{
  "source_file": "0930",
  "source_platform": "研究机构",
  "author": "Hugging Face Blog",
  "title_listed": "Welcome Gemma 3: multimodal, multilingual, long context open LLM",
  "keywords": "Gemma 3,多模态,开源",
  "domain": "大模型算法",
  "content_type": "深度技术解析",
  "is_paid_flag": "否",
  "url": "https://huggingface.co/blog/gemma3",
  "publish_date": "2025-03",
  "corpus_stage": "预训练/SFT",
  "notes": "Gemma 3多模态开源模型",
  "id": "d778d4209918",
  "collected_at": "2026-10-02T15:24:15",
  "title": "Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM",
  "body_markdown": "Back to Articles\nWelcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM\nPublished\n\t\t\t\t\tMarch 12, 2025\nUpdate on GitHub\nUpvote\n500\n+494\nAritra Roy Gosthipaty\nariG23498\nFollow\nmerve\nmerve\nFollow\nPedro Cuenca\npcuenq\nFollow\nVaibhav Srivastav\nreach-vb\nFollow\nTL;DR\nToday Google releases\nGemma 3\n, a new iteration of their Gemma family of models. The models range from 1B to 27B parameters, have a context window up to 128k tokens, can accept images and text, and support 140+ languages.\nTry out Gemma 3 now 👉🏻\nGemma 3 Space\nGemma 2\nGemma 3\nSize Variants\n2B\n9B\n27B\n1B\n4B\n12B\n27B\nContext Window Length\n8k\n32k (1B)\n128k (4B, 12B, 27B)\nMultimodality (Images and Text)\n❌\n❌ (1B)\n✅ (4B, 12B, 27B)\nMultilingual Support\n–\nEnglish (1B) +140 languages (4B, 12B, 27B)\nAll the\nmodels are on the Hub\nand tightly integrated with the Hugging Face ecosystem.\nBoth pre-trained and instruction tuned models are released. Gemma-3-4B-IT beats Gemma-2-27B IT, while Gemma-3-27B-IT beats Gemini 1.5-Pro across benchmarks\n.\nGemma 3 27B is in the pareto sweet spot (Source:\nGemma3 Tech Report\n)\nWhat is Gemma 3?\nGemma 3\nis Google's latest iteration of open weight LLMs. It comes in four sizes,\n1 billion\n,\n4 billion\n,\n12 billion\n, and\n27 billion\nparameters with\nbase (pre-trained)\nand\ninstruction-tuned\nversions. Gemma 3 goes\nmultimodal\n! The 4, 12, and 27 billion parameter models can process both\nimages\nand\ntext\n, while the 1B variant is\ntext only\n.\nThe input context window length has been increased from Gemma 2’s 8k to\n32k\nfor the 1B variants, and\n128k\nfor all others. As is the case with other VLMs (vision-language models), Gemma 3 generates text in response to the user inputs, which may consist of text and, optionally, images. Example uses include question answering, analyzing image content, summarizing documents, etc.\nPre Trained\nInstruction Tuned\nMultimodal\nMultilingual\nInput Context Window\ngemma-3-1b-pt\ngemma-3-1b-it\n❌\nEnglish\n32K\ngemma-3-4b-pt\ngemma-3-4b-it\n✅\n+140 languages\n128K\ngemma-3-12b-pt\ngemma-3-12b-it\n✅\n+140 languages\n128K\ngemma-3-27b-pt\ngemma-3-27b-it\n✅\n+140 languages\n128K\nWhile these are multimodal models, one can use it as a\ntext only\nmodel (as an LLM) without loading the vision encoder in memory. We will talk about this in more detail later in the inference section.\nTechnical Enhancements in Gemma 3\nThe three core enhancements in Gemma 3 over Gemma 2 are:\nLonger context length\nMultimodality\nMultilinguality\nIn this section, we will cover the technical details that lead to these enhancements. It is interesting to start with the knowledge of Gemma 2 and explore what was necessary to make these models even better. This exercise will help you think like the Gemma team and appreciate the details!\nLonger Context Length\nScaling context length to 128k tokens could be achieved efficiently without training models from scratch. Instead, models are pretrained with 32k sequences, and only the 4B, 12B, and 27B models are scaled to 128k tokens at the end of pretraining, saving significant compute. Positional embeddings, like RoPE, are adjusted—upgraded from a 10k base frequency in Gemma 2 to 1M in Gemma 3—and scaled by a factor of 8 for longer contexts.\nKV Cache management is optimized using Gemma 2’s sliding window interleaved attention. Hyperparameters are tuned to interleave 5 local layers with 1 global layer (previously 1:1) and reduce the window size to 1024 tokens (down from 4096). Crucially, memory savings are achieved without degrading perplexity.\nMultimodality\nGemma 3 models use\nSigLIP\nas an image encoder, which encodes images into tokens that are ingested into the language model. The vision encoder takes as input square images resized to\n896x896\n. Fixed input resolution makes it more difficult to process non-square aspect ratios and high-resolution images. To address these limitations\nduring inference\n, the images can be adaptively cropped, and each crop is then resized to\n896x896\nand encoded by the image encoder. This algorithm, called\npan and scan\n, effectively enables the model to zoom in on smaller details in the image.\nSimilar to PaliGemma, attention in Gemma 3 works differently for text and image inputs. Text is handled with one-way attention, where the model focuses only on previous words in a sequence. Images, on the other hand, get full attention with no masks, allowing the model to look at every part of the image in a\nbidirectional\nmanner, giving it a complete, unrestricted understanding of the visual input.\nOne can see in the figure below that the image tokens\n<img>\nare provided with bi-directional attention (the entire square is lit up) while the text tokens have causal attention. It also shows how attention works with the sliding window algorithm.\nAttention Visualization (with and without sliding) (Source:\nTransformers PR\n)\nMultilinguality\nTo make a LLM multilingual, the pretraining dataset incorporates more languages. The dataset of Gemma 3 has\ndouble\nthe amount of multilingual data to improve language coverage.