11.3 hours of 1971 Nixon White House tapes, transcribed in under 3 minutes on one GPU.
NVIDIA's Nemotron 3.5 Transcribe on vLLM (NVFP4 checkpoint), every file sent at once on Hugging Face Jobs. About 3 cents per hour of audio.
huggingface.co/datasets/uv-scripts/transcription
NVIDIA's Nemotron 3.5 Transcribe on vLLM (NVFP4 checkpoint), every file sent at once on Hugging Face Jobs. About 3 cents per hour of audio.
huggingface.co/datasets/uv-scripts/transcription
November 5, 2026 at 9:04 AM
11.3 hours of 1971 Nixon White House tapes, transcribed in under 3 minutes on one GPU.
NVIDIA's Nemotron 3.5 Transcribe on vLLM (NVFP4 checkpoint), every file sent at once on Hugging Face Jobs. About 3 cents per hour of audio.
huggingface.co/datasets/uv-scripts/transcription
NVIDIA's Nemotron 3.5 Transcribe on vLLM (NVFP4 checkpoint), every file sent at once on Hugging Face Jobs. About 3 cents per hour of audio.
huggingface.co/datasets/uv-scripts/transcription
Hydrogen 1.0 🔬 Impressive open model built for scientific discovery
huggingface.co/deep-princip...
- Based on GLM 5.3
- 1M context window
- MIT license + with BF16, NVFP4,MXFP4 variants
huggingface.co/deep-princip...
- Based on GLM 5.3
- 1M context window
- MIT license + with BF16, NVFP4,MXFP4 variants
November 4, 2026 at 8:49 PM
Hydrogen 1.0 🔬 Impressive open model built for scientific discovery
huggingface.co/deep-princip...
- Based on GLM 5.3
- 1M context window
- MIT license + with BF16, NVFP4,MXFP4 variants
huggingface.co/deep-princip...
- Based on GLM 5.3
- 1M context window
- MIT license + with BF16, NVFP4,MXFP4 variants
NVIDIA's DeepSeek-V4.1-Flash NVFP4 Model: Everything You Need to Know
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window.
#hackernews #news #nvidia
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window.
#hackernews #news #nvidia
NVIDIA's DeepSeek-V4.1-Flash NVFP4 Model: Everything You Need to Know
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window.
hackernoon.com
November 5, 2026 at 9:17 AM
NVIDIA's DeepSeek-V4.1-Flash NVFP4 Model: Everything You Need to Know
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window.
#hackernews #news #nvidia
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window.
#hackernews #news #nvidia
QLoRA you know: train quantized - can be a lot faster on modern GPUs which grok NVFP4 precision, but way less stable / accurate.
DyLoRA: train ranks 1, 2, 4, ... *at the same time*, Matroyshka-style, 'cuz the fwd pass over base dominates, extract and pick best, after.
QDyLoRA: both at same time
DyLoRA: train ranks 1, 2, 4, ... *at the same time*, Matroyshka-style, 'cuz the fwd pass over base dominates, extract and pick best, after.
QDyLoRA: both at same time
November 11, 2026 at 1:22 AM
QLoRA you know: train quantized - can be a lot faster on modern GPUs which grok NVFP4 precision, but way less stable / accurate.
DyLoRA: train ranks 1, 2, 4, ... *at the same time*, Matroyshka-style, 'cuz the fwd pass over base dominates, extract and pick best, after.
QDyLoRA: both at same time
DyLoRA: train ranks 1, 2, 4, ... *at the same time*, Matroyshka-style, 'cuz the fwd pass over base dominates, extract and pick best, after.
