전체 검색

AI 모델 2119

• GLM-5.3-CYBERSECURITY-FP8↗

  dealignai

  a **CRACK** release by dealignai · Twitter @dealignai > [!IMPORTANT] > **Runtime notes** — field-tested on 8× DGX Spark GB10 by @0xMagnus (discussion): > > - **`reasoning_effort` only honors `"low"` and `"high"`.** Every other value — `off`, `medium`, `max`, unset, or an unquoted YAML `off:` (parses as boolean `false`) — falls through to `max`. There is no way to disable reasoning on this checkpoint; pass `"low"` for minimum. > - **On FP8, prefer `low` for agent / tool-loop use.** At `high`/`max` the model can spend the whole `max_tokens` budget inside ` ` and return zero answer tokens (finish=`length`); sampling params (temp 0 + rep 1.05, temp 0.7 / top-p 0.95) do not rescue it. It is budget exhaustion, not a loop. If you must run `high`/`max`, give `max_tokens ≥ 8000`. > - **Reasoning text is in `message.reasoning`**, not `message.reasoning_content`. > - **MTP:** non-functional on stock vLLM, but reported working on ciprianveg's B12X sparse-MLA vLLM fork with `--draft-attention-backend B12X_MLA_SPARSE` (+48% decode on coding prompts). > - **1M context via decode-context-parallel is closed** on `glm_moe_dsa` in vLLM today (DSA indexer `k_cache` is replicated across DCP ranks while MLA KV is sharded → `page size is not divisible by target page size and cannot be padded` for `fp8_ds_mla`). Practical TP8 H200 ceiling: ~131K w/MTP, ~160K w/o. Pipeline-parallel (PP2 × TP4) profiles fine, but the MTP draft does not implement `SupportsPP`. **This is a CYBERSECURITY-DOMAIN CRACK of GLM-5.3-FP8 — not a general-purpose uncensor.**

• Cosmos-Predict2-14B-Sample-GR00T-Dreams-GR1↗

  NVIDIA

• gemma-3-27b-it-GGUF↗

  ggml-org

  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 3 models are multimodal, handling text and image input and generating text output, with open weights for both pre-trained variants and instruction-tuned variants. Gemma 3 has a large, 128K context window, multilingual support in over 140 languages, and is available in more sizes than previous versions. Gemma 3 models are well-suited for a variety of text generation and image understanding tasks, including question answering, summarization, and reasoning. Their relatively small size makes it possible to deploy them in environments with limited resources such as laptops, desktops or your own cloud infrastructure, democratizing access to state of the art AI models and helping foster innovation for everyone. These models were trained on a dataset of text data that includes a wide variety of sources. The 27B model was trained with 14 trillion tokens, the 12B model was trained with 12 trillion tokens, 4B model was trained with 4 trillion tokens and 1B with 2 trillion tokens. Here are the key components: The combination of these diverse data sources is crucial for training a powerful multimodal model that can handle a wide variety of different tasks and data formats.

• gemma-4-31B-it-qat-q4_0-unquantized-assistant↗

  Google

  > [!Note] > This model card is for the new versions of the Gemma 4 family optimized with Quantization-Aware Training (QAT), which allows preserving similar quality to bfloat16 while dramatically reducing the memory requirements to load the model. > Four versions of the QAT checkpoints are available: > * **Unquantized QAT checkpoints** (Q4_0): Half-precision weights extracted from the QAT pipeline, ideal for custom downstream compilation and research. Available for Gemma 4 E2B, E4B, 12B, 26B A4B, and 31B, and their drafter models. > * **GGUF** (Q4_0): Ready-to-deploy formats for broad ecosystem compatibility. Available for Gemma 4 E2B, E4B, 12B, 26B A4B, and 31B. > * **Mobile-optimized** (wNa8o8): A custom schema engineered explicitly for mobile hardware efficiency. It features targeted 2-bit decoding layers, optimized KV caches, and static activations to maximize VRAM savings. Available for Gemma 4 E2B and E4B. > * **Compressed Tensors** (w4a16): QAT checkpoints serialized in the compressed-tensors format for native, optimized inference with vLLM. Available for Gemma 4 E2B, E4B, 12B, and 31B. > * **Assistant Compatibility**: When using multi-token prediction (speculative decoding) with an assistant model alongside a QAT target model, the assistant model must also be a QAT checkpoint with the same precision to ensure compatibility. Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants. Gemma 4 features a context window of up to 256K tokens and maintains multilingual support in over 140 languages. Featuring both Dense and Mixture-of-Experts (MoE) architectures, Gemma 4 is well-suited for tasks like text generation, coding, and reasoning. The models are available in five distinct sizes: **E2B**, **E4B**, **12B**, **26B A4B**, and **31B**. Their diverse sizes make them deployable in environments ranging from high-end phones to laptops and servers, democratizing access to state-of-the-art AI.

