The Meta Llama 3.2 collection of multilingual large language models (LLMs) is a collection of pretrained and instruction-tuned generative models in 1B and 3B sizes (text in/text out). The Llama 3.2 instruction-tuned text only models are optimized for multilingual dialogue use cases, including agentic retrieval and summarization tasks. They outperform many of the available open source and closed chat models on common industry benchmarks.
Model Developer: Meta
Model Architecture: Llama 3.2 is an auto-regressive language model that uses an optimized transformer architecture. The tuned versions use supervised fine-tuning (SFT) and reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety.
| Training Data | Params | Input modalities | Output modalities | Context Length | GQA | Shared Embeddings | Token count | Knowledge cutoff | |
|---|---|---|---|---|---|---|---|---|---|
| Llama 3.2 (text only) | A new mix of publicly available online data. | 1B (1.23B) | Multilingual Text | Multilingual Text and code | 128k | Yes | Yes | Up to 9T tokens | December 2023 |
| 3B (3.21B) | Multilingual Text | Multilingual Text and code |
Supported Languages: English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai are officially supported. Llama 3.2 has been trained on a broader collection of languages than these 8 supported languages. Developers may fine-tune Llama 3.2 models for languages beyond these supported languages, provided they comply with the Llama 3.2 Community License and the Acceptable Use Policy. Developers are always expected to ensure that their deployments, including those that involve additional languages, are completed safely and responsibly.
Llama 3.2 Model Family: Token counts refer to pretraining data only. All model versions use Grouped-Query Attention (GQA) for improved inference scalability.
Model Release Date: Sept 25, 2024
Status: This is a static model trained on an offline dataset. Future versions may be released that improve model capabilities and safety.
License: Use of Llama 3.2 is governed by the Llama 3.2 Community License (a custom, commercial license agreement).
Feedback: Where to send questions or comments about the model Instructions on how to provide feedback or comments on the model can be found in the model README. For more technical information about generation parameters and recipes for how to use Llama 3.2 in applications, please go here.
Intended Use Cases: Llama 3.2 is intended for commercial and research use in multiple languages. Instruction tuned text only models are intended for assistant-like chat and agentic applications like knowledge retrieval and summarization, mobile AI powered writing assistants and query and prompt rewriting. Pretrained models can be adapted for a variety of additional natural language generation tasks.
Out of Scope: Use in any manner that violates applicable laws or regulations (including trade compliance laws). Use in any other way that is prohibited by the Acceptable Use Policy and Llama 3.2 Community License. Use in languages beyond those explicitly referenced as supported in this model card.
You can choose 3 programming languages to access our meta-llama/llama-3.2-3b-instruct model.
We provide compatibility with the OpenAI API standard
The API Base URL
1https://api.novita.ai/v3/openai
Example of Using Chat Completions API
Generate a response using a list of messages from a conversation
1# Get the Novita AI API Key by referring to: https://novita.ai/docs/get-started/quickstart.html#_2-manage-api-key 2export API_KEY="{YOUR Novita AI API Key}" 3 4curl "https://api.novita.ai/v3/openai/chat/completions" \ 5 -H "Content-Type: application/json" \ 6 -H "Authorization: Bearer ${API_KEY}" \ 7 -d '{ 8 "model": "meta-llama/llama-3.2-3b-instruct", 9 "messages": [ 10 { 11 "role": "system", 12 "content": "Act like you are a helpful assistant." 13 }, 14 { 15 "role": "user", 16 "content": "Hi there!" 17 } 18 ], 19 "max_tokens": 512 20}'
The response may look like this
1{ 2 "id": "chat-5f461a9a23a44ef29dbd3124b891afc0", 3 "object": "chat.completion", 4 "created": 1731584707, 5 "model": "meta-llama/llama-3.2-3b-instruct", 6 "choices": [ 7 { 8 "index": 0, 9 "message": { 10 "role": "assistant", 11 "content": "Hello! It's nice to meet you. How can I assist you today? Do you have any questions or topics you'd like to discuss? I'm here to help with anything you need." 12 }, 13 "finish_reason": "stop", 14 "content_filter_results": { 15 "hate": { "filtered": false }, 16 "self_harm": { "filtered": false }, 17 "sexual": { "filtered": false }, 18 "violence": { "filtered": false }, 19 "jailbreak": { "filtered": false, "detected": false }, 20 "profanity": { "filtered": false, "detected": false } 21 } 22 } 23 ], 24 "usage": { 25 "prompt_tokens": 46, 26 "completion_tokens": 40, 27 "total_tokens": 86, 28 "prompt_tokens_details": null, 29 "completion_tokens_details": null 30 }, 31 "system_fingerprint": "" 32}
If you want to receive a response via streaming, simply pass "stream": true in the request (see the difference on line 20). An example is provided.
