저장한 도구도구 추천
FOXAI SEARCH

전체 검색

AI 모델 2119

  • pythia-31m↗

    EleutherAI

    - **Developed by:** EleutherAI - **Model type:** Transformer-based Language Model - **Language:** English - **Learn more:** Pythia's GitHub repository for training procedure, config files, and details on how to use. See paper for more evals and implementation details. - **Library:** GPT-NeoX - **License:** Apache 2.0 - **Contact:** Join the EleutherAI Discord and post in `#release-discussion`. For general correspondence: contact@eleuther.ai > **Note:** Prior to Feb 27th, 2026, a model was hosted at this URL that was trained on the deduplicated Pile. That model is now at EleutherAI/pythia-31m-deduped and this model was correctly trained on the standard Pile. We apologize for any confusion this has caused. > **Note:** Pythia-31M was trained after the original Pythia suite at the request of interpretability researchers who wanted a smaller model with the same training setup. It uses the same tokenizer, hyperparameter conventions, and checkpoint schedule as the rest of the Pythia suite. A deduplicated variant is available at `EleutherAI/pythia-31m-deduped`.

  • Kimi-K3-DSpark↗

    Inferact

    **Inferact/Kimi-K3-DSpark** is an **MLA-native DSpark** draft model that accelerates Kimi-K3 on **vLLM**, served through vLLM's native `dspark` speculative method. The draft trains on **target hidden states extracted from vLLM itself** — the same engine that serves it, so the numerics it learns from are the numerics it meets at inference. TorchSpec provides the loop, streaming those hidden states from live target inference into concurrent FSDP draft training. **DSpark** = a block-diffusion backbone of 5 dense layers with non-causal attention, drafting 7 tokens in a single parallel pass, a low-rank sequential **Markov head** supplying the intra-block dependency, and a confidence head for resource-aware scheduling. Mirroring Kimi-K3's own MLA attention means draft and target share one KV layout (a compact 576-element latent per token), so the draft's pages unify with the target's cache — KV offloading and P/D disaggregation work with no separate page format.

  • jina-reranker-v3.5↗

    jinaai

    > [!TIP] > GGUF with quantizations and MLX versions are now available. `jina-reranker-v3.5` is a 0.6B-parameter multilingual listwise document reranker and a drop-in upgrade to `jina-reranker-v3`. It keeps the same *last-but-not-late* (LBNL) listwise interface — ranking a query against many candidates in one forward pass — while improving **domain robustness**, **structured-data ranking**, **multilingual retrieval**, and **inference efficiency**. Built on Qwen3-0.6B with 28 layers (hybrid **3L2G** attention: sliding-window layers interleaved with global layers, window = 1024, pinned terminal global layer required by LBNL) and a lightweight MLP projector (1024→512→512), it supports up to 131K tokens of context and ranks lists of documents jointly via causal self-attention and cosine scoring.

  • Hemmingway-1↗

    HemmingwayAI

    Ask most models for a text to your landlord and you get three options, a preamble, and a paragraph explaining the options. Hemmingway-1 just gives you the text. We built it for the writing people actually do every day: messages, emails, the awkward note to a colleague, the thing you have been putting off. Then we tested it against the biggest models in the world at exactly that, and it came first. We tested on eighty real requests. Every answer went head to head against another model's answer to the same request, shuffled so the judge never knew which was which.

  • Audio8-TTS-Preview-0.6b-GGUF↗

    js-byte

    GGUF conversion of **Audio8/Audio8-TTS-Preview-0.6b** for audio.cpp. No new training. **Original checkpoint:** 0.6B DualAR — Slow 24×896 (14 heads, 2 KV), Fast 4×896, 10×4096 codebooks, 44.1 kHz codec (2048 samples/frame), 2048 context. Languages: yue/zh/nl/en/fr/de/it/ja/ko/pl/es/auto. See original model card for details and evaluation. Both embed `config.json`, `tokenizer.json`, `tokenizer_config.json` and `model_specs/audio8_tts.json`. `q8_0` is the recommended default.

  • sepformer-wsj02mix↗

    speechbrain

    This repository provides all the necessary tools to perform audio source separation with a SepFormer model, implemented with SpeechBrain, and pretrained on WSJ0-2Mix dataset. For a better experience we encourage you to learn more about SpeechBrain. The model performance is 22.4 dB on the test set of WSJ0-2Mix dataset. The system expects input recordings sampled at 8kHz (single channel). If your signal has a different sample rate, resample it (e.g, using torchaudio or sox) before using the interface. To perform inference on the GPU, add `run_opts={"device":"cuda"}` when calling the `from_hparams` method.