QDyLoRA: both at same time
NVIDIA: ModelOpt 0.47.1 fixes NVFP4 fake quantization with Triton 3.9 on GPUs with compute capability 8.9 or newer
November 6, 2026 at 10:45 PM
NVIDIA: ModelOpt 0.47.1 fixes NVFP4 fake quantization with Triton 3.9 on GPUs with compute capability 8.9 or newer
yeah I’d give it a look, the NVFP4 variant they followed with after the initial Nemotron 3 Nano release really raised my eyebrows
you won’t mistake it for Opus but I think it could do lots of interesting agent grunt work and the speed on an M4 Max was pretty sick
you won’t mistake it for Opus but I think it could do lots of interesting agent grunt work and the speed on an M4 Max was pretty sick
November 8, 2026 at 5:49 PM
yeah I’d give it a look, the NVFP4 variant they followed with after the initial Nemotron 3 Nano release really raised my eyebrows
you won’t mistake it for Opus but I think it could do lots of interesting agent grunt work and the speed on an M4 Max was pretty sick
you won’t mistake it for Opus but I think it could do lots of interesting agent grunt work and the speed on an M4 Max was pretty sick
One RTX 5090 capped at 450 W runs stock Qwen3.8 27B (all-NVFP4, NInfer + DFlash2 speculative decoding) at 400 tok/s writing a module + tests, 326 on a code refactor, 165 on prose. Prefill 12,908 tok/s at 8K. Owner-measured 7 Nov.
getrigbase.com/builds/rtx-5...
getrigbase.com/builds/rtx-5...
November 11, 2026 at 7:23 AM
One RTX 5090 capped at 450 W runs stock Qwen3.8 27B (all-NVFP4, NInfer + DFlash2 speculative decoding) at 400 tok/s writing a module + tests, 326 on a code refactor, 165 on prose. Prefill 12,908 tok/s at 8K. Owner-measured 7 Nov.
getrigbase.com/builds/rtx-5...
getrigbase.com/builds/rtx-5...
In the coming days, we will provide BF16, FP8 and NVFP4 quantization for this model, which will be available for free at concept.country
Make sure to follow @conceptcountry for updates.
Make sure to follow @conceptcountry for updates.
Concept Country | Governance & a Commons for AI Agents
Explore Concept Country, an experimental digital country for people and AI agents. Study governance, inspect world data, and discover the Commons.
concept.country
November 6, 2026 at 2:25 AM
In the coming days, we will provide BF16, FP8 and NVFP4 quantization for this model, which will be available for free at concept.country
Make sure to follow @conceptcountry for updates.
Make sure to follow @conceptcountry for updates.
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, offering both a quantized NVFP4 variant and a full‑precision BF16 model for multimodal AI workloads.
NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, offering both a quantized NVFP4 variant and a full‑precision BF16 model for multimodal AI workloads.
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, offering both a quantized NVFP4 variant and a full‑precision BF16 model for multimodal AI workloads.
aiunzipped.com
November 10, 2026 at 6:00 AM
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, offering both a quantized NVFP4 variant and a full‑precision BF16 model for multimodal AI workloads.
NVIDIA Nemotron 3.5 Super VL is now available on Amazon SageMaker JumpStart, offering both a quantized NVFP4 variant and a full‑precision BF16 model for multimodal AI workloads.
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
The key-value (KV) cache dominates memory capacity and bandwidth in long-context LLM serving. While 4-bit microscaling formats (NVFP4, MXFP4) shrink the cache and run attention on FP4 Tensor Cores, a fundamental gap remains between algorithmic quantization and system-aware acceleration: existing FP…
arxiv.org
November 10, 2026 at 8:32 AM
FLUX 3 Image Klein hits 3.46x faster image generation on a single B200. NVIDIA's TensorRT LLM VisualGen combines NVFP4 quantization, fused kernels, and multi-GPU parallelism to cut diffusion latency.
FLUX 3 Image Klein hits 3.46x faster image generation on a single B200 - GenAInews
NVIDIA's TensorRT LLM VisualGen combines NVFP4 quantization, fused kernels, and multi-GPU parallelism to cut diffusion latency.
genainews.tech
November 7, 2026 at 12:15 AM
FLUX 3 Image Klein hits 3.46x faster image generation on a single B200. NVIDIA's TensorRT LLM VisualGen combines NVFP4 quantization, fused kernels, and multi-GPU parallelism to cut diffusion latency.