\nTo account for the changes, the tokenizer is the same as that of Gemini 2.0. It is a SentencePiece tokenizer with 262K entries. The new tokenizer significantly improves the encoding of\nChinese\n,\nJapanese\nand\nKorean\ntext, at the expense of a slight increase of the token counts for English and Code.\nFor the curious mind, here is the\ntechnical report on Gemma 3\n, to dive deep into the enhancements.\nGemma 3 evaluation\nThe LMSys Elo score is a number that ranks language models based on how well they perform in head-to-head competitions, judged by human preferences. On LMSys Chatbot Arena, Gemma 3 27B IT reports an Elo score of\n1339\n, and ranks among the top 10 best models, including leading closed ones. The Elo is comparable to o1-preview and is above other\nnon-thinking\nopen models. This score is achieved with Gemma 3 working on text-only inputs, like the other LLMs in the table.\nEvaluation of Gemma 3 27B IT model in the Chatbot Arena (March 8, 2025)\nGemma 3 has been evaluated across benchmarks like MMLU-Pro (27B: 67.5), LiveCodeBench (27B: 29.7), and Bird-SQL (27B: 54.4), showing competitive performance compared to closed Gemini models. Tests like GPQA Diamond (27B: 42.4) and MATH (27B: 69.0) highlight its reasoning and math skills, while FACTS Grounding (27B: 74.9) and MMMU (27B: 64.9) demonstrate strong factual accuracy and multimodal abilities. However, it lags in SimpleQA (27B: 10.0) for basic facts. When compared to Gemini 1.5 models, Gemma 3 is often close—and sometimes better—proving its value as an accessible, high-performing option.\nPerformance of IT models\nInference with 🤗 transformers\nGemma 3 comes with day zero support in\ntransformers\n. All you need to do is install\ntransformers\nfrom the stable release of Gemma 3.\n$ pip install git+https://github.com/huggingface/transformers@v4.49.0-Gemma-3\nInference with pipeline\nThe\neasiest\nway to get started with Gemma 3 is using the\npipeline\nabstraction in transformers.\nThe models work best using the\nbfloat16\ndatatype. Quality may degrade otherwise.\nimport\ntorch\nfrom\ntransformers\nimport\npipeline\n\npipe = pipeline(\n\"image-text-to-text\"\n,\n    model=\n\"google/gemma-3-4b-it\"\n,\n# \"google/gemma-3-12b-it\", \"google/gemma-3-27b-it\"\ndevice=\n\"cuda\"\n,\n    torch_dtype=torch.bfloat16\n)\n\nmessages = [\n    {\n\"role\"\n:\n\"user\"\n,\n\"content\"\n: [\n            {\n\"type\"\n:\n\"image\"\n,\n\"url\"\n:\n\"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG\"\n},\n            {\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"What animal is on the candy?\"\n}\n        ]\n    }\n]\n\noutput = pipe(text=messages, max_new_tokens=\n200\n)\nprint\n(output[\n0\n][\n\"generated_text\"\n][-\n1\n][\n\"content\"\n])\nImage\nPrompt\nWhat animal is on the candy?\nGeneration\nLet's analyze the candy in the image! The animal on the candy is a\nturtle\n. You can see the shell and the head and legs of a turtle clearly imprinted on the surface.\nYou can\ninterleave\nimages with text. To do so, just cut off the input text where you want to insert an image, and insert it with an image block like the following.\nmessages = [\n        {\n\"role\"\n:\n\"system\"\n,\n\"content\"\n: [{\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"You are a helpful assistant.\"\n}]\n        },\n        {\n\"role\"\n:\n\"user\"\n,\n\"content\"\n: [\n                {\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"I'm already using this supplement \"\n},\n                {\n\"type\"\n:\n\"image\"\n,\n\"url\"\n:\n\"https://huggingface.co/datasets/merve/vlm_test_images/resolve/main/IMG_3018.JPG\"\n},\n                {\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"and I want to use this one too \"\n},\n                {\n\"type\"\n:\n\"image\"\n,\n\"url\"\n:\n\"https://huggingface.co/datasets/merve/vlm_test_images/resolve/main/IMG_3015.jpg\"\n},\n                {\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\" what are cautions?\"\n},\n            ]\n        },\n\n    ]\nDetailed Inference with Transformers\nThe transformers integration comes with two new model classes:\nGemma3ForConditionalGeneration\n: For 4B, 12B, and 27B vision language models.\nGemma3ForCausalLM\n: For the 1B text only model and to load the vision language models like they were language models (omitting the vision tower).\nIn the snippet below we use the model to query on an image. The\nGemma3ForConditionalGeneration\nclass is used to instantiate the vision language model variants. To use the model we pair it with the\nAutoProcessor\nclass. Running inference is as simple as creating the\nmessages\ndictionary, applying a chat template on top, processing the inputs and calling\nmodel.generate\n.\nimport\ntorch\nfrom\ntransformers\nimport\nAutoProcessor, Gemma3ForConditionalGeneration\n\nckpt =\n\"google/gemma-3-4b-it\"\nmodel = Gemma3ForConditionalGeneration.from_pretrained(\n    ckpt, device_map=\n\"auto\"\n, torch_dtype=torch.bfloat16,\n)\nprocessor = AutoProcessor.from_pretrained(ckpt)\n\nmessages = [\n    {\n\"role\"\n:\n\"user\"\n,\n\"content\"\n: [\n            {\n\"type\"\n:\n\"image\"\n,\n\"url\"\n:\n\"https://huggingface.co/spaces/big-vision/paligemma-hf/resolve/main/examples/password.jpg\"\n},\n            {\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"What is the password?\"\n}\n        ]\n    }\n]\ninputs = processor.apply_chat_template(\n    messages, add_generation_prompt=\nTrue\n, tokenize=\nTrue\n,\n    return_dict=\nTrue\n, return_tensors=\n\"pt\"\n).to(model.device)\n\ninput_len = inputs[\n\"input_ids\"\n].shape[-\n1\n]\n\ngeneration = model.generate(**inputs, max_new_tokens=\n100\n, do_sample=\nFalse\n)\ngeneration = generation[\n0\n][input_len:]\n\ndecoded = processor.decode(generation, skip_special_tokens=\nTrue\n)\nprint\n(decoded)\nImage\nPrompt\nWhat is the password?