCount = 0; videoCount = 0; audioCount = 0; const restored = deepClone(request || {}); const restoredStream = restored.stream !== undefined && restored.stream !== null; if (restoredStream) restored.stream = "__STREAM__"; renderedPayload = JSON.stringify(restored, null, 2); if (restoredStream) renderedPayload = replaceAll(renderedPayload, '"__STREAM__"', "stream"); } const hasMedia = usesFileHelpers; %> import axios from 'axios'; <% if (hasMedia) { %>import { existsSync } from 'node:fs'; import { readFile } from 'node:fs/promises'; import { extname } from 'node:path'; <% } %> const invokeUrl = "https://integrate.api.nvidia.com/v1/chat/completions"; const stream = <%- request.stream %>; const headers = { "Authorization": "Bearer $NVIDIA_API_KEY", "Accept": stream ? "text/event-stream" : "application/json" }; <% if (hasMedia) { %> function findMediaPath(kind, index, mimeTypes) { for (const suffix of Object.keys(mimeTypes)) { const path = `${kind}_${index}${suffix}`; if (existsSync(path)) { return path; } } const expected = Object.keys(mimeTypes).join("/"); throw new Error(`Expected ${kind}_${index}${expected}`); } async function readMediaB64(path) { return Buffer.from(await readFile(path)).toString('base64'); } function mediaDataUrl(path, mediaB64, mimeTypes) { return `data:${mimeTypes[extname(path).toLowerCase()]};base64,${mediaB64}`; } <% } %> async function main() { <% if (imageCount > 0) { %> const imageMimeTypes = <%- JSON.stringify(mediaConfig.image.mimeTypes, null, 2) %>; const imagePaths = [<% for (let i = 1; i <= imageCount; i++) { %>findMediaPath("image", <%- i %>, imageMimeTypes)<%- i < imageCount ? ", " : "" %><% } %>]; const imageB64s = await Promise.all(imagePaths.map(readMediaB64)); const imageDataUrls = imagePaths.map((path, index) => mediaDataUrl(path, imageB64s[index], imageMimeTypes)); <% } %><% if (videoCount > 0) { %> const videoMimeTypes = <%- JSON.stringify(mediaConfig.video.mimeTypes, null, 2) %>; const videoPaths = [<% for (let i = 1; i <= videoCount; i++) { %>findMediaPath("video", <%- i %>, videoMimeTypes)<%- i < videoCount ? ", " : "" %><% } %>]; const videoB64s = await Promise.all(videoPaths.map(readMediaB64)); const videoDataUrls = videoPaths.map((path, index) => mediaDataUrl(path, videoB64s[index], videoMimeTypes)); <% } %><% if (audioCount > 0) { %> const audioMimeTypes = <%- JSON.stringify(mediaConfig.audio.mimeTypes, null, 2) %>; const audioPaths = [<% for (let i = 1; i <= audioCount; i++) { %>findMediaPath("audio", <%- i %>, audioMimeTypes)<%- i < audioCount ? ", " : "" %><% } %>]; const audioB64s = await Promise.all(audioPaths.map(readMediaB64)); const audioDataUrls = audioPaths.map((path, index) => mediaDataUrl(path, audioB64s[index], audioMimeTypes)); <% } %> const payload = <%- renderedPayload %>; const response = await axios.post(invokeUrl, payload, { headers: headers, responseType: stream ? 'stream' : 'json' }); if (stream) { response.data.on('data', (chunk) => { console.log(chunk.toString()); }); } else { console.log(JSON.stringify(response.data)); } } main().catch(error => { if (error.response) { console.error(`HTTP ${error.response.status}`); if (error.response.data?.on) { error.response.data.on('data', (chunk) => console.error(chunk.toString())); } else { console.error(error.response.data); } } else { console.error(error); } }); 65:T15bd,<% const deepClone = (value) => { if (Array.isArray(value)) return value.map(deepClone); if (value && typeof value === "object") { const out = {}; for (const key of Object.keys(value)) out[key] = deepClone(value[key]); return out; } return value; }; const mediaConfig = {"image": {"mimeTypes": {".png": "image/png", ".jpg": "image/jpeg", ".jpeg": "image/jpeg", ".webp": "image/webp"}}, "video": {"mimeTypes": {}}, "audio": {"mimeTypes": {".wav": "audio/wav", ".mp3": "audio/mpeg"}, "defaultFormat": "wav"}}; const payloadForSnippet = deepClone(request || {}); const hasMediaKind = (kind) => Object.keys(mediaConfig[kind].mimeTypes || {}).length > 0; const isInlineDataUrl = (url, kind) => ( typeof url === "string" && new RegExp(`^data:${kind}/[^;]+;base64,`, "i").test(url) ); let imageCount = 0; let videoCount = 0; let audioCount = 0; (payloadForSnippet.messages || []).forEach((message) => { if (!Array.isArray(message?.content)) return; message.content.forEach((part) => { if (hasMediaKind("image") && part?.type === "image_url" && isInlineDataUrl(part.image_url?.url, "image")) { part.image_url.url = `__IMAGE_DATA_URL_${imageCount}__`; imageCount += 1; } else if (hasMediaKind("video") && part?.type === "video_url" && isInlineDataUrl(part.video_url?.url, "video")) { part.video_url.url = `__VIDEO_DATA_URL_${videoCount}__`; videoCount += 1; } else if (hasMediaKind("audio") && part?.type === "audio_url" && isInlineDataUrl(part.audio_url?.url, "audio")) { part.audio_url.url = `__AUDIO_DATA_URL_${audioCount}__`; audioCount += 1; } else if (hasMediaKind("audio") && part?.type === "input_audio" && part.input_audio?.data) { part.input_audio.data = `__AUDIO_B64_${audioCount}__`; part.input_audio.format = part.input_audio.format || mediaConfig.audio.defaultFormat; audioCount += 1; } }); }); const replaceAll = (s, a, b) => s.split(a).join(b); let renderedPayload = JSON.stringify(payloadForSnippet, null, 2); for (let i = 0; i < imageCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__IMAGE_DATA_URL_${i}__"`, `"${"${"}image_${i + 1}_data_url}"`); } for (let i = 0; i < videoCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__VIDEO_DATA_URL_${i}__"`, `"${"${"}video_${i + 1}_data_url}"`); } for (let i = 0; i < audioCount; i++) { renderedPayload = replaceAll(renderedPayload, `"__AUDIO_DATA_URL_${i}__"`, `"${"${"}audio_${i + 1}_data_url}"`); renderedPayload = replaceAll(renderedPayload, `"__AUDIO_B64_${i}__"`, `"${"${"}audio_${i + 1}_b64}"`); } // Either file helpers XOR inline base64 - never emit unused helpers. const usesFileHelpers = imageCount + videoCount + audioCount > 0 && [...Array(imageCount).keys()].every((i) => renderedPayload.includes(`image_${i + 1}_data_url`)) && [...Array(videoCount).keys()].every((i) => renderedPayload.includes(`video_${i + 1}_data_url`)) && [...Array(audioCount).keys()].every( (i) => renderedPayload.includes(`audio_${i + 1}_data_url`) || renderedPayload.includes(`audio_${i + 1}_b64`) ); if (!usesFileHelpers) { imageCount = 0; videoCount = 0; audioCount = 0; renderedPayload = JSON.stringify(deepClone(request || {}), null, 2); } const hasMedia = usesFileHelpers; %> stream=<%- request.stream %> if [ "$stream" = true ]; then accept_header='Accept: text/event-stream' else accept_header='Accept: application/json' fi <% if (hasMedia) { %> mime_type_for() { case "$1" in <% for (const [suffix, mimeType] of Object.entries({...mediaConfig.image.mimeTypes, ...mediaConfig.video.mimeTypes, ...mediaConfig.audio.mimeTypes})) { %> *<%- suffix %>) echo "<%- mimeType %>" ;; <% } %> *) echo "Unsupported media file: $1" >&2; return 1 ;; esac } find_media_path() { local kind="$1" local index="$2" shift 2 local ext path for ext in "$@"; do path="${kind}_${index}.${ext}" if [ -f "$path" ]; then echo "$path" return 0 fi done echo "Expected ${kind}_${index}.$*" >&2 return 1 } media_b64() { base64 < "$1" | tr -d '\n' } media_data_url() { local path="$1" local mime_type media_base64 mime_type=$(mime_type_for "$path") || return 1 media_base64=$(media_b64 "$path") || return 1 printf 'data:%s;base64,%s' "$mime_type" "$media_base64" } <% for (let i = 1; i <= imageCount; i++) { %> image_<%- i %>_path=$(find_media_path image <%- i %><% for (const suffix of Object.keys(mediaConfig.image.mimeTypes)) { %> <%- suffix.slice(1) %><% } %>) image_<%- i %>_data_url=$(media_data_url "$image_<%- i %>_path") <% } %><% for (let i = 1; i <= videoCount; i++) { %> video_<%- i %>_path=$(find_media_path video <%- i %><% for (const suffix of Object.keys(mediaConfig.video.mimeTypes)) { %> <%- suffix.slice(1) %><% } %>) video_<%- i %>_data_url=$(media_data_url "$video_<%- i %>_path") <% } %><% for (let i = 1; i <= audioCount; i++) { %> audio_<%- i %>_path=$(find_media_path audio <%- i %><% for (const suffix of Object.keys(mediaConfig.audio.mimeTypes)) { %> <%- suffix.slice(1) %><% } %>) audio_<%- i %>_b64=$(media_b64 "$audio_<%- i %>_path") audio_<%- i %>_data_url=$(media_data_url "$audio_<%- i %>_path") <% } %><% } %> cat > payload.json < JSON curl https://integrate.api.nvidia.com/v1/chat/completions \ -H "Authorization: Bearer $NVIDIA_API_KEY" \ -H "Content-Type: application/json" \ -H "$accept_header" \ -d @payload.json58:[["$","$L41",".0",{"state":{"mutations":[],"queries":[{"dehydratedAt":1790110093823,"state":{"data":"$38:props:state:queries:0:state:data","dataUpdateCount":1,"dataUpdatedAt":1790110093778,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":"$38:props:state:queries:0:queryKey","queryHash":"[\"search\",\"labelsets\",\"ENDPOINT\"]"},{"dehydratedAt":1790110093823,"state":{"data":null,"dataUpdateCount":1,"dataUpdatedAt":1790110093778,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":"$38:props:state:queries:1:queryKey","queryHash":"[\"killswitches/riva-magpie-2.enabled\"]"},{"dehydratedAt":1790110093823,"state":{"data":"$38:props:state:queries:2:state:data","dataUpdateCount":1,"dataUpdatedAt":1790110093778,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":"$38:props:state:queries:2:queryKey","queryHash":"[\"killswitches/media/select-example.models.yaml\"]"},{"dehydratedAt":1790110093823,"state":{"data":"$38:props:state:queries:3:state:data","dataUpdateCount":1,"dataUpdatedAt":1790110093778,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":"$38:props:state:queries:3:queryKey","queryHash":"[\"content/legal/model-terms-of-service.yaml\"]"},{"dehydratedAt":1790110093823,"state":{"data":"$38:props:state:queries:4:state:data","dataUpdateCount":1,"dataUpdatedAt":1790110093779,"error":null,"errorUpdateCount":0,"errorUpdatedAt":0,"fetchFailureCount":0,"fetchFailureReason":null,"fetchMeta":null,"isInvalidated":false,"status":"success","fetchStatus":"idle"},"queryKey":"$38:props:state:queries:4:queryKey","queryHash":"[\"content/legal/playground-disclaimers.yaml\"]"}]},"children":["$","$L60",null,{"initialState":{"artifact":{"artifactType":"ENDPOINT","name":"gemma-3n-e4b-it","displayName":"gemma-3n-e4b-it","publisher":"google","shortDescription":"An edge computing AI model which accepts text, audio and image input, ideal for resource-constrained environments","logo":"https://assets.ngc.nvidia.com/products/api-catalog/images/gemma-3n-e4b-it.jpg","labels":["Visual QA","chat","cloudPartnerType:endpoint:cloud_partner_type_openrouter","cloudPartnerType:endpoint:cloud_partner_type_together","language generation","nimType:endpoint:nim_type_preview","nimType:endpoint:nim_type_upgrade_available","playgroundType:endpoint:playgroundtype_chat","speech recognition"],"attributes":[{"key":"AVAILABLE","value":"false"},{"key":"DEPRECATION","value":"07/27/2026"},{"key":"PREVIEW","value":"true"},{"key":"last_month_api_invocation_count","value":"3061627"},{"key":"last_month_download_count","value":"0"}],"updatedDate":"2025-07-15T22:22:15.788Z","canGuestDownload":true,"createdDate":"2025-06-26T16:44:19.774Z","description":"$61","isPublic":true,"isReadOnly":true,"orgName":"qc69jvmznzxy"},"specAttributes":"$59:props:spec:attributes","modelPath":"/google/gemma-3n-e4b-it","openApiSpec":{"openapi":"3.1.0","info":{"title":"NVIDIA NIM API for google/gemma-3n-e4b-it","description":"The NVIDIA NIM REST API. Please see https://docs.api.nvidia.com/nim/reference/gemma-3n-e4b-it for more details.","version":"1.0.0","termsOfService":"https://nvidia.com/legal/terms-of-use","contact":{"name":"NVIDIA Support","url":"https://help.nvidia.com/"},"license":{"name":"Apache 2.0","url":"https://choosealicense.com/licenses/apache-2.0/"}},"servers":[{"url":"https://integrate.api.nvidia.com/v1"}],"tags":[{"name":"Multimodal API","description":"This API performs inference using multi-modal large language models (MLLMs)."}],"paths":{"/chat/completions":{"post":{"tags":["Multimodal API"],"summary":"Request response from the model","description":"Invokes inference using the model chat parameters.","operationId":"invokeFunction","requestBody":{"content":{"application/json":{"schema":{"properties":{"messages":{"items":{"properties":{"role":{"allOf":[{"type":"string","enum":["assistant","user"],"title":"Role"}],"description":"The role of the message's author."},"content":{"anyOf":[{"type":"string"},{"items":{"anyOf":[{"properties":{"audio_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"AudioURL"},"type":{"type":"string","enum":["audio_url"],"const":"audio_url","title":"Type"}},"type":"object","required":["audio_url","type"],"title":"NIMLLMChatCompletionContentPartAudio"},{"properties":{"image_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"ImageURL"},"type":{"type":"string","enum":["image_url"],"const":"image_url","title":"Type"}},"type":"object","required":["image_url","type"],"title":"NIMLLMChatCompletionContentPartImage"},{"properties":{"text":{"type":"string","title":"Text"},"type":{"type":"string","enum":["text"],"const":"text","title":"Type"}},"type":"object","required":["text","type"],"title":"NIMLLMChatCompletionContentPartText"}]},"type":"array"}],"title":"Content","description":"The contents of the message.\n
To pass media (only with role=`user`):\n
- Use content parts. Text is passed with `type`=`text`.\n
- Images are passed with `type`=`image_url`; set `image_url.url` to an image URL or a base64 data URI like `data:image/{format};base64,{base64encodedimage}`.\n
- Audio is passed with `type`=`audio_url` or `type`=`input_audio`; set `audio_url.url` to an audio URL or a base64 data URI, or set `input_audio.data` to base64 audio data.\n