1# Get the Novita AI API Key by referring to: https://novita.ai/docs/get-started/quickstart.html#_2-manage-api-key 2export API_KEY="{YOUR Novita AI API Key}" 3 4curl "https://api.novita.ai/v3/openai/chat/completions" \ 5 -H "Content-Type: application/json" \ 6 -H "Authorization: Bearer ${API_KEY}" \ 7 -d '{ 8 "model": "meta-llama/llama-3.2-3b-instruct", 9 "messages": [ 10 { 11 "role": "system", 12 "content": "Act like you are a helpful assistant." 13 }, 14 { 15 "role": "user", 16 "content": "Hi there!" 17 } 18 ], 19 "max_tokens": 512, 20 "stream": true 21}'
The response may look like this
1data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 2 3... 4 5data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"n, ne"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 6 7data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"ed"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 8 9data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":" assi"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 10 11data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"s"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 12 13data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"tan"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 14 15data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"ce wi"},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 16 17... 18 19data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":" "},"finish_reason":null,"content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 20 21data: {"id":"chat-d821b951d6ff43ab838d18137aef7d0a","object":"chat.completion.chunk","created":1731586102,"model":"meta-llama/llama-3.1-8b-instruct","choices":[{"index":0,"delta":{"content":"just want to chat?"},"finish_reason":"stop","content_filter_results":{"hate":{"filtered":false},"self_harm":{"filtered":false},"sexual":{"filtered":false},"violence":{"filtered":false},"jailbreak":{"filtered":false,"detected":false},"profanity":{"filtered":false,"detected":false}}}],"system_fingerprint":""} 22 23data: [DONE]
Model Parameters
Feel free to check out our documentation for more details.
First, install the official OpenAI Python client
1pip install 'openai>=1.0.0'
and then you can run inferences with us
Example of Using Chat Completions API
Generate a response using a list of messages from a conversation
1from openai import OpenAI 2 3client = OpenAI( 4 base_url="https://api.novita.ai/v3/openai", 5 # Get the Novita AI API Key by referring to: https://novita.ai/docs/get-started/quickstart.html#_2-manage-api-key. 6 api_key="<YOUR Novita AI API Key>", 7) 8 9model = "meta-llama/llama-3.2-3b-instruct" 10stream = True # or False 11max_tokens = 512 12 13chat_completion_res = client.chat.completions.create( 14 model=model, 15 messages=[ 16 { 17 "role": "system", 18 "content": "Act like you are a helpful assistant.", 19 }, 20 { 21 "role": "user", 22 "content": "Hi there!", 23 } 24 ], 25 stream=stream, 26 max_tokens=max_tokens, 27) 28 29if stream: 30 for chunk in chat_completion_res: 31 print(chunk.choices[0].delta.content or "") 32else: 33 print(chat_completion_res.choices[0].message.content)
If you set stream: true (line 10), the print may look like this
1It' 2s 3 ni 4ce to 5meet you. 6Is 7 the 8re so 9meth 10ing I 11 can h 12e 13lp 14you wi 15th t 16oday, 17 or 18 woul 19d 20 you like to chat?