  • Gemma-4-12B-it-AEON-Abliterated-MLX-8bit↗

    AEON-7

    > **The near-lossless Apple-Silicon build** of `AEON-7/Gemma-4-12B-it-AEON-Abliterated-K4-BF16`. Every language-decoder linear is `mxfp8` (8-bit, group 32); the tied soft-capped head and the vision/audio projectors are kept at **bf16**. Built and validated on a **MacBook Pro M4 Pro (48 GB)**. > > **Target hardware: Apple Silicon (M-series), best on 24 GB+ unified memory.** Full multimodal (text + image + audio) via `mlx-vlm`. > > Want the smallest build for a 16 GB Mac? See the compact FP4 sibling: `…-MLXFP4`. This is the **maximum-fidelity** member of the MLX quant grid: measured **top-1 token agreement 0.924 and median KL ≈ 0.002 nats** against the BF16 source on the model's own greedy output — below the perceptual/sampling-noise floor on the typical token. **Near-lossless**, and the recommended build when quality is paramount. **0 → running on a fresh Mac** (no Python, no tools needed) — `uv` installs a correct Python + the deps for you:

  • MiniCPM-V-2↗

    openbmb

    **MiniCPM-V 2.8B** is a strong multimodal large language model for efficient end-side deployment. The model is built based on SigLip-400M and MiniCPM-2.4B, connected by a perceiver resampler. Our latest version, **MiniCPM-V 2.0** has several notable features. MiniCPM-V 2.0 achieves **state-of-the-art performance** on multiple benchmarks (including OCRBench, TextVQA, MME, MMB, MathVista, etc) among models under 7B parameters. It even **outperforms strong Qwen-VL-Chat 9.6B, CogVLM-Chat 17.4B, and Yi-VL 34B on OpenCompass, a comprehensive evaluation over 11 popular benchmarks**. Notably, MiniCPM-V 2.0 shows **strong OCR capability**, achieving **comparable performance to Gemini Pro in scene-text understanding**, and **state-of-the-art performance on OCRBench** among open-source models. LMMs are known for suffering from hallucination, often generating text not factually grounded in images. MiniCPM-V 2.0 is **the first end-side LMM aligned via multimodal RLHF for trustworthy behavior** (using the recent RLHF-V [CVPR'24] series technique). This allows the model to **match GPT-4V in preventing hallucinations** on Object HalBench.

  • DeepSeek-V4-Flash-0731-REAP-150b-TQ3_4S-GGUF↗

    Tinker-Stack

    A **custom, selectively-quantized GGUF** of the DeepSeek-V4-Flash-0731 architecture, pruned to the 150B "REAP" variant, quantized with a hand-tuned per-tensor recipe and wrapped in the **TurboQuant `TQ3_4S` container** from the `turbo-tan/llama.cpp-tq3` fork. This is **not** a plain uniform quant. It is the result of iterative quantization on a 3× Turing GPU box, and the recipe reflects a specific, measured trade-off: maximize context length (256K) while holding wikitext-2 perplexity within noise of the IQ3_XXS baseline that proved to be the local optimum for this model. > [!WARNING] > **Experimental build.** This is a hand-tuned, selectively-quantized experimental release, > not a polished upstream quant. It requires the non-stock > `turbo-tan/llama.cpp-tq3` fork to load > (`TQ3_4S` container is rejected by stock llama.cpp), the recipe and serving flags are > tuned for our specific 3× Turing hardware (§3, §6), and the benchmarks in §4 are our own > internal measurements, not standardized leaderboard runs. Expect rough edges, and treat > the numbers as relative to sibling builds on the same rig rather than absolute claims. > If you just want a drop-in quant of this model, use the upstream REAP release instead.

  • Qwen3.8-Flash-Next-IQ3_S-GGUF↗

    boxrunner

    A GGUF quantization of Qwen/Qwen3.8-Flash-Next, a 512-expert mixture-of-experts model with a 262 144-token context and image input. It averages 4.081 bits per weight. It is the same file as boxrunner/Qwen3.8-Flash-Next-mixC-GGUF (map C) with one change: the routed experts' gate and up projections are **IQ3_S** instead of IQ2_XXS. Every other tensor is byte-identical to map C. Routed-expert down is IQ4_NL, the n-gram table is IQ4_NL, and the rest is Q5_K, Q6_K and Q8_0. All 24 576 experts together take 53.3 GiB, so on a 64 GB Mac they do not all fit in memory at once. Map C's do (40.4 GiB). The two parts were cut with llama.cpp's `gguf-split`. Put them in one folder and open part 1; the engine finds part 2 itself. `llama-gguf-split --merge` turns them back into one file. The projector and the draft head are the same files as in the map C repository.