For `system` and `assistant` roles, the object list format is not supported.\n"}},"type":"object","required":["role","content"],"title":"NIMLLMChatCompletionMessage"},"type":"array","minItems":1,"title":"Messages","description":"A list of messages comprising the conversation so far."},"model":{"type":"string","title":"Model","description":"The model to use.","default":"google/gemma-3n-e4b-it"},"frequency_penalty":{"anyOf":[{"type":"number","maximum":2,"minimum":-2},{"type":"null"}],"title":"Frequency Penalty","description":"Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.","default":0},"max_tokens":{"anyOf":[{"type":"integer","maximum":8192,"minimum":1},{"type":"null"}],"title":"Max Tokens","description":"The maximum number of tokens that can be generated.","default":512},"presence_penalty":{"anyOf":[{"type":"number","maximum":2,"minimum":-2},{"type":"null"}],"title":"Presence Penalty","description":"Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.","default":0},"seed":{"anyOf":[{"type":"integer","maximum":9223372036854776000,"minimum":-9223372036854776000},{"type":"null"}],"title":"Seed","description":"Changing the seed will produce a different response with similar characteristics. Fixing the seed will reproduce the same results if all other parameters are also kept constant."},"stop":{"anyOf":[{"type":"string"},{"items":{"type":"string"},"type":"array"},{"type":"null"}],"title":"Stop","description":"Sequences where the API will stop generating further tokens."},"stream":{"anyOf":[{"type":"boolean"},{"type":"null"}],"title":"Stream","description":"If set, partial message deltas will be sent, like in ChatGPT. Tokens will be sent as data-only [server-sent events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format) as they become available, with the stream terminated by a `data: [DONE]`","default":false},"temperature":{"anyOf":[{"type":"number","maximum":2,"minimum":0},{"type":"null"}],"title":"Temperature","description":"What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.","default":0.2},"top_p":{"anyOf":[{"type":"number","maximum":1,"exclusiveMinimum":0},{"type":"null"}],"title":"Top P","description":"An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or `temperature` but not both.","default":0.7}},"additionalProperties":false,"type":"object","required":["messages","model"],"title":"NIMLLMChatCompletionRequest"}}},"required":true},"responses":{"200":{"description":"Invocation is fulfilled","content":{"application/json":{"schema":{"properties":{"id":{"type":"string","title":"Id"},"object":{"type":"string","enum":["chat.completion"],"const":"chat.completion","title":"Object","default":"chat.completion"},"created":{"type":"integer","title":"Created"},"model":{"type":"string","title":"Model"},"choices":{"items":{"properties":{"index":{"type":"integer","title":"Index"},"message":{"properties":{"role":{"type":"string","title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","required":["role"],"title":"ChatMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","message"],"title":"ChatCompletionResponseChoice"},"type":"array","title":"Choices"},"usage":{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"}},"additionalProperties":false,"type":"object","required":["model","choices","usage"],"title":"ChatCompletionResponse"}},"text/event-stream":{"schema":{"properties":{"id":{"type":"string","title":"Id"},"object":{"type":"string","enum":["chat.completion.chunk"],"const":"chat.completion.chunk","title":"Object","default":"chat.completion.chunk"},"created":{"type":"integer","title":"Created"},"model":{"type":"string","title":"Model"},"choices":{"items":{"properties":{"index":{"type":"integer","title":"Index"},"delta":{"properties":{"role":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","title":"DeltaMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","delta"],"title":"ChatCompletionResponseStreamChoice"},"type":"array","title":"Choices"},"usage":{"anyOf":[{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"},{"type":"null"}]}},"additionalProperties":false,"type":"object","required":["model","choices"],"title":"ChatCompletionStreamResponse"}}}},"202":{"description":"Result is pending. Client should poll using the requestId.","content":{"application/json":{"example":{},"schema":{}}},"headers":{"NVCF-REQID":{"description":"requestId required for pooling","schema":{"type":"string","format":"uuid","maxLength":36}},"NVCF-STATUS":{"description":"Invocation status","schema":{"type":"string","format":"^[a-zA-Z-]{1,64}$","maxLength":64}}}},"422":{"description":"Validation