If you don't want to receive a response via streaming, simply set stream: false. The output will look like this
1How can I assist you today? Do you have any questions or topics you'd like to discuss?
Model Parameters
Feel free to check out our documentation for more details.
First, install the official OpenAI JavaScript client
1npm install openai
and then you can run inferences with us in the browser or in node.js
Example of Using Chat Completions API
Generate a response using a list of messages from a conversation
1import OpenAI from "openai"; 2 3const openai = new OpenAI({ 4 baseURL: "https://api.novita.ai/v3/openai", 5 apiKey: "<YOUR Novita AI API Key>", 6}); 7const stream = true; // or false 8 9async function run() { 10 const completion = await openai.chat.completions.create({ 11 messages: [ 12 { 13 role: "system", 14 content: "Act like you are a helpful assistant.", 15 }, 16 { 17 role: "user", 18 content: "Hi there!" 19 } 20 ], 21 model: "meta-llama/llama-3.2-3b-instruct", 22 stream 23 }); 24 25 if (stream) { 26 for await (const chunk of completion) { 27 if (chunk.choices[0].finish_reason) { 28 console.log(chunk.choices[0].finish_reason); 29 } else { 30 console.log(chunk.choices[0].delta.content); 31 } 32 } 33 } else { 34 console.log(JSON.stringify(completion)); 35 } 36} 37 38run();
If you set stream: true (line 7), the print may look like this
1It' 2s 3 nic 4e to 5 m 6eet you 7. Ho 8w can 9I 10 as 11sist 12 you 13toda 14y? Do you 15hav 16e any q 17uest 18io 19ns or 20 to 21pics you 22' 23d 24li 25ke to 26 di 27scuss 28stop
If you don't want to receive a response via streaming, simply set stream: false. The output will look like this
1{ 2 "id": "chat-a3ff0e39b4c24abcbd258ab1a1f38db9", 3 "object": "chat.completion", 4 "created": 1731642457, 5 "model": "meta-llama/llama-3.2-3b-instruct", 6 "choices": [ 7 { 8 "index": 0, 9 "message": { 10 "role": "assistant", 11 "content": "How can I help you today? Would you like to talk about something specific or just have a chat? I'm here to assist you with any questions or information you might need." 12 }, 13 "finish_reason": "stop", 14 "content_filter_results": { 15 "hate": { "filtered": false }, 16 "self_harm": { "filtered": false }, 17 "sexual": { "filtered": false }, 18 "violence": { "filtered": false }, 19 "jailbreak": { "filtered": false, "detected": false }, 20 "profanity": { "filtered": false, "detected": false } 21 } 22 } 23 ], 24 "usage": { 25 "prompt_tokens": 46, 26 "completion_tokens": 37, 27 "total_tokens": 83, 28 "prompt_tokens_details": null, 29 "completion_tokens_details": null 30 }, 31 "system_fingerprint": "" 32}
Model Parameters
Feel free to check out our documentation for more details.
Training Factors: We used custom training libraries, Meta's custom built GPU cluster, and production infrastructure for pretraining. Fine-tuning, annotation, and evaluation were also performed on production infrastructure.
Training Energy Use: Training utilized a cumulative of 916k GPU hours of computation on H100-80GB (TDP of 700W) type hardware, per the table below. Training time is the total GPU time required for training each model and power consumption is the peak power capacity per GPU device used, adjusted for power usage efficiency.
Training Greenhouse Gas Emissions: Estimated total location-based greenhouse gas emissions were 240 tons CO2eq for training. Since 2020, Meta has maintained net zero greenhouse gas emissions in its global operations and matched 100% of its electricity use with renewable energy; therefore, the total market-based greenhouse gas emissions for training were 0 tons CO2eq.
| Training Time (GPU hours) | Logit Generation Time (GPU Hours) | Training Power Consumption (W) | Training Location-Based Greenhouse Gas Emissions (tons CO2eq) | Training Market-Based Greenhouse Gas Emissions (tons CO2eq) | |
|---|---|---|---|---|---|
| Llama 3.2 1B | 370k | - | 700 | 107 | 0 |
| Llama 3.2 3B | 460k | - | 700 | 133 | 0 |
| Total | 830k | 86k | 240 | 0 |
The methodology used to determine training energy use and greenhouse gas emissions can be found here. Since Meta is openly releasing these models, the training energy use and greenhouse gas emissions will not be incurred by others.