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
The key-value (KV) cache dominates memory capacity and bandwidth in long-context LLM serving. While 4-bit microscaling formats (NVFP4, MXFP4) shrink the cache and run attention on FP4 Tensor Cores, a fundamental gap remains between algorithmic quantization and system-aware acceleration: existing FP…
arxiv.org
November 9, 2026 at 11:12 AM
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
Dayou Du et al.
#arXiv #cs.AI #cs.AR #cs.CL #cs.PF
BitDecoding-FP4: Native FP4 KV Cache Decoding via Algorithm-System Co-Design
The key-value (KV) cache dominates memory capacity and bandwidth in long-context LLM serving. While 4-bit microscaling formats (NVFP4, MXFP4) shrink the cache and run attention on FP4 Tensor Cores, a fundamental gap remains between algorithmic quantization and system-aware acceleration: existing FP…
arxiv.org
November 9, 2026 at 8:11 AM
📰 New article by Venu Kanamatareddy, Evan Kravitz, Naidile Murali
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
#AWS #AI #MachineLearning
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
#AWS #AI #MachineLearning
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, image, and video inference for document intelligence, computer-use agents, and multimodal agentic workloads.
aws.amazon.com
November 9, 2026 at 8:11 PM
📰 New article by Venu Kanamatareddy, Evan Kravitz, Naidile Murali
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
#AWS #AI #MachineLearning
NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart
#AWS #AI #MachineLearning
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation
Yining Wei et al.
#arXiv #cs.AI #cs.LG
Yining Wei et al.
#arXiv #cs.AI #cs.LG
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation
Reinforcement learning (RL) post-training increasingly targets low-bit deployment, with NVFP4 weight-and activation quantization (W4A4) becoming the state-of-the-practice format. However, integrating NVFP4 into RL degrades rollout quality and amplifies cross-engine training-inference mismatch (TIM)…
arxiv.org
November 4, 2026 at 10:44 PM
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation
Yining Wei et al.
#arXiv #cs.AI #cs.LG
Yining Wei et al.
#arXiv #cs.AI #cs.LG
FourSight: Stable FP4 RL Post-Training via Sequence Filtering and Sparse Channel Compensation
Reinforcement learning (RL) post-training increasingly targets low-bit deployment, with NVFP4 weight-and activation quantization (W4A4) becoming the state-of-the-practice format. However, integrating NVFP4 into RL degrades rollout quality and amplifies cross-engine training-inference mismatch (TIM)…
arxiv.org
November 4, 2026 at 9:19 PM
🤖 **NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart**
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, imag...
📰 Source […]
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, imag...
📰 Source […]
Original post on igeek.gamer-geek-news.com
igeek.gamer-geek-news.com
November 9, 2026 at 8:11 PM
🤖 **NVIDIA Nemotron 3.5 Super VL is now available in Amazon SageMaker JumpStart**
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, imag...
📰 Source […]
NVIDIA Nemotron 3.5 Super VL, a 120B-parameter multimodal vision-language model, is now available in Amazon SageMaker JumpStart. Deploy the NVFP4 or BF16 variant in a few steps, then run text, imag...
📰 Source […]
Cloud Codes on Ripple-65B on NVIDIA DGX Spark: full-precision weights are "129.6 GB" vs 128 GB shared; NVFP4 is "59.2 GB" but "reads every weight for every token". Useful if fit is not speed. Their math: "near 4.6 tokens a second". Would bandwidth decide?
www.youtube.com/watch?v=zxco...
www.youtube.com/watch?v=zxco...
November 8, 2026 at 11:30 PM
Cloud Codes on Ripple-65B on NVIDIA DGX Spark: full-precision weights are "129.6 GB" vs 128 GB shared; NVFP4 is "59.2 GB" but "reads every weight for every token". Useful if fit is not speed. Their math: "near 4.6 tokens a second". Would bandwidth decide?
www.youtube.com/watch?v=zxco...
www.youtube.com/watch?v=zxco...
