Results for “fp”

Quantize diffusion transformers with NVIDIA ModelOpt, using FP8 or NVFP4 lower-precision numbers, and prepare the saved weights for SGLang Diffusion. Follow a workflow for loading, checking output quality, and comparing speed under matching conditions.

• Preparing ModelOpt FP8 weights for SGLang Diffusion
• Checking quality after diffusion model quantization
• Comparing BF16 and quantized generation speed

RFP creation, distribution, evaluation, and shortlisting for in-house legal teams selecting outside counsel. Draft a new legal services RFP from scratch, evaluate firm responses against weighted criteria, produce a shortlist recommendation with selection rationale for GC sign-off, or design the end-to-end RFP process from scratch. Trigger on: 'draft an RFP', 'run an RFP process', 'evaluate firm responses', 'score the RFP submissions', 'which firms should we shortlist', 'selection recommendation', 'how do we run a panel RFP', 'RFP for legal services', 'pitch process', 'firm selection process', 'compare the proposals', 'we're reviewing our panel', 'panel refresh RFP', 'we need to go to market', 'request for proposal', 'evaluate the pitches', 'which firm won the RFP', 'write a legal RFP', 'RFP evaluation criteria', 'design the RFP process', 'how do we weight the criteria'.

The English summary is being prepared. Showing the original description.

Guides large language model reinforcement learning with miles, covering environment setup, low-precision training, consistency between training and inference, and speculative decoding, where a smaller model proposes tokens for a larger model to verify.

• Configure reinforcement learning for large MoE models
• Investigate instability in FP8 training
• Set up speculative decoding for faster rollout generation

Structure responses to buyer questionnaires and proposal requests by classifying requirements, matching claims to verifiable evidence, and exposing gaps. Develop buyer-focused reasons to choose your offer and estimate win probability to support bid decisions.

• Classify requirements in an RFP
• Map questionnaire answers to supporting evidence
• Review missing capabilities before bidding

Turn browser screenshots captured during real interactions into a 60 fps, 4:3 4K demo video. Add a natural macOS cursor, timed clicks, and gentle zooms, then verify video properties and a sample frame.

• Creating a browser-based product demo
• Showing clicks and zooms in a UI walkthrough
• Re-recording a browser interaction sequence

Check a Proofpack release candidate before packaging by running tests, validating the candidate, and inspecting the package locally. Compare the source state and command results with the evidence log, then report passed, failed, or unknown checks.

• Checking a Proofpack candidate before packaging
• Comparing local verification with recorded evidence
• Finding unverified checks or uncovered changes

Integrate Flash Attention into Transformer models to speed up attention, the calculation that relates input elements, while reducing GPU memory use. Covers PyTorch integration, performance measurement, output comparison, and a separate H100 optimization path.

• Optimize training with long input sequences
• Reduce memory use during long-context inference
• Measure speed and compare model outputs

Route pricing, discount, partnership, sales channel, proposal response, and bookings forecast questions to one of seven specialist skills. Receive a concise digest of findings, recommended actions, and output files to support commercial decisions and human approval.

• Review pricing models and product packages
• Assess deal discounts and approval requirements
• Compare partnership terms and channel profitability

Draft responses to requests for proposals and sales questionnaires using approved past answers and product documentation. Organize answers in Google Docs, flag uncertain questions, and prepare an unsent Gmail reply for human review and submission.

• Preparing a first draft of a sales proposal questionnaire
• Reusing approved answers about pricing and support
• Identifying questions that need human review

Guides large language model reinforcement learning with miles, including MoE models that route work to specialized components. Covers low-precision training, matching training and inference behavior, and accelerating training response generation with draft models.

• Setting up reinforcement learning for large MoE models
• Configuring FP8 or INT4 training
• Checking consistency between training and inference

Troubleshoot PDF creation by trying pandoc, then reportlab, then fpdf2. Check tool availability, install missing Python libraries as needed, run generation with an explicit working directory, and verify that the output file exists.

• Converting Markdown or HTML to PDF
• Trying another tool after PDF generation fails
• Checking PDF tools and installing missing libraries

Use when working on imp's quantization formats, loaders, or dequant paths - GGUF Q4_0...Q8_0/Q*_K/IQ4/MXFP4, NVFP4 two-level scaling (Modelopt vs compressed-tensors layouts), FP8 E4M3, StorageTier, decode cache, KV-cache dtypes (auto/fp16/fp8/int8/int4/nvfp4/mxfp4), "which quant should I use", "which KV dtype", scale-factor layout, dequant kernel wiring, imp-quantize / AWQ, judging quant quality (PPL). Do NOT use for writing/optimizing GEMM/GEMV kernels (sm120-cuda-expert) or measuring quant perf (benchmark-cuda).

The English summary is being prepared. Showing the original description.

Configure LLM inference and serving on NVIDIA GPUs with TensorRT-LLM. Covers quantization, which reduces numerical precision, batching requests during generation, and distributing models across GPUs to improve processing capacity and response times.

• Setting up an LLM server on NVIDIA GPUs
• Configuring FP8 or INT4 quantization
• Distributing large models across multiple GPUs

Review proposal responses for missing requirements, compare competing products, and plan proof-of-concept trials that test customer needs. Supports technical sales work from requirement discovery and demo preparation through evaluation scorecards and technical proposals.

