> [!WARNING] > **Withdrawn from Strata NVFP4's installer (2026-10-09): this quant often loops in long thinking.** > We captured a real Claude Code request and replayed it. It had a 72K-token prompt, thinking on, and no temperature > sent, so the server decoded greedily, which is its default. > - **This quant looped in 9 of 10 runs**, on Strata NVFP4 v0.1.41-nvfp4.2 and on v0.1.40.3-nvfp4.1 alike. It > repeated the same 372-word passage of its reasoning until `max_tokens`, instead of closing the thinking and > answering. > - With Qwen's thinking-mode sampling (temperature 0.6, top_p 0.95, top_k 20) it still looped in 2 of 5 runs. > - Our OrcaRouter quant, > made with the same recipe (every expert NVFP4, GPTQ), **looped in 0 of 5 runs** on the same request. The installer > now offers that one instead (`--family orca-nvfp4`). > > We have not established whether the cause is huihui's abliteration itself or its quantization. The abliteration > edited the experts' `down_proj`, which is NVFP4 here. > > If you run this model anyway: > - send a temperature with your requests; > - turn on the server's loop guard, `"reasoning_loop_recovery": "recover"` in the server config. It brought the > reply to an answer in 2 of 2 looping runs. **Installer:** Strata NVFP4 v0.1.41-nvfp4.1 and .2 listed this repository as `--family huihui-nvfp4`. From v0.1.41-nvfp4.3 the installer no longer offers it (see the warning above) and installs `orca-nvfp4` by default. To run this model by hand, see Run it. - **Chain:** `Qwen/Qwen3.8-Flash-Next` → `huihui-ai/Huihui-Qwen3.8-Flash-Next-abliterated` (abliteration, BF16) → this quantization. - **4.5 bits a weight for the experts, 63.3 GiB:** the size and the decode cost of a ModelOpt NVFP4 checkpoint, about **34% of plain rounding's error** on held-out tokens (below). No layer is kept in 8 bits. - **Only for Strata NVFP4.** The files are Strata's own runtime formats (an expert pack, GGUFs the fork reads, the MTP draft head). They do not load in transformers, vLLM or llama.cpp.
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
> [!WARNING] > **Withdrawn from Strata NVFP4's installer (2026-10-10): as an agent, this abliterated build is much worse than the > original Qwen.** > - **A multi-step Claude Code task** (building an interactive 3D page), same prompt on both models: the original > Qwen3.8-Flash-Next (our GPTQ of it) > ran 175 steps, checking and fixing its work from screenshots, and gave a clearly better result. This build > stopped after 10-70 steps (3 runs) and declared the work done. > - **It closes its thinking sooner.** Over 7,869 positions of the original model's thinking (teacher-forced), this > pack raises the probability of ` ` by +0.12 nats on average. > - On a captured 72K-token Claude Code request (5 greedy runs each) it thought 997 tokens on average, the original > 1,699. > - Its style of thinking is the same; it just stops earlier. > - The likely cause is OrcaRouter's weight edit: it removes the refusal direction from all 149 residual-writing > matrices of every layer, including the late layers whose edit pushes ` `. > - **What the installer offers instead.** From Strata NVFP4 0.1.41-nvfp4.5: > - the original Qwen (`--family qwen-nvfp4-gptq`), with an opt-in switch (`--uncensored on`, off by default); > - the switch removes refusals at run time, in layers 8-33 only, and leaves the weights untouched; > - with thinking on it refused 0 of 104 English and 0 of 30 Ukrainian held-out harmful requests; > - ` ` +0.02 nats; thinking length as the original's. > > An existing install of this model keeps working, and the files here are unchanged. **Installer:** Strata NVFP4 v0.1.41-nvfp4.1 to .4 listed this repository as `--family orca-nvfp4` (the default from .3). From v0.1.41-nvfp4.5 the installer no longer offers it (see the warning above). To run this model by hand, see Run it. - **Chain:** `Qwen/Qwen3.8-Flash-Next` → `orcarouter/Qwen3.8-Flash-Next-Uncensored` (abliteration, BF16) → this quantization. - **Same size and the same decode cost as the ModelOpt NVFP4 experts** (63.3 GiB, 4.5 bits a weight), about **33% of their error** on held-out tokens: the same format, other codes and scales. - End to end, an earlier build of the same recipe (2026-10-02, the fork's docs/NVFP4.md) had a median per-answer KL of 0.0070 against an all-Q8_0 reference where ModelOpt's experts had 0.0107 (65%), better in 12 of 12 prompts. This build was checked as described under Checks, not re-
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
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.
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
Qwen's original Qwen3.8-Flash-Next with **our NVFP4 GPTQ experts, and Q8_0 in the layers that move the answers most**, packed for Strata NVFP4. It is the most accurate model that fits a **128 GB PC** with every expert in RAM. Its most likely next token is the Q8_0 model's 95.0% of the time, against 93.6% for the all-NVFP4 GPTQ, and it moves the answers 43% less (KL). It decodes at about half the GPTQ model's speed. **Quick start (Windows):** from the Strata NVFP4 release 0.1.42-nvfp4.2 or newer, run `START-HERE.bat --family qwen-nvfp4-mix`: it checks the PC, downloads this repository (only its experts if `qwen-nvfp4-gptq` is installed: the other files are the same) and starts the model. `pack/mix.json` has the plan. - **Everything but the experts is the same as in the NVFP4 GPTQ repository**, byte for byte: the dense GGUF (Q8_0 projections; router, small tensors and norms BF16 / F32), Qwen's own FP8 PLE table, the BF16 embedding and the MTP draft head. - **The original model, censored, with a switch.** It refuses what Qwen refuses. The fork can remove the refusal direction at run time (layers 8-33, the weights untouched): `--uncensored on` at install, the web app's "Disable censorship" switch, or `"uncensored": true` in a request. Checked on this pack: 5 of 5 held-out harmful requests refused with it off, 0 of 5 with it on. - **Only for Strata NVFP4.** The files are Strata's runtime formats. They do not load in transformers, vLLM or llama.cpp.
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
> **Quantized by NVIDIA.** These are the weights of NVIDIA's checkpoint > `nvidia/Qwen3.8-Flash-Next-NVFP4` (NVIDIA Model > Optimizer), unchanged. This repository only converts their file format for Strata NVFP4. It is not made or endorsed by > NVIDIA. - **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 eve
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
Qwen's original Qwen3.8-Flash-Next with **every routed expert in Q8_0**, rounded from the BF16 checkpoint, packed for Strata NVFP4. It is the fork's accuracy reference: the other models' accuracy below is measured against it. It needs a PC with **128 GB of RAM**, and it decodes at about a quarter of the NVFP4 GPTQ model's speed. **Quick start (Windows):** from the Strata NVFP4 release 0.1.42-nvfp4.2 or newer, run `START-HERE.bat --family qwen-q8`: it checks the PC, downloads this repository (only its experts if `qwen-nvfp4-gptq` is installed: the other files are the same) and starts the model. Strata NVFP4 0.1.42-nvfp4.1 on an RTX 5090 (32 GB), Ryzen 9 9950X3D, 128 GB of DDR5-5600, the arguments under Run it.
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
Small dummy LLama2-type Model useable for Unit/Integration tests. Suitable for CPU only machines, see H2O LLM Studio for an example integration test. Use fixtures with session scope to load the model only once. This will decrease test runtime further.
翻訳待ち・原文を表示
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 画像
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト · 画像→ 出力 テキスト
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
**Maya-1 is an open-weight text-to-speech model from Maya Research that combines natural-language voice descriptions with inline expressive controls.** It generates speech as discrete audio tokens, which are decoded into a 24 kHz waveform. Developers can describe a voice and its delivery in text instead of choosing only from a fixed speaker menu. Voice description Female, in her 30s with an American accent and is an event host, energetic, clear diction This card documents the **Maya-1 checkpoint**. Other Maya Research models have separate documentation.
翻訳待ち・原文を表示
入力 テキスト→ 出力 音声
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 音声
Veena is a state-of-the-art neural text-to-speech (TTS) model developed by Maya Research, designed for English and Indian languages. Built on a Llama architecture backbone, Veena generates natural, expressive speech with emotional tone, remarkable quality, and ultra-low latency. It represents the foundation of our voice intelligence work, bringing human-like voice to the two most spoken languages in the world. **Veena** is a 3B parameter autoregressive transformer model based on the Llama architecture. It is designed to synthesize high-quality speech from text in Hindi and English, including code-mixed scenarios. The model outputs audio at a 24kHz sampling rate using the SNAC neural codec. To use Veena, you need to install the `transformers`, `torch`, `torchaudio`, `snac`, and `bitsandbytes` libraries.
翻訳待ち・原文を表示
入力 テキスト→ 出力 音声
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
分類準備中
入力 未確認→ 出力 未確認
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
分類準備中
入力 未確認→ 出力 未確認
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
分類準備中
入力 未確認→ 出力 未確認
基础档案已收录，详细介绍后续补充。可先查看厂商、分类与来源。
基本情報を収録・詳細は順次追加
入力 テキスト→ 出力 テキスト
