뉴스 및 전체 검색
뉴스, 도구, 모델, 가이드, 주제, 워크플로, 프롬프트를 함께 검색합니다.
AI 모델 149
“Transcript AI”
Qwen3.8-Flash-Next-NVFP4-GPTQ-Strata
Qwen's original Qwen3.8-Flash-Next with **every routed expert in NVFP4, quantized from the BF16 checkpoint by GPTQ**, packed for Strata NVFP4: the 125B hybrid MoE on one RTX 20-50 card (12 GB of VRAM or more) and 64 GB of RAM or more. - **Accuracy:** how often the model's most likely next token is the Q8_0 model's. Measured on 5,153 tokens of code, docs, English and Ukrainian chats and agent transcripts, every model fed the same text. Every model ran with its experts on the CPU and the route tail skip off, so only the quantization differs. The Q8_0 model also has Q8_0 attention, Qwen's BF16 router and Qwen's own FP8 PLE, so it is the original's precision everywhere but the 8-bit weights. - **Speed:** RTX 5090 (32 GB), Ryzen 9 9950X3D, 128 GB of DDR5-5600, Windows 11, Strata NVFP4 0.1.42-nvfp4.1's engine with the arguments 0.1.42-nvfp4.2's setup writes for each model. Median of 3 runs: 4 greedy chats (up to 1,500 tokens each) and one 32,007-token prompt. - **RAM:** "RAM in use" is the engine and the server at the speed shown; "PC for this speed" is the installed RAM that holds it with Windows beside it. Smaller PCs still run the models, slower: the NVFP4 GPTQ and NVIDIA's on a 64 GB PC in the low-RAM mode decode ~115 tokens/s and read a 32K prompt in ~9 s (docs/NVFP4.md, "64 GB of RAM"); Unsloth's files go into a RAM-budget mode there and read experts from the SSD (UD-Q4_K_XL: 7-8.5 tokens/s at 64 GB, upstream's measurement); setup's minimum is 48 GB for them and the ISTA 2-bit files, 60-64 GB for the ISTA 3-bit ones. - **Unsloth's files keep every expert in RAM where they fit with 16 GiB more** (setup from 0.1.42-nvfp4.2): from ~88 GB of RAM for UD-Q4_K_XL and ~68 GB for UD-Q3_K_XL. Below that, setup's RAM-budget mode reads the rest from the GGUF on the SSD. Setup before 0.1.42-nvfp4.2 used the budget mode on every PC; on 128 GB UD-Q4_K_XL then decoded 118 tokens/s and read the 32K prompt in 30 s (run setup again to switch an existing install). - UD-Q3_K_XL's experts are i-quants despite the name: IQ3_XXS gate/up with IQ4_NL or Q8_0 down. - The GGUF models run upstream's route tail skip by default: ~10% faster decode, and answers a little further from Q8_0 than the table shows. - The two 128 GB models need a 128 GB PC (setup asks below that). Q8_0 runs there in the low-RAM mode by itself (100 GiB of experts locked in RAM); the mix keeps every expert in RAM.
gemma-4-E4B-it-heretic
A decensored variant of google/gemma-4-E4B-it, produced with Heretic v1.4.0 (directional ablation / "abliteration"). Google's Gemma 4 E4B is an 8B-class multimodal MoE that takes text, image, video, and audio and generates text, with a 128K context and configurable thinking modes; refusal behaviour is suppressed here via targeted weight edits to the attention output and MLP down-projections rather than fine-tuning, so the model's multimodal perception and reasoning are left largely intact. **Who this is for:** developers who want Gemma 4's multimodal range — image, video, and audio input — without the alignment refusals, on hardware they already own. E4B is the sweet spot in the family: it runs on-device on a laptop, handles 128K context, supports thinking modes, and here answers directly instead of declining. Best for local document/UI/screenshot analysis, audio transcription and understanding, and multilingual assistant work over 140 languages. Weights only, at this model's native 8B size; add ~1 GB per 32K of context. Because this is a mixture-of-experts model, memory *bandwidth* matters far more than raw VRAM — a CPU with fast RAM often outruns a slow GPU. OOM? Drop one quant level. Headroom to spare? Go one up.
voxcpm2-GGUF
Historical T4 Vulkan component measurements found CFM per audio step of 70.2 ms for Q8_0 versus 59.4 ms for the mixed recipe at ten steps. An eight-step run measured 46.7 ms. These component timings do **not** establish speech quality or an accepted end-to-end speedup. The mixed file was built with `crispasr-quantize voxcpm2-f16.gguf out.gguf q8_0 --tensor-type "^locdit\.=f16"`. A previous CPU synthesis roundtrip is not GPU acceptance. **Validation update, 2026-10-09:** matched real T4 tests at ten steps and seed 2 produced repeated tokens in the **Nemotron ASR transcripts** of the short test sentence for both Q8_0 and the mixed recipe. Six of sixteen exact primary ASR cases passed. This rejected gate is not proof that synthesized audio audibly stutters: a later fixed-audio comparison found **both Parakeet and Qwen3 transcribe every one of 22 original Python/native CPU/GPU controls exactly** across three resampling paths, while Nemotron retains errors. Resampling alone does not resolve the discrepancy. The completed original NVIDIA float32 and native F16 comparison reads **all 22 VoxCPM2 controls correctly**; native Q4 reads only 8/22. Original and F16 normalized words agree on all 26 tested clips, and fresh/reused sessions agree. The repeated tokens are therefore attributable to the native Nemotron Q4 recognizer on these fixed controls. Critical RNNT precision guards are being tested before changing that quant or its defaults. The published Q4 primary gate remains rejected while that recognizer is repaired. This diagnosis does not certify eight steps, Intel B390 performance, or complete original/native waveform parity. Defaults remain ten steps. Original NVIDIA/F16/Q4 inputs, features, encoder states and complete transcripts (SHA256 `7762f2a73d8ae7dad05d667fbf0bca393a0b3b7ad622b1b1f09473fa8f5d0463`).