failed, provided entity could not be processed.","content":{"application/json":{"schema":{"properties":{"object":{"type":"string","title":"Object","default":"error"},"message":{"type":"string","title":"Message"},"type":{"type":"string","title":"Type"},"param":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Param"},"code":{"type":"integer","title":"Code"}},"additionalProperties":false,"type":"object","required":["message","type","code"],"title":"ErrorResponse"},"example":{"type":"urn:nvcf-worker-service:problem-details:unprocessable-entity","title":"Unprocessable Entity","status":422,"detail":"string","instance":"/v2/nvcf/pexec/functions/e598bfc1-b058-41af-869d-556d3c7e1b48","requestId":"3fa85f64-5717-4562-b3fc-2c963f66afa6"}}}},"500":{"description":"The invocation ended with an error.","content":{"application/json":{"schema":{"properties":{"object":{"type":"string","title":"Object","default":"error"},"message":{"type":"string","title":"Message"},"type":{"type":"string","title":"Type"},"param":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Param"},"code":{"type":"integer","title":"Code"}},"additionalProperties":false,"type":"object","required":["message","type","code"],"title":"ErrorResponse"},"example":{"type":"urn:nvcf-worker-service:problem-details:internal-server-error","title":"Internal Server Error","status":500,"detail":"string","instance":"/v2/nvcf/pexec/functions/e598bfc1-b058-41af-869d-556d3c7e1b48","requestId":"3fa85f64-5717-4562-b3fc-2c963f66afa6"}}}}},"x-nvai-meta":{"name":"Create chat completion","returns":"Returns a [chat completion](/docs/api-reference/chat/object) object, or a streamed sequence of [chat completion chunk](/docs/api-reference/chat/streaming) objects if the request is streamed.\n","path":"create","examples":[{"name":"image_captioning","input":{"text":"Describe this image.","images":["https://assets.ngc.nvidia.com/products/api-catalog/gemma3n/lake_louise.jpg"]}},{"name":"multi_image_comparison","input":{"text":"What are common about these two images?","images":["https://assets.ngc.nvidia.com/products/api-catalog/gemma3n/new_york.jpg","https://assets.ngc.nvidia.com/products/api-catalog/gemma3n/merlion.png"]}},{"name":"speech_recognition","input":{"text":"Transcribe the following speech segment in English:","audios":["https://assets.ngc.nvidia.com/products/api-catalog/gemma3n/speech.wav"]}},{"name":"audio_visual_understanding","input":{"text":"","audios":["https://assets.ngc.nvidia.com/products/api-catalog/phi-4-mini-mm/what_is_the_traffic_sign_in_the_image.wav"],"images":["https://assets.ngc.nvidia.com/products/api-catalog/phi-4-mini-mm/traffic_sign.jpg"]}}],"templates":[{"title":"Default","response":"{\n \"choices\": [{\n \"index\": 0,\n \"model\": \"google/gemma-3n-e4b-it\",\n \"message\": {\n \"role\": \"assistant\",\n \"content\": \"...\",\n },\n \"finish_reason\": \"stop\"\n }],\n \"usage\": {\n \"prompt_tokens\": 9,\n \"completion_tokens\": 12,\n \"total_tokens\": 21\n }\n}\n","requestEjs":{"python":"$62","langChain":"$63","node.js":"$64","curl":"$65"}}]}}},"/status/{requestId}":{"get":{"tags":["Multimodal API"],"summary":"Gets the result of an earlier function invocation request that returned a status of 202.","operationId":"getFunctionInvocationResult","parameters":[{"name":"requestId","in":"path","description":"requestId to poll results","required":true,"schema":{"type":"string","format":"uuid","maxLength":36}}],"responses":{"200":{"description":"Invocation is fulfilled","content":{"application/json":{"schema":{"properties":{"id":{"type":"string","title":"Id"},"object":{"type":"string","enum":["chat.completion"],"const":"chat.completion","title":"Object","default":"chat.completion"},"created":{"type":"integer","title":"Created"},"model":{"type":"string","title":"Model"},"choices":{"items":{"properties":{"index":{"type":"integer","title":"Index"},"message":{"properties":{"role":{"type":"string","title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","required":["role"],"title":"ChatMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","message"],"title":"ChatCompletionResponseChoice"},"type":"array","title":"Choices"},"usage":{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"}},"additionalProperties":false,"type":"object","required":["model","choices","usage"],"title":"ChatCompletionResponse"}}}},"202":{"description":"Result is pending. Client should poll using the requestId.","content":{"application/json":{"example":{},"schema":{}}},"headers":{"NVCF-REQID":{"description":"requestId required for