搜索“RAG”

safety without eroding the capabilities the earlier stages built. In addition to human preference and safety alignment, this stage also applies a reasoning-length penalty to discourage overly verbose reasoning behavior acquired during earlier stages. How the Three Sizes Differ Same method and infrastructure; the difference is how far up the ladder each model goes. 3B is a strong foundational-RL model; 8B and 30B add the agentic-RL block on top, learning to act with tools in real environments. Re

匹配字段正文

ntic analytics and intelligent agents. Chapters cover topics including: Evolving data strategy for agentic AI and scaling data products Building on existing infrastructure with a pragmatic, business-first approach Unlocking value with classical ML, semantic layers, and cross-team alignment Real-world perspectives from leaders across different industries Get your free digital copy today . Image source: Ideogram / The Rundown The Rundown: The Rundown Roundtable is a weekly feature where we poll me

匹配字段正文

In Figure 2, we show the final impact on latency by testing function calling on the BFCL dataset [4]. Across various multi-tool scenarios, DSpark reduces the latency by 57% on average. We use the same test setup as for LFM2.5-1.2B-Instruct. In Table 3, we present the throughput results for LFM2.5-1.2B-Instruct. Note that, in contrast to the 2.6B model, 1.2B is a non-reasoning model. For 1.2B, we observe much more variance in dataset acceptance rates, so speedup varies by as much as 52% dependin

匹配字段正文

ers, ends fading back to pure acoustic gentle close, heartfelt true emotion without over-dramatizing, subtle restrained Chinese-style parental love, proud yet tender unspoken encouragement, strong Taiwanese accent in vocals, emotional, sincere, warm parental perspective, builds to uplifting hopeful ending, soft whispered intimate outro like a gentle murmur calling home, nostalgic yet empowering family love song, seasoned experienced voices, no youthful bright tone More Natural Vocals Vocals are

匹配字段正文

3.7 Flash, Gemini Spark can turn ideas into action more efficiently by consolidating files, drafting emails, and updating status documents. We continually work to improve the coverage and robustness of Frontier Safety safeguards . Gemini 3.7 Flash is shipping with updated safeguards against misuse in the domains of Chemical, Biological, Radiological, and Nuclear (CBRN) and cyber offense, while enabling beneficial use cases, in accordance with our approach to bioresilience and our cyber program

匹配字段正文

es of our previous release with four major improvements: Screen/UI understanding. LFM2.5-VL-3B has a strong understanding of digital screens across mobile, web, and desktop. It averages 80.7 on ScreenSpot-v2, far ahead of the much larger Gemma-4-E4B (51.2) and Qwen 3.5 4B (78.5) and close behind the larger InternVL-3.5-4B (84.1) Function calling. New to our VL line, LFM2.5-VL-3B is significantly stronger at tool use and function calling, for both text-only and vision-text inputs. ToolSandbox mor

匹配字段正文

Technical Blog sessionStorage.removeItem('facets.search.text'); sessionStorage.removeItem('facets.search.text-copy'); sessionStorage.removeItem('facets.search.filters'); NVIDIA Nemotron 3.5 Lightning Delivers Fast, Accurate Specialized Task Execution for Long-Running Agents Aug 11, 2026 NVIDIA Nemotron 3.5 Lightning is an open 30B mixture-of-experts model with 3B active parameters optimized for high-volume, low-latency execution in long-running AI agents. The model includes speculative decoding

匹配字段正文

atency you measure is the server-side latency you actually control, with no managed-service round-trip in the number. Source: NVIDIA TTS NIM Performance documentation (v26.07), average of three trials, on-prem. TTFA = latency to first audio; RTFX = throughput as a multiple of real time. At 32ms on B200, Magpie's TTFA leaves the rest of the latency budget for ASR and LLM processing — keeping total end-to-end latency within the sub-200ms window natural conversation requires. Across NVIDIA GPUs, Ma

匹配字段正文

Technical Blog sessionStorage.removeItem('facets.search.text'); sessionStorage.removeItem('facets.search.text-copy'); sessionStorage.removeItem('facets.search.filters'); Aug 10, 2026 Muse Glimmer is a 30B open-weight dense model with a 120K+ context window built for local AI agentic work. The dense architecture activates every parameter for each token, delivering reliable instruction following, long-context coherence, predictable latency, and fewer failure modes for multi-step agentic workloads.