• Checking proposal responses for missing requirements
• Comparing product features against competitors
• Preparing a demo around customer priorities

Configure LLM inference and serving on NVIDIA GPUs with TensorRT-LLM. Covers quantization, which reduces numerical precision, batching requests during generation, and distributing models across GPUs to improve processing capacity and response times.

• Setting up an LLM server on NVIDIA GPUs
• Configuring FP8 or INT4 quantization
• Distributing large models across multiple GPUs

Optimize transformer attention, the calculation that relates tokens to one another, using Flash Attention. The skill guides PyTorch integration, GPU memory reduction, performance measurement, and output comparisons for training or inference with long sequences.

• Reducing GPU memory use with long sequences
• Integrating Flash Attention into PyTorch models
• Comparing runtime and outputs before and after changes

Extract PDF text when read_file returns unreadable binary content or little usable output. Try pdftotext with layout preservation, then use pdfplumber as an alternative when the command is unavailable or produces insufficient text.

• Extracting text after read_file returns binary content
• Retrying PDF extraction with empty or very short output
• Extracting text while preserving the PDF layout

Trigger for: FFP IGCE, firm-fixed-price estimate, FFP cost model, proposed FFP rate validation, wrap-rate analysis, Agency BPA rate comparison, price-reasonableness memo, or fair-and-reasonable analysis. Build auditable Firm-Fixed-Price federal estimates using BLS OEWS wages, layered fringe/overhead/G&A/profit, GSA CALC+ positioning, and GSA Per Diem travel. Use for FFP-by-period, FFP-by-deliverable, SOW/PWS decomposition, implied-multiplier analysis, and fixed-price scenario comparisons. Do NOT use for Labor Hour, T&M, or cost-reimbursement IGCEs (use IGCE Builder LH/T&M or IGCE Builder CR). Do NOT use for grant budgets. Requires the bls-oews, gsa-calc, and gsa-perdiem MCP servers.

The English summary is being prepared. Showing the original description.

Drafts answers to the questions in a received RFP, RFQ or tender from the organisation's past responses, answer library, case studies, policies and other documents provided: one row per buyer question with a draft answer marked Reused, Adapted or Missing, its source and date, and a legal or pricing review flag on every answer touching terms, liability, warranties, insurance, data protection or price. Use when the user asks to "draft our RFP response", "answer this tender from our past proposals", "pre-fill the bid questions", "first pass at the RFQ answers" or "reuse our previous bid content". Do not use for writing the buyer's own RFP or requirements, use rfp-requirements-pack instead; for security or due diligence questionnaires use vendor-security-questionnaire-prefill; for scoring received supplier responses use rfp-comparison-pack. Drafts for human review; never approves, authorises or signs off.

The English summary is being prepared. Showing the original description.

Use when the user asks about speakers, talks, schedules, or CFP dates for any Devoxx or VoxxedDays conference. Covers: speaker bios/socials, talk topics/tracks/levels, daily program or room agenda, most popular talks, upcoming or past events, and CFP open/close dates. Queries the public CFP API at [event-slug].cfp.dev and the devoxxians.com event registry. Example triggers: who speaks about Kubernetes at Devoxx UK, what is on Wednesday in Room 8, find AI talks at Voxxed Zurich, when does the CFP open for Devoxx Belgium.

The English summary is being prepared. Showing the original description.

Build a single HTML visualizer for particles, fractals, simulations, or data. It moves heavy calculations into a Web Worker, a background browser process, and draws on a canvas while targeting smooth 60fps rendering.

• Prototyping a particle or physics simulation
• Visualizing Mandelbrot or Julia fractals
• Building an audio or data visualizer

Expert blueprint for First-Person Shooters: fps_camera_look, fps_movement_logic, hitscan_weapon_query, weapon_bobbing_system, procedural recoil, and FPS NEVER rules. Shared TPS/cover theory links to godot-genre-shooter. Keywords: FPS, movement physics, weapon bobbing, camera sway, hitscan, viewmodel, air control.

The English summary is being prepared. Showing the original description.

Expert blueprint for First-Person Shooters: fps_camera_look, fps_movement_logic, hitscan_weapon_query, weapon_bobbing_system, procedural recoil, and FPS NEVER rules. Shared TPS/cover theory links to godot-genre-shooter. Keywords: FPS, movement physics, weapon bobbing, camera sway, hitscan, viewmodel, air control.

The English summary is being prepared. Showing the original description.

Use when the user wants to write text in a specific 16-minds (MBTI) personality's voice — articles, blog posts, essays, SNS投稿, replies, or rewriting existing AI-generated text to feel less AI-like. Triggers on Japanese phrases like 「ENFPで書いて」「INTJ風の記事」「人間っぽく書き直して」「AI臭抜いて」, English equivalents like "write as INFJ" / "humanize", or any 16-minds command with a `--write` / `--rewrite` flag. Personality (one of intj/intp/entj/entp/infj/infp/enfj/enfp/istj/isfj/estj/esfj/istp/isfp/estp/esfp) must be specified or inferable.

The English summary is being prepared. Showing the original description.