全站搜索

AI 模型 2119

• Qwen3.8-Flash-Next-Gyro-abliterated-Huihui-GGUF↗

  agentionai

  > **Updated 2026-10-13: both files.** Same file names, sizes and tensor types, better weights from an improved encode. This is huihui-ai/Huihui-Qwen3.8-Flash-Next-abliterated encoded the same way as our Gyro files. Abliteration is huihui-ai's work: it removes the refusal direction from the model's weights. We only quantized it. **Which setup fits your hardware, NVIDIA and AMD instructions, context limits, vision and speculative decoding are all on the main Gyro card and apply unchanged here; only the file names differ.**

• gemma-4-31B-it-qat-w4a16-ct↗

  Google

  Hugging Face | GitHub | Launch Blog | Documentation | Technical Report > [!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.

• Fun-ASR-MLT-Nano-2512-gguf↗

  handy-computer

  GGUF conversions of FunAudioLLM/Fun-ASR-MLT-Nano-2512 for use with transcribe.cpp. Ported from upstream commit cf67a93, pinned 2026-05-06. Validated against the FunASR reference at transcribe.cpp commit f094d28 on 2026-05-06. Offline speech-to-text covering 31 languages, with focused optimization on East and Southeast Asian languages: Chinese, English, Cantonese, Japanese, Korean, Vietnamese, Indonesian, Thai, Malay, Filipino, plus Arabic, Hindi, and 19 European languages (Bulgarian, Croatian, Czech, Danish, Dutch, Estonian, Finnish, Greek, Hungarian, Irish, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Swedish). Same architecture as Fun-ASR-Nano-2512 (~800M trainable parameters: frozen SenseVoiceEncoderSmall + 2-layer audio adaptor + bundled Qwen3-0.6B LLM); trained on a smaller multilingual corpus ("hundreds of thousands of hours" per the model card, vs Nano's "tens of millions"). Takes a 16 kHz mono WAV and emits text. Not streaming, no translation, no timestamps. ITN (inverse text normalization) is on by default, matching upstream, for readable casing, punctuation, and digits; pass `--no-itn` for spoken-form output.

• slimsam-50-uniform↗

  nielsr

• Qwen3.8-27B-MTPLX-Optimized-Speed↗

  Youssofal

  **MTPLX: the fastest way to run Qwen 3.8 on a Mac. Native multi-token-prediction speculative decoding on Apple Silicon, two to three times the speed of plain decoding, exact at any temperature.** This is the model MTPLX recommends for coding on a modern Mac with 32 GB or more. It is Qwen3.8-27B with its native multi-token-prediction head kept, so MTPLX can draft several tokens ahead and verify them in one pass. Same model, same output distribution, a lot more tokens per second. This is a Qwen3.8-27B MLX pack for MTPLX, the fastest way to run Qwen 3.8 27B on a Mac. All numbers on a MacBook Pro M5 Max, fans verified at maximum, sampled at the model's own settings. Conditions and sources: mtplx.com/benchmarks.

• bert-large-cased-whole-word-masking-finetuned-squad↗

  google-bert

  Pretrained model on English language using a masked language modeling (MLM) objective. It was introduced in this paper and first released in this repository. This model is cased: it makes a difference between english and English. Differently to other BERT models, this model was trained with a new technique: Whole Word Masking. In this case, all of the tokens corresponding to a word are masked at once. The overall masking rate remains the same. The training is identical -- each masked WordPiece token is predicted independently.