pooling","schema":{"type":"string","format":"uuid","maxLength":36}},"NVCF-STATUS":{"description":"Invocation status","schema":{"type":"string","format":"^[a-zA-Z-]{1,64}$","maxLength":64}}}},"422":{"description":"The invocation ended with an error.","content":{"application/json":{"schema":{"properties":{"object":{"type":"string","title":"Object","default":"error"},"message":{"type":"string","title":"Message"},"type":{"type":"string","title":"Type"},"param":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Param"},"code":{"type":"integer","title":"Code"}},"additionalProperties":false,"type":"object","required":["message","type","code"],"title":"ErrorResponse"},"example":{"type":"urn:nvcf-worker-service:problem-details:unprocessable-entity","title":"Unprocessable Entity","status":422,"detail":"string","instance":"/v2/nvcf/pexec/functions/e598bfc1-b058-41af-869d-556d3c7e1b48","requestId":"3fa85f64-5717-4562-b3fc-2c963f66afa6"}}}},"500":{"description":"The invocation ended with an error.","content":{"application/json":{"schema":{"properties":{"object":{"type":"string","title":"Object","default":"error"},"message":{"type":"string","title":"Message"},"type":{"type":"string","title":"Type"},"param":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Param"},"code":{"type":"integer","title":"Code"}},"additionalProperties":false,"type":"object","required":["message","type","code"],"title":"ErrorResponse"},"example":{"type":"urn:nvcf-worker-service:problem-details:internal-server-error","title":"Internal Server Error","status":500,"detail":"string","instance":"/v2/nvcf/pexec/functions/e598bfc1-b058-41af-869d-556d3c7e1b48","requestId":"3fa85f64-5717-4562-b3fc-2c963f66afa6"}}}}}}}},"security":[{"Token":[]}],"components":{"securitySchemes":{"Token":{"type":"http","scheme":"bearer"}},"schemas":{"ChatCompletionResponse":{"properties":{"id":{"type":"string","title":"Id"},"object":{"type":"string","enum":["chat.completion"],"const":"chat.completion","title":"Object","default":"chat.completion"},"created":{"type":"integer","title":"Created"},"model":{"type":"string","title":"Model"},"choices":{"items":{"properties":{"index":{"type":"integer","title":"Index"},"message":{"properties":{"role":{"type":"string","title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","required":["role"],"title":"ChatMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","message"],"title":"ChatCompletionResponseChoice"},"type":"array","title":"Choices"},"usage":{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"}},"additionalProperties":false,"type":"object","required":["model","choices","usage"],"title":"ChatCompletionResponse"},"ChatCompletionResponseChoice":{"properties":{"index":{"type":"integer","title":"Index"},"message":{"properties":{"role":{"type":"string","title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","required":["role"],"title":"ChatMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","message"],"title":"ChatCompletionResponseChoice"},"ChatCompletionResponseStreamChoice":{"properties":{"index":{"type":"integer","title":"Index"},"delta":{"properties":{"role":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","title":"DeltaMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","delta"],"title":"ChatCompletionResponseStreamChoice"},"ChatCompletionStreamResponse":{"properties":{"id":{"type":"string","title":"Id"},"object":{"type":"string","enum":["chat.completion.chunk"],"const":"chat.completion.chunk","title":"Object","default":"chat.completion.chunk"},"created":{"type":"integer","title":"Created"},"model":{"type":"string","title":"Model"},"choices":{"items":{"properties":{"index":{"type":"integer","title":"Index"},"delta":{"properties":{"role":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","title":"DeltaMessage"},"finish_reason":{"anyOf":[{"type":"string","enum":["stop","length"]},{"type":"null"}],"title":"Finish Reason"},"stop_reason":{"anyOf":[{"type":"integer"},{"type":"string"},{"type":"null"}],"title":"Stop