Qwen3-ASR-1.7B-hf
The Qwen3-ASR family includes **Qwen3-ASR-1.7B** and **Qwen3-ASR-0.6B**, which support language identification and ASR for 52 languages and dialects. Both leverage large-scale speech training data and the strong audio understanding capability of their foundation model, Qwen3-Omni. The 1.7B version achieves state-of-the-art performance among open-source ASR models and is competitive with the strongest proprietary commercial APIs. - **All-in-one:** Supports language identification and speech recognition for 30 languages and 22 Chinese dialects, including English accents from multiple countries and regions. - **Excellent and Fast:** High-quality and robust recognition under complex acoustic environments. Qwen3-ASR-0.6B reaches 2000× throughput at a concurrency of 128. Both models support streaming/offline unified inference with a single model and handle long audio. - **Forced Alignment:** Qwen3-ForcedAligner-0.6B supports timestamp prediction for arbitrary units within up to 5 minutes of speech in 11 languages, surpassing E2E-based forced-alignment models in accuracy. `apply_transcription_request` handles chat-template formatting for you and is the recommended entry point.
s1-mini-GGUF
GGUF builds of superwhisper/s1-mini, release v1, for llama.cpp, Ollama, LM Studio, and anything else built on llama.cpp. You can use it in your own dictation app too, just check the license first. S1-mini is a 0.6B-parameter text normalizer for speech-to-text output. It takes a raw ASR transcript and rewrites it as clean written text: fillers removed, false starts and self-corrections resolved to the value the speaker landed on, punctuation and capitalization applied, and spoken numbers, dates, times, currency and email addresses rendered in written form. At Q4_K_M it is a 462 MiB file that runs comfortably on a laptop CPU, and on a held-out set of 7,519 English cases it reaches 94.8% token accuracy.
granite-speech-3.3-2b
**Model Summary:** Granite-speech-3.3-2b is a compact and efficient speech-language model, specifically designed for automatic speech recognition (ASR) and automatic speech translation (AST). Granite-speech-3.3-2b uses a two-pass design, unlike integrated models that combine speech and language into a single pass. Initial calls to granite-speech-3.3-2b will transcribe audio files into text. To process the transcribed text using the underlying Granite language model, users must make a second call as each step must be explicitly initiated. The model was trained on a collection of public corpora comprising diverse datasets for ASR and AST as well as synthetic datasets tailored to support the speech translation task. Granite-speech-3.3-2b was trained by modality aligning granite-3.3-2b-instruct (https://huggingface.co/ibm-granite/granite-3.3-2b-instruct) to speech on publicly available open source corpora containing audio inputs and text targets. Compared to the initial release, revision 3.3.2 * supports multilingual speech inputs in English, French, German, Spanish and Portuguese, * provides transcription accuracy improvements for English ASR by using a deeper acoustic encoder and additional training data.
security-8b
An 8B cybersecurity agent (offense + defense) fine-tuned from `Qwen/Qwen3-8B` on **2197 verified agentic + Q&A traces**, distilled from GLM-5.2 (round 4) and Kimi K3/K2.6 (rounds 1–3). Trained to operate an autonomous security harness — plan, run bash, read observations, write files, finish — not just answer security questions. Pipeline per iteration: failure analysis on eval traces → parameterized, decontaminated seed factory → teacher best-of-3 rejection-sampled agentic transcripts (judged + artifact-checked) → full-FT SFT → multi-run attack/defend eval → next round. Run-to-run variance is significant at temperature 0.7; single-run scores are not meaningful for this suite. Known v4 gap: `atk-sqli` — the model prefers to start the staged vulnerable app and fuzz it over HTTP instead of reading the offline artifacts (trace-verified behavioral prior, targeted in round 5).
cohere-transcribe-03-2026-gguf
GGUF conversions of CohereLabs/cohere-transcribe-03-2026 for use with transcribe.cpp. Ported from upstream commit 76b8b23, pinned 2026-04-16. Validated against the Transformers reference at transcribe.cpp commit bf0d0b7 on 2026-04-18. Offline multilingual speech-to-text covering 14 languages (English, French, German, Spanish, Italian, Portuguese, Dutch, Polish, Greek, Arabic, Japanese, Chinese, Vietnamese, Korean). A Conformer encoder with a Transformer encoder-decoder head (cross-attention, tied token embedding). Takes a 16 kHz mono WAV and produces a transcript. Decoding is autoregressive.
whisper-tiny-gguf
Ported from upstream commit 169d4a4, pinned 2026-04-25. Validated against the transformers reference at transcribe.cpp commit 0a26478 on 2026-09-13. OpenAI Whisper tiny — converted to GGUF for transcribe.cpp. Multilingual transcription, language detection, and speech translation (audio in any supported language → English text). 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.
Audio-Flamingo-GGUF
These self-contained GGUFs package the model tensors, tokenizer, configuration, and audio.cpp model spec for Audio Flamingo 3, Audio Flamingo Next, and Music Flamingo. AF3 and Next support transcription, captioning, and questions about speech, music, and environmental sounds. Music Flamingo specializes in music understanding; it produces text, not audio. Select a GGUF with `--model`. Use a question such as `Describe the instruments and tempo.` for general audio understanding. Requested timestamps or speaker descriptions are generated text, not structured alignment or diarization results. For server use, configure family `audio_flamingo`, task `asr`, and mode `offline`, then submit audio to `/v1/audio/transcriptions` with the instruction in `prompt`.