• Qwen-Image-2.1-Turbo-GGUF↗

  unsloth

  This is a GGUF quantized version of Qwen-Image-2.1-Turbo, the 8-step distilled checkpoint of Qwen-Image-2.1. unsloth/Qwen-Image-2.1-Turbo-GGUF uses Unsloth Dynamic 2.0 methodology for SOTA performance. - Important layers are upcasted to higher precision, per tensor, from a measured sensitivity scan. Every file uses the same per-tensor layout as the matching file in unsloth/Qwen-Image-2.1-GGUF, with an importance matrix collected on Turbo itself at its own 8-step schedule. - Run these with Unsloth Desktop, ComfyUI (ComfyUI-GGUF), stable-diffusion.cpp and more. A GGUF is the denoiser only, so it needs the VAE and the Qwen3-VL text encoder alongside it. Turbo uses the same text encoder and VAE as Qwen-Image-2.1. - VAE: unsloth/Qwen-Image-2.1-FP8 `vae/qwen_image_2.1_vae_bf16.safetensors`. Text encoder: unsloth/Qwen3-VL-8B-Instruct-GGUF `Qwen3-VL-8B-Instruct-UD-Q4_K_XL.gguf`, the Dynamic 2.0 4-bit rung rather than the uniform `Q4_K_M`. - Turbo is meant for **8 steps at CFG 1** with the sampling schedule it ships (`sample_sigmas` in model_index.json). Unsloth Desktop applies it automatically. LPIPS (AlexNet, lower is better) against the bf16 Turbo transformer, same text encoder, VAE, prompts and seeds, 1024x1024, 8 steps, CFG 1, rendered through Unsloth Desktop's diffusers route. Mean with 95% bootstrap CI over 8 prompts x 2 seeds; the held-out column uses 8 further prompts x 2 seeds that were not part of the importance-matrix calibration. Two bf16 runs of the same seeds give 0.000 (run-to-run floor).

• d3-mini↗

  vllm-sr

  **d3-mini** is the 4B multimodal foundation decision model of Decision 3.0, the decision models of vLLM Semantic Router. Give it an input (text or JSON, optionally with images and videos) and the questions you need answered: pick one of several options, say yes or no, or rate on a scale. It answers them all in one call and returns a probability for every answer, without generating text. **Try it free** at decision.vllm-sr.ai · Official Space: vllm-sr/Decision-3.0 - **Jev Decision Index 0.3.2: 52.6 on the official text board**, first among the entries of similar size and on the board's Pareto frontier; public suite 54.90, measured with the official 0.3 kit on the released weights: all 140,178 public requests answered, none unsupported. - **+7.7 over Decision 2.0** (its 4B model: 45.0 on the official board): +10.7 on the public suite, ahead in all five of its areas, and +8.2 and +3.4 on the private same-skill tests and new-domain tasks. - **Reads images:** multiple images per request (PNG, JPEG or WebP), given as paths, URLs, PIL images or base64 data URLs; every question of the request sees all of them. - **Reads videos:** multiple videos per request (MP4, WebM, MOV or MKV), given as paths, URLs, base64 data URLs or frame arrays, read at 2 frames per second; **Perception Test (multiple-choice video QA): 70.3** (internal evaluation). - **Speed:** a median of 18.1 ms for a text request, 112 ms for a request with an image and 336 ms for a request with a 10-second video, on one AMD Instinct MI325X GPU, one request at a time with AutoModel. - **Serving:** 151 requests/s with p99 under 1 s on one AMD Instinct MI325X with vLLM Semantic Router, 2.7× AutoModel (one-question text, with batching, now the default); with one request per forward, the former default, it gave answers identical to AutoModel's (Serving performance). - **Many questions, one call:** Choice, Yes / No and Score questions about the same input are answered together, each from its own forward pass over the input, with a probability for every option.

• whisper-small.en-gguf↗

  handy-computer

  Ported from upstream commit e872752, pinned 2026-04-25. Validated against the transformers reference at transcribe.cpp commit 0a26478 on 2026-09-13. OpenAI Whisper small.en — converted to GGUF for transcribe.cpp. English-only; faster than the multilingual model at the same size. Encoder-decoder transformer; 30-second windows with chunked long-form decoding. WER on the full LibriSpeech test-clean split (2,620 utterances), batch size 1, timestamps none. Figures without a commit were published before provenance was recorded.

• SAM2.1-Hiera-Tiny-Mask-Decoder↗

  litert-community