Reason"}},"additionalProperties":false,"type":"object","required":["index","delta"],"title":"ChatCompletionResponseStreamChoice"},"type":"array","title":"Choices"},"usage":{"anyOf":[{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"},{"type":"null"}]}},"additionalProperties":false,"type":"object","required":["model","choices"],"title":"ChatCompletionStreamResponse"},"ChatMessage":{"properties":{"role":{"type":"string","title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","required":["role"],"title":"ChatMessage"},"DeltaMessage":{"properties":{"role":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Role"},"content":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Content"}},"additionalProperties":false,"type":"object","title":"DeltaMessage"},"ErrorResponse":{"properties":{"object":{"type":"string","title":"Object","default":"error"},"message":{"type":"string","title":"Message"},"type":{"type":"string","title":"Type"},"param":{"anyOf":[{"type":"string"},{"type":"null"}],"title":"Param"},"code":{"type":"integer","title":"Code"}},"additionalProperties":false,"type":"object","required":["message","type","code"],"title":"ErrorResponse"},"HTTPValidationError":{"properties":{"detail":{"items":{"properties":{"loc":{"items":{"anyOf":[{"type":"string"},{"type":"integer"}]},"type":"array","title":"Location"},"msg":{"type":"string","title":"Message"},"type":{"type":"string","title":"Error Type"}},"type":"object","required":["loc","msg","type"],"title":"ValidationError"},"type":"array","title":"Detail"}},"type":"object","title":"HTTPValidationError"},"AudioURL":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"AudioURL"},"InputAudio":{"properties":{"data":{"type":"string","title":"Data"},"format":{"type":"string","enum":["wav","mp3"],"title":"Format"}},"type":"object","required":["data","format"],"title":"InputAudio"},"ImageURL":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"ImageURL"},"NIMLLMChatCompletionMessage":{"properties":{"role":{"allOf":[{"type":"string","enum":["assistant","user"],"title":"Role"}],"description":"The role of the message's author."},"content":{"anyOf":[{"type":"string"},{"items":{"anyOf":[{"properties":{"audio_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"AudioURL"},"type":{"type":"string","enum":["audio_url"],"const":"audio_url","title":"Type"}},"type":"object","required":["audio_url","type"],"title":"NIMLLMChatCompletionContentPartAudio"},{"properties":{"image_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"ImageURL"},"type":{"type":"string","enum":["image_url"],"const":"image_url","title":"Type"}},"type":"object","required":["image_url","type"],"title":"NIMLLMChatCompletionContentPartImage"},{"properties":{"text":{"type":"string","title":"Text"},"type":{"type":"string","enum":["text"],"const":"text","title":"Type"}},"type":"object","required":["text","type"],"title":"NIMLLMChatCompletionContentPartText"}]},"type":"array"}],"title":"Content","description":"The contents of the message.\n
To pass media (only with role=`user`):\n
- Use content parts. Text is passed with `type`=`text`.\n
- Images are passed with `type`=`image_url`; set `image_url.url` to an image URL or a base64 data URI like `data:image/{format};base64,{base64encodedimage}`.\n
- Audio is passed with `type`=`audio_url` or `type`=`input_audio`; set `audio_url.url` to an audio URL or a base64 data URI, or set `input_audio.data` to base64 audio data.\n
For `system` and `assistant` roles, the object list format is not supported.\n"}},"type":"object","required":["role","content"],"title":"NIMLLMChatCompletionMessage"},"NIMLLMChatCompletionRequest":{"properties":{"messages":{"items":{"properties":{"role":{"allOf":[{"type":"string","enum":["assistant","user"],"title":"Role"}],"description":"The role of the message's author."},"content":{"anyOf":[{"type":"string"},{"items":{"anyOf":[{"properties":{"audio_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"AudioURL"},"type":{"type":"string","enum":["audio_url"],"const":"audio_url","title":"Type"}},"type":"object","required":["audio_url","type"],"title":"NIMLLMChatCompletionContentPartAudio"},{"properties":{"image_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"ImageURL"},"type":{"type":"string","enum":["image_url"],"const":"image_url","title":"Type"}},"type":"object","required":["image_url","type"],"title":"NIMLLMChatCompletionContentPartImage"},{"properties":{"text":{"type":"string","title":"Text"},"type":{"type":"string","enum":["text"],"const":"text","title":"Type"}},"type":"object","required":["text","type"],"title":"NIMLLMChatCompletionContentPartText"}]},"type":"array"}],"title":"Content","description":"The contents of the message.\n
To pass media (only with role=`user`):\n
- Use content parts. Text is passed with `type`=`text`.\n
- Images are passed with `type`=`image_url`; set `image_url.url` to an image URL or a base64 data URI like `data:image/{format};base64,{base64encodedimage}`.\n
- Audio is passed with `type`=`audio_url` or `type`=`input_audio`; set `audio_url.url` to an audio URL or a base64 data URI, or set `input_audio.data` to base64 audio data.\n