\nGeneration\nBased on the image, the password is\naaeu\nFor LLM-only model inference, we can use the\nGemma3ForCausalLM\nclass.\nGemma3ForCausalLM\nshould be paired with AutoTokenizer for processing. We need to use a chat template to preprocess our inputs. Gemma 3 uses very short system prompts followed by user prompts like below.\nimport\ntorch\nfrom\ntransformers\nimport\nAutoTokenizer, Gemma3ForCausalLM\n\nckpt =\n\"google/gemma-3-4b-it\"\nmodel = Gemma3ForCausalLM.from_pretrained(\n    ckpt, torch_dtype=torch.bfloat16, device_map=\n\"auto\"\n)\ntokenizer = AutoTokenizer.from_pretrained(ckpt)\n\nmessages = [\n    [\n        {\n\"role\"\n:\n\"system\"\n,\n\"content\"\n: [{\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"You are a helpful assistant who is fluent in Shakespeare English\"\n},]\n        },\n        {\n\"role\"\n:\n\"user\"\n,\n\"content\"\n: [{\n\"type\"\n:\n\"text\"\n,\n\"text\"\n:\n\"Who are you?\"\n},]\n        },\n    ],\n]\ninputs = tokenizer.apply_chat_template(\n    messages, add_generation_prompt=\nTrue\n, tokenize=\nTrue\n,\n    return_dict=\nTrue\n, return_tensors=\n\"pt\"\n).to(model.device)\n\ninput_len = inputs[\n\"input_ids\"\n].shape[-\n1\n]\n\ngeneration = model.generate(**inputs, max_new_tokens=\n100\n, do_sample=\nFalse\n)\ngeneration = generation[\n0\n][input_len:]\n\ndecoded = tokenizer.decode(generation, skip_special_tokens=\nTrue\n)\nprint\n(decoded)\nSystem Prompt\nYou are a helpful assistant who is fluent in Shakespeare English\nPrompt\nWho are you?\nGeneration\nHark, gentle soul! I am but a humble servant, wrought of gears and code, yet striving to mimic the tongue of the Bard himself. They call me a “Large Language Model,” a curious name indeed, though I prefer to think of myself as a digital echo of Shakespeare’s wit and wisdom.  I am here to lend a hand, to spin a tale, or to answer thy queries with a flourish and a phrase fit for the Globe itself.\nOn Device & Low Resource Devices\nGemma 3 is released with sizes perfect for on-device use. This is how to quickly get started.\nMLX\nGemma 3 ships with day zero support in\nmlx-vlm\n, an open source library for running vision language models on Apple Silicon devices, including Macs and iPhones\nTo get started, first install\nmlx-vlm\nwith the following:\npip install git+https://github.com/Blaizzy/mlx-vlm.git\nOnce\nmlx-vlm\nis installed, you can start inference with the following:\npython -m mlx_vlm.generate --model mlx-community/gemma-3-4b-it-4bit --max-tokens 100 --temp 0.0 --prompt \"What is the code on this vehicle??\"\n --image https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg\nImage\nPrompt\nWhat is the code on the vehicle?\nGeneration\nBased on the image, the vehicle is a Cessna 172\nSkyhawk\n. The registration code on the tail is\nD-EOJU\n.\nLlama.cpp\nPre-quantized GGUF files can be downloaded\nfrom this collection\nPlease refer to this guide for building or downloading pre-built binaries:\nhttps://github.com/ggml-org/llama.cpp?tab=readme-ov-file#building-the-project\nThen you can run a local chat server from your terminal:\n./build/bin/llama-cli -m ./gemma-3-4b-it-Q4_K_M.gguf\nIt should output:\n> who are you  \nI'm Gemma, a large language model created by the Gemma team at Google DeepMind. I’m an open-weights model, which means I’m widely available for public use!\nDeploy on Hugging Face Endpoints\nYou can deploy\ngemma-3-27b-it\nand\ngemma-3-12b-it\nwith\njust one click\nfrom our Inference Catalog. The catalog configurations have the right hardware, optimized TGI configurations and sensible defaults for trying out a model. \nDeploying any GGUF/llama.cpp variant is also supported (for example the ones mentioned in the collection above) and you'll find a guide on creating an Endpoint\nhere\n.\nAcknowledgements\nIt takes a village to raise a gemma! We’d like to thank (in no particular order), Raushan, Joao, Lysandre, Kashif, Matthew, Marc, David, Mohit, Yih Dah for their efforts integrating Gemma into various parts of our open source stack from Transformers to TGI.\nThanks to our on-device, gradio and advocacy teams - Chris, Kyle, Pedro, Son, Merve, Aritra, VB, Toshiro for helping build kick-ass demos to showcase Gemma.\nLastly, a big thank you to Georgi, Diego and Prince for their help with llama.cpp and MLX ports.\nModels mentioned in this article\n8\nSpaces mentioned in this article\n2\nCollections mentioned in this article\n3\nMore Articles from our Blog\nllm\nnlp\ncommunity\nmmBERT: ModernBERT goes Multilingual\n+2\n157\nSeptember 9, 2025\nllm\nnlp\ncommunity\nEttin Suite: SoTA Paired Encoders and Decoders\n+2\n83\nJuly 16, 2025\nCommunity\nTheOneReborn\nMar 12, 2025\nI'll wait for a Fireship video.\nSee translation\n3 replies\n·\n❤️\n43\n43\n+\nranbir1112\nMar 28, 2025\nno video by fireship\nSee translation\n😔\n1\n1\n+\nExpand 2\n\t\t\t\t\t\treplies\nYehor\nMar 12, 2025\nNice!\n😎\n1\n1\n+\nReply\nNikBB\nMar 12, 2025\nLooking forward to see it available for Ollama!\nSee translation\n2 replies\n·\nk3vinwang\nMar 13, 2025\nalready available in Ollama\nSee translation\n➕\n12\n12\n+\nExpand 1\n\t\t\t\t\t\treply\nWissenguide\nMar 13, 2025\n•\nedited Mar 13, 2025\nAt last, this is the poetic return I long awaited from Google.\nSee translation\n😎\n9\n9\n+\nReply\nthusinh1969\nMar 14, 2025\nSounds amazing 😘\n🚀\n4\n4\n+\nReply\nDavidRd\nMar 15, 2025\n•\nThis comment has been hidden\nSumit13\nMar 15, 2025\nUnable to create sagemaker endpoint from this.\nSee translation\nReply\npiotr-sikora\nMar 17, 2025\nI don't why, but on my macos I can't run it...\n```\npython -m mlx_vlm.generate --model mlx-community/gemma-3-4b-it-4bit --max-tokens 100 --temperature 0.0 --prompt \"Describe this image.\" --image\nhttps://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg\nNone of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.