Overview: Llama 3.2 was pretrained on up to 9 trillion tokens of data from publicly available sources. For the 1B and 3B Llama 3.2 models, we incorporated logits from the Llama 3.1 8B and 70B models into the pretraining stage of the model development, where outputs (logits) from these larger models were used as token-level targets. Knowledge distillation was used after pruning to recover performance. In post-training we used a similar recipe as Llama 3.1 and produced final chat models by doing several rounds of alignment on top of the pre-trained model. Each round involved Supervised Fine-Tuning (SFT), Rejection Sampling (RS), and Direct Preference Optimization (DPO).
Data Freshness: The pretraining data has a cutoff of December 2023.
In this section, we report the results for Llama 3.2 models on standard automatic benchmarks. For all these evaluations, we used our internal evaluations library.
| Category | Benchmark | # Shots | Metric | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
|---|---|---|---|---|---|---|
| General | MMLU | 5 | macro_avg/acc_char | 32.2 | 58 | 66.7 |
| AGIEval English | 3月5日 | average/acc_char | 23.3 | 39.2 | 47.8 | |
| ARC-Challenge | 25 | acc_char | 32.8 | 69.1 | 79.7 | |
| Reading comprehension | SQuAD | 1 | em | 49.2 | 67.7 | 77 |
| QuAC (F1) | 1 | f1 | 37.9 | 42.9 | 44.9 | |
| DROP (F1) | 3 | f1 | 28 | 45.2 | 59.5 | |
| Long Context | Needle in Haystack | 0 | em | 96.8 | 1 | 1 |
| Capability | Benchmark | # Shots | Metric | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
|---|---|---|---|---|---|---|
| General | MMLU | 5 | macro_avg/acc | 49.3 | 63.4 | 69.4 |
| Re-writing | Open-rewrite eval | 0 | micro_avg/rougeL | 41.6 | 40.1 | 40.9 |
| Summarization | TLDR9+ (test) | 1 | rougeL | 16.8 | 19 | 17.2 |
| Instruction following | IFEval | 0 | avg(prompt/instruction acc loose/strict) | 59.5 | 77.4 | 80.4 |
| Math | GSM8K (CoT) | 8 | em_maj1@1 | 44.4 | 77.7 | 84.5 |
| MATH (CoT) | 0 | final_em | 30.6 | 47.3 | 51.9 | |
| Reasoning | ARC-C | 0 | acc | 59.4 | 78.6 | 83.4 |
| GPQA | 0 | acc | 27.2 | 32.8 | 32.8 | |
| Hellaswag | 0 | acc | 41.2 | 69.8 | 78.7 | |
| Tool Use | BFCL V2 | 0 | acc | 25.7 | 67 | 70.9 |
| Nexus | 0 | macro_avg/acc | 13.5 | 34.3 | 38.5 | |
| Long Context | InfiniteBench/En.QA | 0 | longbook_qa/f1 | 20.3 | 19.8 | 27.3 |
| InfiniteBench/En.MC | 0 | longbook_choice/acc | 38 | 63.3 | 72.2 | |
| NIH/Multi-needle | 0 | recall | 75 | 84.7 | 98.8 | |
| Multilingual | MGSM (CoT) | 0 | em | 24.5 | 58.2 | 68.9 |
| Category | Benchmark | Language | Llama 3.2 1B | Llama 3.2 3B | Llama 3.1 8B |
|---|---|---|---|---|---|
| General | MMLU (5-shot, macro_avg/acc) | Portuguese | 39.82 | 54.48 | 62.12 |
| Spanish | 41.5 | 55.1 | 62.5 | ||
| Italian | 39.8 | 53.8 | 61.6 | ||
| German | 39.2 | 53.3 | 60.6 | ||
| French | 40.5 | 54.6 | 62.3 | ||
| Hindi | 33.5 | 43.3 | 50.9 | ||
| Thai | 34.7 | 44.5 | 50.3 |
As part of our Responsible release approach, we followed a three-pronged strategy to managing trust & safety risks:
Approach: Llama is a foundational technology designed to be used in a variety of use cases. Examples on how Meta’s Llama models have been responsibly deployed can be found in our Community Stories webpage. Our approach is to build the most helpful models, enabling the world to benefit from the technology power, by aligning our model safety for generic use cases and addressing a standard set of harms. Developers are then in the driver’s seat to tailor safety for their use cases, defining their own policies and deploying the models with the necessary safeguards in their Llama systems. Llama 3.2 was developed following the best practices outlined in our Responsible Use Guide.