匹配字段正文

uctured outputs and how to fine-tune on images. We also experimented with running the model on OpenCode with AsyncGRPO example . Model shows strong coding capabilities, so we encourage you to try training with coding environments in OpenEnv and TRL. Demos Here are some fun ways to try out Muse Glimmer. In our opinion, the coolest thing about this model is that it is a local scale personal assistant that can code. That means you can make it do things like, quantize itself, find quantized weights

匹配字段正文

pect from day one. Running it well: Meta recommends using the following settings. temperature = 1.0, top p = 0.95, top k = 64 Reasoning effort is set in the system prompt as Reasoning strength: , with low, medium, high, and xhigh available - use high or xhigh for agentic and coding work. All benchmark numbers in the table above are at high. The era of agentic AI is here. We encourage you to start building today with Muse Glimmer model on Fireworks. Check out our documentation and start building!

匹配字段正文

ther details about FLUX 3 here. Video Models Must be Reality Models. Reality is inherently multimodal; but every snapshotrepresentation (e.g. image, video, sound) captures only a fragment of it. No single fragment is complete on its own. This is why FLUX 3 - a model designed to model reality as accurately as possible - is natively multimodal, and designed to model reality without collapsing into a particular uniform subset or style. The results are video outputs that are not limited to a cinemat

匹配字段正文

the safety model's quality. What's next Shieldstral is a step toward moderation that adapts to context instead of forcing every product through one frozen taxonomy. We're continuing to push on multilingual coverage, longer-document robustness, and broader multimodal safety — and we'd love to see what the community builds on top of it. BTW, we're hiring! If you want to help make AI better, see our careers page. Shieldstral Apache 2.0 Text and image safety classification against a supplied policy.

匹配字段正文

-Policy Distillation (MOPD), and Agentic Reinforcement Learning (Agentic RL). Supervised fine-tuning. Post-training begins with two consecutive SFT stages: starting with broad coverage across all domains, followed by targeted shaping on priority skills like agentic tasks, reasoning, and tool use. Across the two stages, the SFT training mix is about seven times the size of the one used for LFM2.5-8B-A1B, with heavier weighting toward agentic tasks such as tool use, web search, software engineerin

匹配字段正文

t reference, motion reference, and style reference; voice, sound effects, and music in audio generation were largely studied as separate domains; and video generation was further fragmented into text-to-video, image-to-video, first-and-last-frame, subject reference, motion reference, voice reference, video editing, and more, with clear boundaries between image, video, and audio as well. These siloed tasks, capabilities, and modalities constrained creative freedom in practice. And, on the trainin

匹配字段正文

while tests are still failing. Thinking mode makes the trade-off visible too. It lifts Terminal-Bench 2.1 from 60.4% to 70.2%, and DeepSWE from 16.5% to 40.4%, the latter at an average of 249,000 completion tokens per trajectory. In one 50-minute run, the model took 181 steps, without human intervention, to build an HTML and CSS rendering engine in JavaScript. It implemented a tokenizer, DOM, cascade, layout system and canvas renderer. The model has no vision capability yet, so it could not look

匹配字段正文

ditors":["eltonwilliams"],"editorAvatarUrls":["https://cdn-avatars.huggingface.co/v1/production/uploads/noauth/1Yp81w2J8OBsuzrHcQoZu.jpeg"],"reactions":[{"reaction":"🚀","users":["DragonFace","Nourivex"],"count":2}],"isReport":false}},{"id":"6a6897aba4402b1ff4994633","author":{" id":"67a4fea096628f7a5a4f8d21","avatarUrl":"https://cdn-avatars.huggingface.co/v1/production/uploads/no-auth/ 9MyBcbHOkpztwiNml71W.png","fullname":"Kiran Bhattacharyya","name":"kdb6df","type":"user","isPro":false,"isHf":f

匹配字段正文

e you'd get a low quality image should be dramatically reduced; Personalization will "understand" your personal taste much better. Especially if you have a ton of ratings under your profiles; Creating personalization profiles for V8.2 now has a much larger and improved pool of images to choose from Thanks for your help rating images to make this possible! We hope you all have fun and we strongly encourage you to try both new and old personalization profiles with the latest release. Published by:

匹配字段正文

5 is a clear step up from Opus 4.8 in both accuracy and efficiency. Its performance floor is materially higher, especially on deep finance domain logic. Across effort levels it averaged 9 percentage points higher accuracy with a third fewer turns and tool calls and 60% less time.Richard Pham Evals and Product Lead Claude Opus 5 checks its own work the way a real frontend developer would. On our benchmark it opened its pages in a browser at desktop and phone widths, caught a product hidden below

匹配字段正文

le of solving complex, real-world challenges. Microsoft AI partners directly with domain specialists to adapt and refine our models for their unique needs: MAI is partnering with Dragon Copilot, a solution used by 170,000 medical providers, which processed 28 million patient encounters last quarter.MAI-Transcribe-1.5 now supports Dragon Copilot’s multilingual workflow, replacing the previous model with our own best-in-class model across 58 languages. In internal evaluations on multilingual recor

匹配字段正文