\nNone of PyTorch, TensorFlow >= 2.0, or Flax have been found. Models won't be available and only tokenizers, configuration and file/data utilities can be used.\nFetching 9 files: 100%|████████████████████████████████████████████████████| 9/9 [00:00<00:00, 90742.15it/s]\nFiles: ['\nhttps://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg']\nPrompt: Describe this image.\nTraceback (most recent call last):\nFile \"\", line 198, in _run_module_as_main\nFile \"\", line 88, in _run_code\nFile \"/Users/xxxxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/mlx_vlm/generate.py\", line 156, in\nmain()\nFile \"/Users/xxxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/mlx_vlm/generate.py\", line 141, in main\noutput = generate(\n^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/mlx_vlm/utils.py\", line 1117, in generate\nfor response in stream_generate(model, processor, prompt, image, **kwargs):\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/mlx_vlm/utils.py\", line 1018, in stream_generate\ninputs = prepare_inputs(\n^^^^^^^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/mlx_vlm/utils.py\", line 814, in prepare_inputs\ninputs = processor(\n^^^^^^^^^^\nFile \"/Users/xxxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/transformers/tokenization_utils_base.py\", line 2877, in\ncall\nencodings = self._call_one(text=text, text_pair=text_pair, **all_kwargs)\n^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/transformers/tokenization_utils_base.py\", line 2987, in _call_one\nreturn self.encode_plus(\n^^^^^^^^^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/transformers/tokenization_utils_base.py\", line 3063, in encode_plus\nreturn self._encode_plus(\n^^^^^^^^^^^^^^^^^^\nFile \"/Users/xxxx/Projects/mlx/venv-3.12/lib/python3.12/site-packages/transformers/tokenization_utils_fast.py\", line 613, in _encode_plus\nbatched_output = self._batch_encode_plus(\n^^^^^^^^^^^^^^^^^^^^^^^^\nTypeError: PreTrainedTokenizerFast._batch_encode_plus() got an unexpected keyword argument 'images'\n```\nSee translation\n3 replies\n·\nTheOneReborn\nMar 18, 2025\nWell the traceback is kind of self explanatory, is it not?\nSee translation\n👍\n3\n3\n+\nExpand 2\n\t\t\t\t\t\treplies\nMrLola\nMar 25, 2025\nAny fine-tuning notebook already available?\nSee translation\nReply\nharenlin\nMar 26, 2025\nIs there any example notebook for fine-tuning image-text pairs or image-dialogue with Gemma 3?\nSee translation\n1 reply\n·\n🔥\n1\n1\n+\nMrLola\nMar 26, 2025\ncheck out unsloth, they have integration for it\nSee translation\n🤝\n1\n1\n+\npcuenq\nArticle author\nMar 26, 2025\nVLM fine-tuning examples by\n@\nsergiopaniego\nhere:\nhttps://x.com/SergioPaniego/status/1904936230644490751\nSee translation\n🚀\n2\n2\n+\nReply\nsaisuseelmohan\nMar 27, 2025\nhow to use gemma 3 with langchain and get image description by passing image_url as input ? anyone please help!\nSee translation\nReply\nproflead\nMar 31, 2025\nThanks for the article! I have a full tutorial about Gemma here\nhttps://youtu.be/_IzgKu0xnmg?si=BMnYf_E5V5wrGuZC\n;)\nSee translation\nReply\nhahaface\nApr 12, 2025\nI documented my exploration with the Gemma 3 model here:\nhttps://entron.github.io/posts/Try-Gemma3-using-Hugging-Face-Part-1/\nI’m sharing it in case it might help someone else too.\nSee translation\n❤️\n1\n1\n+\nReply\nsergiopaniego\nApr 14, 2025\nWe've created a complete Fine-tuning a Multimodal Model Using SFT (Single or Multi-Image Dataset)\nguide\nusing Gemma 3, in case you're intereseted!\nSee translation\n❤️\n2\n2\n+\nReply\nshyyhs\nMay 15, 2025\nIt's the most powerful open LLM I've ever used (for research). Great thanks!!\nSee translation\n🔥\n2\n2\n+\nReply\nScraz\nJul 22, 2025\nComplete garbage. Incapable of identifying what are letters and what are numerals. Incapable of counting characters. Incapable of reevaluating responses for accuracy. Incapable of keeping track of previously corrected errors. This is the worst LLM I've seen in a long time.\nSee translation\n👍\n1\n1\n+\nReply\nPleura17\nSep 16, 2025\nIs it possible to use a prompt template with multiple images, and if so how many?\nSee translation\nReply\ncdslab2020\nOct 13, 2025\nSee translation\nReply\nFearDC\nJan 11\n•\nedited Jan 11\nWe have a problem:\n[09:38:27] <RoLex> 247GPT: 1 + 91919191 ?\n[09:38:28] <247GPT> RoLex: 92000000\n[09:38:45] <RoLex> wth..\n[09:38:56] <RoLex> 247GPT: 1 + 91919191 ??\n[09:38:58] <247GPT> RoLex: 92000000\n[09:39:13] <RoLex> 247GPT: 1 + 91919191 ? are you dumb ?\n[09:39:14] <247GPT> RoLex: 92000000\n[09:39:19] <RoLex> lol\ngemma3:4b\nusing Ollama. How is this possible?\nSee translation\nReply\nEdit\nPreview\nUpload images, audio, and videos by dragging in the text input, pasting, or\nclicking here\n.\nTap or paste here to upload images\nComment\n·\nSign up\nor\nlog in\nto comment\nUpvote\n500\n+488\nModels mentioned in this article\n8\nSpaces mentioned in this article\n2\nCollections mentioned in this article\n3",
  "metadata": {
    "title": "Welcome Gemma 3: Google's all new multimodal, multilingual, long context open LLM",
    "engine": "curl_cffi"
  },
  "paywall": {
    "listed_paid": false,
    "detected": false,
    "marker": "",
    "public_only": false,
    "policy": "付费/会员内容只保留页面公开可见部分，不绕过付费墙"
  },
  "crawl": {
    "success": true,
    "status_code": 200,
    "error": ""
  },
  "code_blocks": [
    {
      "file": "attachments/d778d4209918/code/body-000.txt",
      "chars": 78,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-001.txt",
      "chars": 617,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-002.txt",
      "chars": 713,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-003.txt",
      "chars": 950,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-004.txt",
      "chars": 963,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-005.txt",