For `system` and `assistant` roles, the object list format is not supported.\n"}},"type":"object","required":["role","content"],"title":"NIMLLMChatCompletionMessage"},"type":"array","minItems":1,"title":"Messages","description":"A list of messages comprising the conversation so far."},"model":{"type":"string","title":"Model","description":"The model to use.","default":"google/gemma-3n-e4b-it"},"frequency_penalty":{"anyOf":[{"type":"number","maximum":2,"minimum":-2},{"type":"null"}],"title":"Frequency Penalty","description":"Number between -2.0 and 2.0. Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.","default":0},"max_tokens":{"anyOf":[{"type":"integer","maximum":8192,"minimum":1},{"type":"null"}],"title":"Max Tokens","description":"The maximum number of tokens that can be generated.","default":512},"presence_penalty":{"anyOf":[{"type":"number","maximum":2,"minimum":-2},{"type":"null"}],"title":"Presence Penalty","description":"Number between -2.0 and 2.0. Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.","default":0},"seed":{"anyOf":[{"type":"integer","maximum":9223372036854776000,"minimum":-9223372036854776000},{"type":"null"}],"title":"Seed","description":"Changing the seed will produce a different response with similar characteristics. Fixing the seed will reproduce the same results if all other parameters are also kept constant."},"stop":{"anyOf":[{"type":"string"},{"items":{"type":"string"},"type":"array"},{"type":"null"}],"title":"Stop","description":"Sequences where the API will stop generating further tokens."},"stream":{"anyOf":[{"type":"boolean"},{"type":"null"}],"title":"Stream","description":"If set, partial message deltas will be sent, like in ChatGPT. Tokens will be sent as data-only [server-sent events](https://developer.mozilla.org/en-US/docs/Web/API/Server-sent_events/Using_server-sent_events#Event_stream_format) as they become available, with the stream terminated by a `data: [DONE]`","default":false},"temperature":{"anyOf":[{"type":"number","maximum":2,"minimum":0},{"type":"null"}],"title":"Temperature","description":"What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.","default":0.2},"top_p":{"anyOf":[{"type":"number","maximum":1,"exclusiveMinimum":0},{"type":"null"}],"title":"Top P","description":"An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered. We generally recommend altering this or `temperature` but not both.","default":0.7}},"additionalProperties":false,"type":"object","required":["messages","model"],"title":"NIMLLMChatCompletionRequest"},"NIMLLMChatCompletionContentPartAudio":{"properties":{"audio_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"AudioURL"},"type":{"type":"string","enum":["audio_url"],"const":"audio_url","title":"Type"}},"type":"object","required":["audio_url","type"],"title":"NIMLLMChatCompletionContentPartAudio"},"NIMLLMChatCompletionContentPartInputAudio":{"properties":{"input_audio":{"properties":{"data":{"type":"string","title":"Data"},"format":{"type":"string","enum":["wav","mp3"],"title":"Format"}},"type":"object","required":["data","format"],"title":"InputAudio"},"type":{"type":"string","const":"input_audio","title":"Type"}},"type":"object","required":["input_audio","type"],"title":"NIMLLMChatCompletionContentPartInputAudio"},"NIMLLMChatCompletionContentPartImage":{"properties":{"image_url":{"properties":{"url":{"type":"string","title":"Url"}},"type":"object","required":["url"],"title":"ImageURL"},"type":{"type":"string","enum":["image_url"],"const":"image_url","title":"Type"}},"type":"object","required":["image_url","type"],"title":"NIMLLMChatCompletionContentPartImage"},"NIMLLMChatCompletionContentPartText":{"properties":{"text":{"type":"string","title":"Text"},"type":{"type":"string","enum":["text"],"const":"text","title":"Type"}},"type":"object","required":["text","type"],"title":"NIMLLMChatCompletionContentPartText"},"Role":{"type":"string","enum":["assistant","user"],"title":"Role"},"UsageInfo":{"properties":{"prompt_tokens":{"type":"integer","title":"Prompt Tokens","default":0},"total_tokens":{"type":"integer","title":"Total Tokens","default":0},"completion_tokens":{"anyOf":[{"type":"integer"},{"type":"null"}],"title":"Completion Tokens","default":0}},"additionalProperties":false,"type":"object","title":"UsageInfo"},"ValidationError":{"properties":{"loc":{"items":{"anyOf":[{"type":"string"},{"type":"integer"}]},"type":"array","title":"Location"},"msg":{"type":"string","title":"Message"},"type":{"type":"string","title":"Error Type"}},"type":"object","required":["loc","msg","type"],"title":"ValidationError"}}}},"pageLayout":"v3"},"children":["$L66","$L67","$L68"]}]}]]