全站搜索

AI 模型 2119

• rinalmo-giga↗

  multimolecule

  This is an UNOFFICIAL implementation of the RiNALMo: General-Purpose RNA Language Models Can Generalize Well on Structure Prediction Tasks by Rafael Josip Penić, et al. > [!TIP] > The MultiMolecule team has confirmed that the provided model and checkpoints are producing the same intermediate representations as the original implementation. **The team releasing RiNALMo did not write this model card for this model so this model card has been written by the MultiMolecule team.**

• qwen3-asr-arabic-uae↗

  vadimbelsky

  Fine-tuned Qwen/Qwen3-ASR-1.7B for **UAE Emirati Arabic** dialect speech recognition. - Matches informal Emirati dialect style (شي vs شيء, الاماكن vs الأماكن) - Removes spurious punctuation that the base model adds - Better handling of dialect-specific words and expressions - **Base model**: Qwen/Qwen3-ASR-1.7B (2B params, audio encoder + 1.7B LLM decoder) - **Training data**: ~22,500 UAE Emirati Arabic dialect samples from vadimbelsky/UAE_Arabic_English_Bilingual_Dataset_40k - **Strategy**: Audio encoder frozen, only LLM decoder fine-tuned (84.4% of params) - **Precision**: bfloat16 - **Epochs**: 3 - **Effective batch size**: 32 (batch 2 × gradient accumulation 16) - **Learning rate**: 2e-5 with linear schedule - **Gradient checkpointing**: enabled - **Text normalization**: Diacritics removed, alef/teh marbuta normalized, punctuation stripped

• Falcon3-3B-Instruct↗

  tiiuae

  **Falcon3** family of Open Foundation Models is a set of pretrained and instruct LLMs ranging from 1B to 10B parameters. **Falcon3-3B-Instruct** achieves strong results on reasoning, language understanding, instruction following, code and mathematics tasks. Falcon3-3B-Instruct supports 4 languages (English, French, Spanish, Portuguese) and a context length of up to 32K. We report in the following table our internal pipeline benchmarks. - We use lm-evaluation harness. - We report **raw scores** obtained by applying chat template and fewshot_as_multiturn. - We use same batch-size across all models.

• GLM-5.3-Flash-UNCENSORED-FP8↗

  dealignai

  > [!IMPORTANT] > **Serving note — read before your first request (vLLM / SGLang with a reasoning parser).** > > - **`reasoning_effort` only accepts `"low"` and `"high"`.** Anything else — `medium`, `xhigh`, > `off`, or omitting it — silently resolves to **`max`**, the deepest mode. Template line: > `reasoning_effort ... in ['low','high'] else 'max'`. > - **At `max`, a small `max_tokens` returns an EMPTY reply.** The model spends the whole budget > inside ` ` and you get `finish_reason="length"` with `content=""`. In multi-turn the empty > turn poisons the history and the conversation does not recover. Measured: `max` + `max_tokens > 2000` dies from turn ~4; `max` + `6000` is clean; `low` and `high` are clean at 2000. > **It is budget exhaustion, not a loop** — sampling changes do not rescue it. > - **Do NOT pass `enable_thinking`.** This template has no such variable (only `clear_thinking` and > `reasoning_effort`). Passing it — or a top-level `reasoning_effort: "none"` — turns the parser off > while the model still thinks, dumping raw reasoning into `content`. > - **Reasoning is returned in `message.reasoning`**, not `message.reasoning_content` > (vLLM renamed the output field; the input side still accepts both). > - `clear_thinking` must be nested inside `chat_template_kwargs` — a top-level key is ignored. > > > Use `"high"` for hard tasks; if you want `max`, give it `max_tokens >= 6000`. > > > **Video and image parameters — measured, because several are accepted then silently ignored.** > > - 🔴 **`media_io_kwargs.video.fps` must stay BELOW the clip's own frame rate.** Requesting a value > at or above it makes the placeholder builder count 3x the real tokens and **kills the engine**: > `ValueError: Attempted to assign N = N multimodal tokens to 3N placeholders` -> `EngineDeadError`, > server down until restarted. Upstream: vLLM #55644 / #55647. `fps: 2` is the safe, accurate choice. > - **`max_frames` and `num_frames` are not interchangeable across builds.** On vLLM nightly > `max_frames` is honoured and `num_frames` ignored; on the older `0.1.dev*` line it is the exact > reverse. A client that sets only one silently gets full-rate sampling on the other build. Set > `fps`, which every build honours. > - **Pixel controls: use `mm_processor_kwargs.max_image_tokens` / `min_image_tokens`.** > `max_pixels`, `min_pixels`, `size` and `detail` return HTTP 200 and

• GIST-all-MiniLM-L6-v2↗

  avsolatorio

  *GISTEmbed: Guided In-sample Selection of Training Negatives for Text Embedding Fine-tuning* The model is fine-tuned on top of the sentence-transformers/all-MiniLM-L6-v2 using the MEDI dataset augmented with mined triplets from the MTEB Classification training dataset (excluding data from the Amazon Polarity Classification task). The model does not require any instruction for generating embeddings. This means that queries for retrieval tasks can be directly encoded without crafting instructions.

• twitter-roberta-large-2022-154m↗

  cardiffnlp

  This is a RoBERTa-large model trained on 154M tweets until the end of December 2022 (from original checkpoint, no incremental updates). A base model trained on the same datais available here. These 154M tweets result from filtering 220M tweets obtained exclusively from the Twitter Academic API, covering every month between 2018-01 and 2022-12. Filtering and preprocessing details are available in the TimeLMs paper. Below, we provide some usage examples using the standard Transformers interface. For another interface more suited to comparing predictions and perplexity scores between models trained at different temporal intervals, check the TimeLMs repository.

• Qwen-Image-2.1-Turbo-mflux-8bit↗

  OsaurusAI

  An Osaurus 8-bit mflux quantization of Qwen/Qwen-Image-2.1-Turbo for Apple Silicon (17 GB, closest to the original; recommended). This is not an official Qwen release and is not endorsed by the Qwen team. Turbo is the 8-step distilled checkpoint of Qwen-Image-2.1. One bundle does both **text-to-image** and **reference editing** (up to 10 ordered reference images); the vision encoder is included. Requires **Osaurus 0.25.21 or newer**. Older versions treat this bundle as Qwen-Image-2.1 and sample it on that model's 40-step schedule, which gives degraded images.

• Qwen3.6-35B-A3B-DFlash2↗

  incoai

  This repository contains the DFlash 2 draft model for `Qwen/Qwen3.6-35B-A3B`. It is not a standalone language model: it runs inside a speculative decoding server and drafts tokens for the target model to verify. DFlash 2 is a block-diffusion drafter for speculative decoding. It predicts a whole block of tokens in a single pass and keeps the top candidates at every position. A lightweight selector then traces one coherent path through them. Two-tap dynamic convolutions in the backbone keep the draft from decaying toward the end of the block. Decoding is lossless: greedy output matches the target model exactly, and sampling preserves its distribution. We compare autoregressive decoding, Qwen3.6's built-in MTP with seven speculative steps, and DFlash 2. Both speculative methods propose seven draft tokens per verification step.

• Tess-2.0-Llama-3-70B-v0.2↗

  migtissera

  Tess, short for Tesoro (Treasure in Italian), is a general purpose Large Language Model series. Tess-2.0-Llama-3-70B-v0.2 was trained on the meta-llama/Meta-Llama-3-70B base. The change between v0.1 and this version, v0.2 is that v0.2 has undergone an additional step of uncensoring. Tess-2.0-Llama-3 was trained on the (still curating) Tess-2.0 dataset. Tess-2.0 dataset and the training methodology follows LIMA (Less-Is-More) principles, and contains ~100K high-quality code and general training samples. The dataset is highly uncensored, hence the model will almost always follow instructions. The model was only fine-tuned for 2-epochs with a low learning rate to try and preserve its entropy as much as possible. In v0.2, an uncensored dataset was also added to the mix.

• embeddinggemma-2-bf16↗

  mlx-community

  MLX conversion of Google EmbeddingGemma 2 in **BF16** format, retaining text, image, audio, and video encoders. The model produces normalized 768-dimensional embeddings. See the original model card for intended use, training information, evaluation, and limitations. Weight storage: 1.489 GB (decimal). Source revision: `914f7f89142e33e77833254d9c9b90c3cef7303b`. Converted with MLX-VLM revision `3d87e884`, from branch `pc/embeddinggemma-2`, and MLX 0.32.3. Apply the appropriate task prefixes from `config_sentence_transformers.json`. Queries and documents must use the same embedding dimension. Keep non-quantized weights and activations in BF16; do not cast this model to float16.