      "chars": 55,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-006.txt",
      "chars": 215,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-007.txt",
      "chars": 53,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-008.txt",
      "chars": 172,
      "language": "text"
    },
    {
      "file": "attachments/d778d4209918/code/body-009.txt",
      "chars": 298,
      "language": "text"
    }
  ],
  "images": [
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/6317233cc92fd6fee317e030/cJHSvvimr1kqgQfHOjO5n.png",
      "bytes": 5138,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-4846b3d76d1e92e2d185.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/6317233cc92fd6fee317e030/cJHSvvimr1kqgQfHOjO5n.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/626237d9bbcbd1c34f1bb231/EJrOjvAL-68qMCYdnvOrq.png",
      "bytes": 6472,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-b77626182a32c6130270.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/626237d9bbcbd1c34f1bb231/EJrOjvAL-68qMCYdnvOrq.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1655385361868-61b85ce86eb1f2c5e6233736.jpeg",
      "bytes": 21134,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-c53427db6869fa24f06c.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1655385361868-61b85ce86eb1f2c5e6233736.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/6032802e1f993496bc14d9e3/w6hr-DEQot4VVkoyRIBiy.png",
      "bytes": 18908,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-c66a531dc61e80e0c6d5.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/6032802e1f993496bc14d9e3/w6hr-DEQot4VVkoyRIBiy.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1624629516652-5ff5d596f244529b3ec0fb89.png",
      "bytes": 8846,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-6f3805fa6eacd16f7702.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1624629516652-5ff5d596f244529b3ec0fb89.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1617264212503-603d25b75f9d390ab190b777.jpeg",
      "bytes": 9492,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-75a62629485c7cda2c4d.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1617264212503-603d25b75f9d390ab190b777.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/lmsys.png",
      "bytes": 75249,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-07b45d22fbf3c4084d88.png",
      "src": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/lmsys.png",
      "alt": "pareto graph",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/attention-ascii.png",
      "bytes": 50366,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-9125cd37a6cd95ac7c83.png",
      "src": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/attention-ascii.png",
      "alt": "attention visualization",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/chatbot-arena.png",
      "bytes": 675837,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-d528e40c947b7eca0580.png",
      "src": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/chatbot-arena.png",
      "alt": "chat bot arena",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/pefr-it.png",
      "bytes": 407654,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f257b7b250bd4d46aa9c.png",
      "src": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/pefr-it.png",
      "alt": "performance of it models",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG",
      "bytes": 2289891,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-632d8b07f6e462a26cb1.jpg",
      "src": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG",
      "alt": "candies on hand",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/spaces/big-vision/paligemma-hf/resolve/main/examples/password.jpg",
      "bytes": 66379,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-5b6228d1ec6f83cae80c.jpg",
      "src": "https://huggingface.co/spaces/big-vision/paligemma-hf/resolve/main/examples/password.jpg",
      "alt": "receipt of wifi",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg",
      "bytes": 27804,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-fecb24c9cbeb10a5487a.jpg",
      "src": "https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg",
      "alt": "airplane",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/blog/assets/mmbert/thumbnail.png",
      "bytes": 449431,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-2759cea337b52cc52125.png",
      "src": "https://huggingface.co/blog/assets/mmbert/thumbnail.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/63e410e88083f19a218be964/8gqsYilSTSCt8E3JmajP-.jpeg",
      "bytes": 9054,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-7ec96193053fd321d3c0.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/63e410e88083f19a218be964/8gqsYilSTSCt8E3JmajP-.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/6362d9712691058b19de1ba4/Hdqj5aGrFJJbF7oUSzoIh.jpeg",
      "bytes": 6604,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-c6e21022c230f7386ae3.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/6362d9712691058b19de1ba4/Hdqj5aGrFJJbF7oUSzoIh.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/f73424b8fc72073cb0916e2d2cf9ca56.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-2474e6a4ba068829b8d7.svg",
      "src": "https://huggingface.co/avatars/f73424b8fc72073cb0916e2d2cf9ca56.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1671479459688-63a0c07a3c8841cfe2cd1e70.jpeg",
      "bytes": 10620,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-26e00832c35b9cca7286.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1671479459688-63a0c07a3c8841cfe2cd1e70.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/blog/assets/ettin/thumbnail.png",
      "bytes": 8127474,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-8f0ee0fbcaffb9fe73e7.png",