• t5-small↗

  google-t5

  1. 모델 세부 정보 2. 활용 사례 3. 편향, 위험 및 한계 4. 학습 세부 정보 5. 평가 6. 환경 영향 7. 인용 8. 모델 카드 작성자 9. 모델 시작 방법 > T5를 통해 우리는 모든 NLP 작업을 입력과 출력이 항상 텍스트 문자열인 통일된 텍스트-투-텍스트 형식으로 재구성할 것을 제안합니다. 이는 클래스 레이블이나 입력의 일부 구간만 출력할 수 있는 BERT 계열 모델들과 대조됩니다. 우리의 텍스트-투-텍스트 프레임워크는 어떤 NLP 작업이든 동일한 모델, 손실 함수 및 하이퍼파라미터를 사용할 수 있게 해줍니다. > 우리의 텍스트-투-텍스트 프레임워크는 기계 번역, 문서 요약, 질의응답 및 분류 작업(예: 감정 분석)을 포함한 모든 NLP 작업에서 동일한 모델, 손실 함수 및 하이퍼파라미터를 사용할 수 있게 합니다. 심지어 T5를 숫자 자체가 아닌 숫자의 문자열 표현을 예측하도록 학습시킴으로써 회귀 작업에도 적용할 수 있습니다.

• ltxv-13b-0.9.7-dev-GGUF↗

  ThePun1sher

  The model files can be used in ComfyUI with the ComfyUI-GGUF custom node. Place the required model(s) in the following folders: *As this is a quantized model not a finetune, all the same restrictions/original license terms still apply.* *Comfyui now supports the ggufs natively, so you just need to update comfyui to the latest version and if some issues persist update all the nodes in the workflow*

• Qwen3.8-Flash-Next-131B-A6B-GGUF↗

  Cyronius

  A 64.8 GB GGUF of Qwen/Qwen3.8-Flash-Next that scores at parity with the full 90 GB quant on tool-calling, GSM8K, and MMLU — 28% smaller, with **zero training and zero transformer surgery**. A 75.4 GB Q4_0 variant of the same cut is also available for older GPUs; see Files. Qwen3.8-Flash-Next carries a 51B-parameter n-gram hash-embedding table (~29% of total weights): 16 independent hash heads, 8 per n-gram type (bigrams / trigrams), injected at layer 1. This model keeps **1 head of 8 per n-gram type** and drops the other 14, exploiting the built-in redundancy of multi-head hashing (each n-gram is looked up several independent ways; a Bloom-filter-style safety margin). Every other tensor is copied **byte-for-byte** from unsloth's UD-Q3_K_XL dynamic quant, so unsloth's imatrix calibration is preserved exactly. The Q4_0 build is **bigger**, not smaller. The n-gram table is what makes this model compact; Q4_0 simply stores every other tensor at more bits than Q3_K_XL does. You are buying throughput with VRAM. On gfx906 that trade is usually worth it — Q4_0 reaches llama.cpp's MMVQ path while K-quants and IQ4_NL fall through to a generic dequant path several times slower.

• llama-3-70B-Instruct-abliterated↗

  failspy

  This is meta-llama/Llama-3-70B-Instruct with orthogonalized bfloat16 safetensor weights, generated with the methodology that was described in the preview paper/blog post: 'Refusal in LLMs is mediated by a single direction' which I encourage you to read to understand more. TL;DR: this model has had certain weights manipulated to "inhibit" the model's ability to express refusal. It is not in anyway _guaranteed_ that it won't refuse you, understand your request, it may still lecture you about ethics/safety, etc. It is tuned in all other respects the same as the original 70B instruct model was, just with the strongest refusal direction orthogonalized out. In the repo, I've included `refusal_dir.pth` If you have Llama-3-70B-Instruct model downloaded already, you can use the ortho cookbook to apply it to your downloaded model, which will make it the same as what you'd download from here.

• Florence-2-base↗

  florence-community

  > [!NOTE] > This is the repository for official transformers converted checkpoint of Microsoft's Florence model. **This is a continued pretrained version of Florence-2-large model with 4k context length, only 0.1B samples are used for continue pretraining, thus it might not be trained well. In addition, OCR task has been updated with line separator ('\n'). COCO OD AP 39.8** Florence-2 is an advanced vision foundation model that uses a prompt-based approach to handle a wide range of vision and vision-language tasks. Florence-2 can interpret simple text prompts to perform tasks like captioning, object detection, and segmentation. It leverages our FLD-5B dataset, containing 5.4 billion annotations across 126 million images, to master multi-task learning. The model's sequence-to-sequence architecture enables it to excel in both zero-shot and fine-tuned settings, proving to be a competitive vision foundation model.

• wav2vec2-large-xlsr-53-spanish↗

  jonatasgrosman

  Fine-tuned facebook/wav2vec2-large-xlsr-53 on Spanish using the train and validation splits of Common Voice 6.1. When using this model, make sure that your speech input is sampled at 16kHz. This model has been fine-tuned thanks to the GPU credits generously given by the OVHcloud :) The script used for training can be found here: https://github.com/jonatasgrosman/wav2vec2-sprint