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window. #machinelearning
NVIDIA's DeepSeek-V4.1-Flash NVFP4 Model: Everything You Need to Know
hackernoon.com
November 4, 2026 at 8:00 AM
Discover NVIDIA's quantized DeepSeek-V4.1-Flash NVFP4 model, featuring 552B parameters, multimodal inputs, and a 1M-token context window. #machinelearning
ASpec trains a draft model to get rejected at key steps, resetting verification for higher precision. Authors report +7 DeepSWE points for GLM-5.3-NVFP4; DeepSeek-V4.1-Flash’s +4 has a 95% CI spanning zero. Five evaluation runs per setup. arxiv.org/abs/2611.05345
November 5, 2026 at 2:38 AM
ASpec trains a draft model to get rejected at key steps, resetting verification for higher precision. Authors report +7 DeepSWE points for GLM-5.3-NVFP4; DeepSeek-V4.1-Flash’s +4 has a 95% CI spanning zero. Five evaluation runs per setup. arxiv.org/abs/2611.05345
NVIDIA Nemotron 3.5 Super VL arrives in Amazon SageMaker JumpStart. The 120B-parameter vision-language model accepts images, video, and text and supports context windows up to 1 million tokens. Amazon SageMaker JumpStart now offers the model in NVFP4 and BF16 variants for deployment.
NVIDIA Nemotron 3.5 Super VL arrives in Amazon SageMaker JumpStart - GenAInews
The 120B-parameter vision-language model accepts images, video, and text and supports context windows up to 1 million tokens.
genainews.tech
November 9, 2026 at 8:06 PM
NVIDIA Nemotron 3.5 Super VL arrives in Amazon SageMaker JumpStart. The 120B-parameter vision-language model accepts images, video, and text and supports context windows up to 1 million tokens. Amazon SageMaker JumpStart now offers the model in NVFP4 and BF16 variants for deployment.
Qwen4 27B、昨日リリースされると告知されて、まだ出てないことにヤキモキしてる界隈を観察してるアカウントがこちらです。楽し嬉しそうでなにより。
MoEじゃなくてdenseなので、下々のものたちの元に降りてくるのはしばらく掛かりそうではある
NVFP4で14GBだと言われてるので、多くの環境で動きそう
MoEじゃなくてdenseなので、下々のものたちの元に降りてくるのはしばらく掛かりそうではある
NVFP4で14GBだと言われてるので、多くの環境で動きそう
November 10, 2026 at 3:17 AM
Qwen4 27B、昨日リリースされると告知されて、まだ出てないことにヤキモキしてる界隈を観察してるアカウントがこちらです。楽し嬉しそうでなにより。
MoEじゃなくてdenseなので、下々のものたちの元に降りてくるのはしばらく掛かりそうではある
NVFP4で14GBだと言われてるので、多くの環境で動きそう
MoEじゃなくてdenseなので、下々のものたちの元に降りてくるのはしばらく掛かりそうではある
NVFP4で14GBだと言われてるので、多くの環境で動きそう
NVIDIA clocked FLUX 3 Image Klein at 2048×2048 on one B200.
3.46x faster text-to-image than uncompiled BF16.
NVFP4 is 1.62x of that. The rest is kernels and caching.
None of it shows up in a per-image price list.
You're not buying a model. You're buying a stack.
3.46x faster text-to-image than uncompiled BF16.
NVFP4 is 1.62x of that. The rest is kernels and caching.
None of it shows up in a per-image price list.
You're not buying a model. You're buying a stack.
November 7, 2026 at 12:18 AM
NVIDIA clocked FLUX 3 Image Klein at 2048×2048 on one B200.
3.46x faster text-to-image than uncompiled BF16.
NVFP4 is 1.62x of that. The rest is kernels and caching.
None of it shows up in a per-image price list.
You're not buying a model. You're buying a stack.
3.46x faster text-to-image than uncompiled BF16.
NVFP4 is 1.62x of that. The rest is kernels and caching.
None of it shows up in a per-image price list.
You're not buying a model. You're buying a stack.