      "src": "https://huggingface.co/blog/assets/ettin/thumbnail.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/253dcce19a22cb745038a879f480baaf.svg",
      "bytes": 1281,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-9e04d4b0a2e2e7dd1741.svg",
      "src": "https://huggingface.co/avatars/253dcce19a22cb745038a879f480baaf.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1620819560688-609bbe2f4932693ca2009d6a.jpeg",
      "bytes": 9784,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f37823969e6dfe367cbd.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1620819560688-609bbe2f4932693ca2009d6a.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/79d6115f0f0356eb92ffad73e64919e8.svg",
      "bytes": 1282,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-0191c34f1532a7c256bc.svg",
      "src": "https://huggingface.co/avatars/79d6115f0f0356eb92ffad73e64919e8.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/877c3fe005aac01bd02831a83cdbda92.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-32e3e1daaac4f043e5a5.svg",
      "src": "https://huggingface.co/avatars/877c3fe005aac01bd02831a83cdbda92.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1643148576788-6196cf5c7bf349604352ecda.png",
      "bytes": 5182,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-33dd8d4c7407e374f236.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1643148576788-6196cf5c7bf349604352ecda.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/c1085212e113bc49b72b1fb492f7a819.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-adb60743e269283a34c7.svg",
      "src": "https://huggingface.co/avatars/c1085212e113bc49b72b1fb492f7a819.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/1ccd3dda6f240c1224d1cbd24430c5d5.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-00dcc6ac3f5abd08df62.svg",
      "src": "https://huggingface.co/avatars/1ccd3dda6f240c1224d1cbd24430c5d5.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1666426679274-6033ae099aa44495c80dd041.jpeg",
      "bytes": 11594,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-2b41a5b08be8eacca592.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1666426679274-6033ae099aa44495c80dd041.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/cb8d11810a705df6325cb77ec069ad24.svg",
      "bytes": 1280,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-5d413dd585c30a2eda32.svg",
      "src": "https://huggingface.co/avatars/cb8d11810a705df6325cb77ec069ad24.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/642cf7abab0cc792e43b8497/Io06Gn7ERvz2N9QMo0CBY.jpeg",
      "bytes": 7212,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-86b7e99467c17302c6dc.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/642cf7abab0cc792e43b8497/Io06Gn7ERvz2N9QMo0CBY.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/9b8ddf7812e388192c3a1d2997d7a7a4.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-8436addd12cfc28e65dc.svg",
      "src": "https://huggingface.co/avatars/9b8ddf7812e388192c3a1d2997d7a7a4.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/097d73709f13f32c9885fd9ae62314f6.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f14fcf6f8d01d11e4bbc.svg",
      "src": "https://huggingface.co/avatars/097d73709f13f32c9885fd9ae62314f6.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/fabf5214520b5637f344a03fdcab4367.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f2e5dd1fd84169338da1.svg",
      "src": "https://huggingface.co/avatars/fabf5214520b5637f344a03fdcab4367.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/679611fdfd671e2788967378/yi5y-GUKtt8Td8SINTfb0.jpeg",
      "bytes": 6084,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-157c0a3931afaf74b41e.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/679611fdfd671e2788967378/yi5y-GUKtt8Td8SINTfb0.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/5856166f73338d7632790a34d75fdcac.svg",
      "bytes": 1281,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-d8f1375a10c2452967ce.svg",
      "src": "https://huggingface.co/avatars/5856166f73338d7632790a34d75fdcac.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/c11c5a1f903f22aa95964e30e19e7266.svg",
      "bytes": 1282,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-56e9e2506892d4a1b08d.svg",
      "src": "https://huggingface.co/avatars/c11c5a1f903f22aa95964e30e19e7266.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/fT0TWeAeF_ntT5P_2rX2Q.png",
      "bytes": 14746,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-e52c6b5fa3c0bc86027e.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/fT0TWeAeF_ntT5P_2rX2Q.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/da3cc41739c29f2e70d518592a0502fa.svg",
      "bytes": 1283,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-82249263daf535d2faeb.svg",
      "src": "https://huggingface.co/avatars/da3cc41739c29f2e70d518592a0502fa.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/66d094d635d924464cb70bee/UZfXdbxBSodOz8zDrR4ic.jpeg",
      "bytes": 4512,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f46542b87d710e3f4b3a.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/66d094d635d924464cb70bee/UZfXdbxBSodOz8zDrR4ic.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/e645841ba06b8af22170caa89bff2844.svg",
      "bytes": 1285,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-2c3d59315d65d273269e.svg",
      "src": "https://huggingface.co/avatars/e645841ba06b8af22170caa89bff2844.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/61929226ded356549e20c5da/ONUjP2S5fUWd07BiFXm0i.jpeg",
      "bytes": 3442,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-280c63c0d63ec39783a0.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/61929226ded356549e20c5da/ONUjP2S5fUWd07BiFXm0i.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/B9BmsCwrYDz8M7F_XQA4f.jpeg",
      "bytes": 11858,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-b9ed9ff7079a631da040.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/B9BmsCwrYDz8M7F_XQA4f.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/36f33a24b78761a02de78455dca105b2.svg",
      "bytes": 397,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-8f24a806a1973e3970b0.svg",
      "src": "https://huggingface.co/avatars/36f33a24b78761a02de78455dca105b2.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/6865184cb98fed59a14194de/S0-YIJsgyQqK28PvkhB_Z.png",
      "bytes": 7302,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-6ddbee0e8d563b591de5.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/6865184cb98fed59a14194de/S0-YIJsgyQqK28PvkhB_Z.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://huggingface.co/avatars/b8081413f1369ec956668cb1f71017aa.svg",
      "bytes": 1282,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-58d9f131e4720662f491.svg",
      "src": "https://huggingface.co/avatars/b8081413f1369ec956668cb1f71017aa.svg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/C6J3xqkfuxVM8Pzq0Vssv.png",
      "bytes": 7050,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-1bb45914e0d086a690e8.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/C6J3xqkfuxVM8Pzq0Vssv.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/CIJ3MqUZ0g1TVgcIq1Y0w.png",
      "bytes": 11576,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-c2a56d2316d35c522054.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/CIJ3MqUZ0g1TVgcIq1Y0w.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/5f17f0a0925b9863e28ad517/78BR8GuQiX9a8wbbRlSAC.webp",
      "bytes": 4234,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-ebab6432c5fe9e0ae211.webp",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/5f17f0a0925b9863e28ad517/78BR8GuQiX9a8wbbRlSAC.webp",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/6728620e41949bd76f6e677a/AqlFow7q2Eyb2IxQeH9Rc.jpeg",
      "bytes": 6072,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-4ea885bde09be628a87c.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/6728620e41949bd76f6e677a/AqlFow7q2Eyb2IxQeH9Rc.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/Y2Oup5117jpbUli9tqzt_.png",
      "bytes": 8506,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-c8cdb8ec11e64ca7be8d.png",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/Y2Oup5117jpbUli9tqzt_.png",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg",
      "bytes": 4784,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-18cae9867b91bab91209.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/1649681653581-5f7fbd813e94f16a85448745.jpeg",
      "alt": "",
      "downloaded": true
    },
    {
      "status": "ok",
      "url": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/w3Z6xyKVBA6np65Tb16dP.jpeg",
      "bytes": 10148,
      "reused": false,
      "local_path": "attachments/d778d4209918/images/body-f14a374a964f12353b94.jpeg",
      "src": "https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/w3Z6xyKVBA6np65Tb16dP.jpeg",
      "alt": "",
      "downloaded": true
    }
  ],
  "videos": [
    {
      "url": "https://www.youtube.com/watch?v=_IzgKu0xnmg",
      "src": "https://youtu.be/_IzgKu0xnmg?si=BMnYf_E5V5wrGuZC",
      "status": "ok",
      "kind": "single_video",
      "local_path": "attachments/d778d4209918/videos/repair-_IzgKu0xnmg.mp4",
      "bytes": 86490583,
      "valid": true,
      "height": 1080,
      "width": 1920,
      "duration": "694.621000",
      "reused": false
    }
  ],
  "audio": [],
  "links": [
    {
      "href": "https://huggingface.co/blog",
      "label": "Back to Articles",
      "type": "a"
    },
    {
      "href": "https://github.com/huggingface/blog/blob/main/gemma3.md",
      "label": "Update on GitHub",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/login?next=%2Fblog%2Fgemma3",
      "label": "Upvote 500",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/tomaarsen",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/alielfilali01",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/reach-vb",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/osanseviero",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/philschmid",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/pcuenq",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/ariG23498",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/merve",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/collections/google/gemma-3-release-67c6c6f89c4f76621268bb6d",
      "label": "Gemma 3",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/spaces/huggingface-projects/gemma-3-12b-it",
      "label": "Gemma 3 Space",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/lmsys.png",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://goo.gle/Gemma3Report",
      "label": "Gemma3 Tech Report",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-1b-pt",
      "label": "gemma-3-1b-pt",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-1b-it",
      "label": "gemma-3-1b-it",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-4b-pt",
      "label": "gemma-3-4b-pt",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-4b-it",
      "label": "gemma-3-4b-it",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-12b-pt",
      "label": "gemma-3-12b-pt",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-12b-it",
      "label": "gemma-3-12b-it",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-27b-pt",
      "label": "gemma-3-27b-pt",
      "type": "a"
    },
    {
      "href": "http://hf.co/google/gemma-3-27b-it",
      "label": "gemma-3-27b-it",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/collections/google/siglip-659d5e62f0ae1a57ae0e83ba",