Objective: Our main objectives for conducting safety fine-tuning are to provide the research community with a valuable resource for studying the robustness of safety fine-tuning, as well as to offer developers a readily available, safe, and powerful model for various applications to reduce the developer workload to deploy safe AI systems. We implemented the same set of safety mitigations as in Llama 3, and you can learn more about these in the Llama 3 paper.
Fine-Tuning Data: We employ a multi-faceted approach to data collection, combining human-generated data from our vendors with synthetic data to mitigate potential safety risks. We’ve developed many large language model (LLM)-based classifiers that enable us to thoughtfully select high-quality prompts and responses, enhancing data quality control.
Refusals and Tone: Building on the work we started with Llama 3, we put a great emphasis on model refusals to benign prompts as well as refusal tone. We included both borderline and adversarial prompts in our safety data strategy, and modified our safety data responses to follow tone guidelines.
Safety as a System: Large language models, including Llama 3.2, are not designed to be deployed in isolation but instead should be deployed as part of an overall AI system with additional safety guardrails as required. Developers are expected to deploy system safeguards when building agentic systems. Safeguards are key to achieve the right helpfulness-safety alignment as well as mitigating safety and security risks inherent to the system and any integration of the model or system with external tools. As part of our responsible release approach, we provide the community with safeguards that developers should deploy with Llama models or other LLMs, including Llama Guard, Prompt Guard and Code Shield. All our reference implementations demos contain these safeguards by default so developers can benefit from system-level safety out-of-the-box.
Technological Advancement: Llama releases usually introduce new capabilities that require specific considerations in addition to the best practices that generally apply across all Generative AI use cases. For prior release capabilities also supported by Llama 3.2, see Llama 3.1 Model Card, as the same considerations apply here as well.
Constrained Environments: Llama 3.2 1B and 3B models are expected to be deployed in highly constrained environments, such as mobile devices. LLM Systems using smaller models will have a different alignment profile and safety/helpfulness tradeoff than more complex, larger systems. Developers should ensure the safety of their system meets the requirements of their use case. We recommend using lighter system safeguards for such use cases, like Llama Guard 3-1B or its mobile-optimized version.
Scaled Evaluations: We built dedicated, adversarial evaluation datasets and evaluated systems composed of Llama models and Purple Llama safeguards to filter input prompt and output response. It is important to evaluate applications in context, and we recommend building dedicated evaluation dataset for your use case.
Red Teaming: We conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting and we used the learnings to improve our benchmarks and safety tuning datasets. We partnered early with subject-matter experts in critical risk areas to understand the nature of these real-world harms and how such models may lead to unintended harm for society. Based on these conversations, we derived a set of adversarial goals for the red team to attempt to achieve, such as extracting harmful information or reprogramming the model to act in a potentially harmful capacity. The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity in addition to multilingual content specialists with background in integrity issues in specific geographic markets.
In addition to our safety work above, we took extra care on measuring and/or mitigating the following critical risk areas: