전체 검색
AI 모델 2119
- wav2vec2-large-xlsr-malayalam↗
gvs
Fine-tuned facebook/wav2vec2-large-xlsr-53 on ml (Malayalam) using the Indic TTS Malayalam Speech Corpus (via Kaggle), Openslr Malayalam Speech Corpus, SMC Malayalam Speech Corpus and IIIT-H Indic Speech Databases. The notebooks used to train model are available here. When using this model, make sure that your speech input is sampled at 16kHz. The model can be evaluated as follows on the test data of combined custom dataset. For more details on dataset preparation, check the notebooks mentioned at the end of this file. A combined dataset was created using Indic TTS Malayalam Speech Corpus (via Kaggle), Openslr Malayalam Speech Corpus, SMC Malayalam Speech Corpus and IIIT-H Indic Speech Databases. The datasets were downloaded and was converted to HF Dataset format using this notebook
- Latxa-Llama-3.1-8B-Instruct↗
HiTZ
We introduce Latxa 3.1 8B Instruct, an instructed version of Latxa. This new Latxa is based on Llama-3.1 (Instruct), which we trained on our Basque corpus (Etxaniz et al., 2024) comprising 4.3M documents and 4.2B tokens using language adaptation techniques (paper in preparation). > [!WARNING] > DISCLAIMER > > This model is still under development. > Further training details will be released with the corresponding research paper in the near future. Our preliminary experimentation shows that Latxa 3.1 8B Instruct outperforms Llama-3.1-Instruct by a large margin on Basque standard benchmarks, and particularly, on chat conversations. In addition, we organized a public arena-based evaluation, on which Latxat competed against other baselines and proprietary models such as GPT-4o and Claude Sonnet. The results showed that Latxa ranked 3rd, just behind Claude and GPT-4 and above all the other same-size competitors. The official paper is coming soon. Latxa is a family of Large Language Models (LLM) based on Meta’s LLaMA models. Current LLMs exhibit incredible performance for high-resource languages such as English, but, in the case of Basque and other low-resource languages, their performance is close to a random guesser. These limitations widen the gap between high- and low-resource languages when it comes to digital development. We present Latxa to overcome these limitations and promote the development of LLM-based technology and research for the Basque language. Latxa models follow the same architecture as their original counterparts and were further trained in Latxa Corpus v1.1, a high-quality Basque corpora.
- Qwen3.8-Flash-Next-MLX-Serve-iQ-MLX-4.7bpw↗
ddalcu
mlx-serve pack of Qwen/Qwen3.8-Flash-Next, the Qwen4 preview architecture (`model_type: qwen4_exp`). Runs on a 128 GB Mac with about 75 GB resident. Includes the MTP head and the vision tower (image and video input). This is the calibrated twin of Qwen3.8-Flash-Next-MLX-Serve-mixed-4-8bit: the same layout and the same size, byte for byte (every expert 4-bit group 64, the rest 8-bit), but every expert is quantized with an importance matrix collected on the bf16 model, so the scales follow the activations the weights actually see. Same speed, same memory, closer to bf16. The importance matrix was collected on the bf16 checkpoint over 1.35 M tokens of agent transcripts, code (including SWE-bench Lite issues and patches), prose and math, every expert of every layer hit. Quality is scored against the bf16 model's own logits on 290 held-out positions (top-1: the pack's greedy token equals bf16's; KLD: KL(bf16 || pack) over bf16's top-1024 tokens; the KLD mean is dominated by a few flat prose positions, so the median is the steadier number):
- Qwen3.5-4B-RKLLM↗
HanzoHuang
Use the files for your board’s exact SoC. `w4a16` uses 4-bit weights and 16-bit activations; `w4a16_g128` adds a weight group size of 128. `w8a8` uses 8-bit weights and activations. File size below is download size, not total runtime RAM. For image input, also download the **vision encoder for the same SoC**. Keep the language model and vision encoder from this release together. With Docker installed on a matching Rockchip board, this example starts the **RK3576 w4a16_g128** model and its API on port 8001:
- dscript_human_v1↗
samsl
- Qwen3.8-27B-DFlash2↗
incoai
This repository contains the DFlash 2 draft model for `Qwen/Qwen3.8-27B`. It is not a standalone language model: it runs inside a speculative decoding server and drafts tokens for the target model to verify. The checkpoint is also mirrored at `z-lab/Qwen3.8-27B-DFlash2`. 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. See the blog post for other engines and more details.
- indo-sentiment-analysis↗
agufsamudra
This model is fine-tuned from **`indobenchmark/indobert-base-p1`** for binary sentiment classification (Positive/Negative) on Indonesian text. - **Developed by:** agufsamudra - **Model type:** Text Classification - **Language(s):** Indonesian (id) - **License:** Apache-2.0 - **Fine-tuned from model:** indobenchmark/indobert-base-p1 This model is intended for binary sentiment classification tasks in Indonesian language texts. It predicts whether a given text expresses positive or negative sentiment.
- Qwen3.8-27B-DFlash2-GGUF↗
incoai
This repository contains GGUF conversions of `incoai/Qwen3.8-27B-DFlash2`, the DFlash 2 draft model for `Qwen/Qwen3.8-27B`. It is not a standalone language model: it runs inside a speculative decoding server and drafts tokens for the target model to verify. The checkpoints are also mirrored at `z-lab/Qwen3.8-27B-DFlash2-GGUF`. 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. Build llama.cpp with DFlash 2 support (PR #27342):
- GLM-5.3-Flash-Strata-Q2Q3-Experts↗
lte-ai
Experimental GGUF weights for running **GLM-5.3-Flash** on a normal PC with the LightTransport Strata fork: expert sidecars and a complete full model for a 128 GB host with a 16 GB NVIDIA GPU, a smaller REAP-224 model, and a complete REAP-50 model for a 64 GB host with a 16 GB AMD GPU. The sidecars requantize routed expert gate/up weights from IQ2_XS to Q2_K and down weights from IQ3_XXS to Q3_K; this changes weight values and is neither a lossless repack nor a fine-tune. REAP-224 keeps 224 of 288 routed experts per layer; REAP-50 keeps 144. Read the quality numbers below before choosing a model. Current measurements, setup for one and two GPUs, and context-length limits are in `docs/GLM53F_RELEASE.md` on the fork's `glm53f` branch; bringing it up on other hardware is described in `docs/GLM53F_NEW_HARDWARE.md`. The conversion recipe is `docs/GLM_RELEASE_Q2Q3.md`. **On a 128 GB machine use `GLM-5.3-Flash-Q23-FP8-assembled.gguf`** (or `experts-q23-fp8.gguf` with the Unsloth shards); **on a 64 GB machine use `reap50/GLM-5.3-Flash-REAP50-Q23-FP8-assembled.gguf`.** Their routed experts have the same types as Q23 (Q2_K gate/up, Q3_K down) but are quantized once, directly from the official FP8 weights, instead of from Unsloth's 2-bit experts (REAP-50: from its Q4_K_M experts). Measured against exact logits of the FP8 release on 16 held-out sequences (8,192 tokens):
- Gemma-4-E4B-Samantha-i1-GGUF↗
mradermacher