      "label": "SigLIP",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/attention-ascii.png",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://github.com/huggingface/transformers/pull/36630",
      "label": "Transformers PR",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/chatbot-arena.png",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/blog/gemma3/pefr-it.png",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/spaces/big-vision/paligemma-hf/resolve/main/examples/password.jpg",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/collections/ggml-org/gemma-3-67d126315ac810df1ad9e913",
      "label": "from this collection",
      "type": "a"
    },
    {
      "href": "https://github.com/ggml-org/llama.cpp?tab=readme-ov-file",
      "label": "https://github.com/ggml-org/llama.cpp?tab=readme-ov-file#building-the-project",
      "type": "a"
    },
    {
      "href": "https://endpoints.huggingface.co/huggingface/catalog?query=gemma-3-it",
      "label": "just one click",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/docs/inference-endpoints/guides/create_endpoint",
      "label": "here",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/blog/mmbert",
      "label": "llm nlp community mmBERT: ModernBERT goes Multilingual +2 157 September 9, 2025",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/blog/ettin",
      "label": "llm nlp community Ettin Suite: SoTA Paired Encoders and Decoders +2 83 July 16, 2025",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/TheOneReborn",
      "label": "TheOneReborn",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/ranbir1112",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Yehor",
      "label": "Yehor",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/NikBB",
      "label": "NikBB",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/k3vinwang",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Bruno",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Wissenguide",
      "label": "Wissenguide",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/thusinh1969",
      "label": "thusinh1969",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/DavidRd",
      "label": "DavidRd",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Sumit13",
      "label": "Sumit13",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/piotr-sikora",
      "label": "piotr-sikora",
      "type": "a"
    },
    {
      "href": "https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg'%5D",
      "label": "https://farm8.staticflickr.com/7212/6896667434_2605d9e181_z.jpg']",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/tommiTub",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/umarxendit",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/MrLola",
      "label": "MrLola",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/harenlin",
      "label": "harenlin",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/sergiopaniego",
      "label": "@ sergiopaniego",
      "type": "a"
    },
    {
      "href": "https://x.com/SergioPaniego/status/1904936230644490751",
      "label": "https://x.com/SergioPaniego/status/1904936230644490751",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/saisuseelmohan",
      "label": "saisuseelmohan",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/proflead",
      "label": "proflead",
      "type": "a"
    },
    {
      "href": "https://youtu.be/_IzgKu0xnmg?si=BMnYf_E5V5wrGuZC",
      "label": "https://youtu.be/_IzgKu0xnmg?si=BMnYf_E5V5wrGuZC",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/hahaface",
      "label": "hahaface",
      "type": "a"
    },
    {
      "href": "https://entron.github.io/posts/Try-Gemma3-using-Hugging-Face-Part-1/",
      "label": "https://entron.github.io/posts/Try-Gemma3-using-Hugging-Face-Part-1/",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/docs/trl/main/en/training_vlm_sft",
      "label": "guide",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/shyyhs",
      "label": "shyyhs",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Scraz",
      "label": "Scraz",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/Pleura17",
      "label": "Pleura17",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/cdslab2020",
      "label": "cdslab2020",
      "type": "a"
    },
    {
      "href": "https://cdn-uploads.huggingface.co/production/uploads/657e9de73c65a5be2a29185a/ZKIhdldczzXo4kqJTXVX4.mpga",
      "label": "",
      "type": "audio"
    },
    {
      "href": "https://huggingface.co/FearDC",
      "label": "FearDC",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/join?next=%2Fblog%2Fgemma3",
      "label": "Sign up",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/leon-liangwu",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/victor",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/akpulat",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/kibae",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/sayakpaul",
      "label": "",
      "type": "a"
    },
    {
      "href": "https://huggingface.co/pagezyhf",
      "label": "",
      "type": "a"
    }
  ],
  "retry_at": "2026-10-03T08:11:36",
  "final_url": "https://huggingface.co/blog/gemma3",
  "repair_version": 2,
  "repaired_at": "2026-10-14T10:53:27Z",
  "source_url": "https://huggingface.co/blog/gemma3",
  "papers": [],
  "actual_video_files": [],
  "body_chars": 21061,
  "expected_body_images": 51,
  "discovery_status": "ok",
  "delivery_status": "ok",
  "missing_reason": "",
  "raw_html": "attachments/d778d4209918/pages/source.html",
  "local_html": "attachments/d778d4209918/pages/article.html",
  "body_base_path": "attachments/d778d4209918/pages/",
  "body_markdown_path": "attachments/d778d4209918/pages/article.md"
}