What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
In an August 29, 2026 report to the llama.cpp issue tracker, a user found that llama-server returned logprob: 0.0 and empty top_logprobs arrays for most generated tokens when speculative decoding was enabled. Those values are not evidence that the model was certain: a log probability of 0 corresponds to probability 1, while an empty alternatives list contains no requested top-token data. Treat this as a reported bug signature, not a proven problem in every build.
What the reported logprobs mean
The issue author’s example compared a speculative run using a Gemma 4 target, an MTP drafter, and --spec-type draft-mtp with a run after removing the drafter flags. In the speculative run, 23 of 24 generated tokens reportedly had a logprob of 0.0 and empty top alternatives. In the comparison, none of 24 tokens had a logprob of 0.0, and the displayed samples included three alternatives. The speculative run also reported draft_n_accepted=14 out of draft_n=24. These are counts from one user’s example, not a benchmark or a guarantee about other configurations. Issue #27972
Mathematically, log(1.0) = 0. So a returned logprob of zero implies probability 1 if interpreted literally. But paired with an empty top_logprobs array, it is a suspicious combination: the response lacks the requested alternatives, and the zero should not be read as genuine certainty.
Why speculative decoding may produce placeholder values
Speculative decoding lets a draft model propose multiple tokens, which the target model then verifies in a batch. The llama.cpp speculative decoding documentation describes this process and notes support for multiple implementations in llama-server.
#1 Best Overall
- 【High-Performance APU】The MS-S1 MAX features an AMD Ryzen AI Max+ 395 APU, integrating a Zen 5 architecture CPU (up to 5.1GHz, 16C/32T, 64M L3 Cache), an RDNA 3.5 GPU, and an NPU (50 TOPS). The total system output is 126 TOPS. It provides powerful parallel computing capabilities for demanding AI workflows. It is ideal for running local LLMs, multimodal models, and computationally intensive tasks
- 【128GB UMA Memory】Equipped with up to 128GB of LPDDR5x-8000MT/s unified memory, it enables the CPU and GPU to access a shared, high-bandwidth memory pool with extremely low latency. Ideal for large-scale AI inference, 3D workloads, and complex timelines in video editing. It eliminates traditional VRAM bottlenecks, ensuring smoother data transfer during high-intensity computations. The UMA design maximizes performance stability under high loads
- 【Flexible Expansion】The MS-S1 MAX features USB4 V2 (up to 80Gbps), dual 10GbE LAN, HDMI 2.1 (up to 8K60), a full-length PCIe x16 expansion slot, and dual M.2 slots supporting up to 16TB RAID 0/1. Wi-Fi 7 provides stronger signal coverage and a more stable wireless experience. The slide-out design facilitates upgrades and maintenance. It easily adapts to personal, studio, or rack-mount enterprise environments
- 【High-Efficiency Cooling System】Utilizing an aerospace-grade aluminum alloy chassis, copper base plate, six heat pipes, dual turbine fans, and advanced PCM thermal conductive material, it maintains stable cooling performance even under continuous load. This system supports 130W continuous power and 160W peak power operation, with a built-in 320W power supply. It boasts multiple global certifications including CCC, FCC, UL, CE, and UKCA, ensuring stable and reliable operation in various environments
- 【Cluster Design】Two MS-S1 MAX units can be configured as a dual-unit cluster to run a large 235B Q4 model locally, achieving an output speed of 10.87 tok/s. Supporting 2U rack deployment, multiple MS-S1 MAX units can be cascaded into a distributed cluster to create a high-efficiency AI computing center. A cluster of four MS-S1 MAX units successfully ran a DeepSeek-R1 671B Q4 large model. A reserved cluster power-on interface allows for unified start-up and shutdown
The issue author’s code-path analysis points to the speculative “main model sample and accept” path: it reportedly sets result.prob = 1.0f but does not fill result.probs before the response is serialized. That would explain a zero logprob and missing alternatives. The issue quotes a nearby code comment, // TODO: set result.probs. This is the reporter’s explanation of the implicated path, not an independently verified analysis of every commit or speculative mode. Issue #27972
What the server API is supposed to return
The llama-server README describes n_probs as requesting the top N token probabilities for each generated token. It also documents post_sampling_probs, which reports probabilities after the sampling chain; in that mode, the response fields are named prob and top_probs rather than logprob and top_logprobs. These documented semantics describe the intended response, but do not establish that every release correctly populates the fields in every speculative path.
Rank #2
- [3352 AI TOPS, 5th Gen Tensor Cores, AI Content Creation] Built for AI-assisted photo and video workflows including upscaling, denoise, background removal, masking, and generative AI creation for faster creator productivity.
- [32GB GDDR7 VRAM, Local LLM Inference, Larger Models] Run local LLM inference and on-device AI tools with massive VRAM headroom for larger models, longer context, and heavier multitasking across AI and creator apps.
- [28 Gbps, 512-bit, 21760 CUDA Cores] High-throughput next-gen memory and core resources for demanding creator projects, complex timelines, large assets, and GPU-accelerated ML experimentation and inference pipelines.
- [Quad-Fan Force, Vapor Chamber, Phase-Change Thermal Pad] Designed for sustained performance under heavy loads with quad-fan cooling, a patented vapor chamber, and a phase-change GPU thermal pad to help lower temps and reduce hotspots.
- [DP 2.1b x3, HDMI 2.1b x2, Bundle GPU Holder] Multi-display ready with up to 4 displays and up to 7680 x 4320 max digital resolution, plus an included GPU Holder to help reduce GPU sag and improve long-term build stability.
How to check whether your build is affected
- Record the exact setup. Note the
llama.cppversion and commit, target model, speculative implementation and flags, sampling parameters, requestedn_probs, and returned per-token probability fields. - Make a controlled comparison. Send otherwise identical requests to the same server build with speculative decoding enabled and disabled. Change only the speculative settings; keep the prompt, target model, sampling configuration, and requested top-N count the same.
- Inspect the response token by token. Check whether speculative output pairs
logprob: 0.0with emptytop_logprobs, while the non-speculative response contains populated alternatives and varying probabilities. Plausible generated text alone does not validate the probability fields. - Keep acceptance data in context. If the response reports draft and accepted-token counts, record them alongside the probabilities. Acceptance indicates that speculative decoding is occurring; it does not demonstrate that the returned logprobs are correct.
This is a diagnostic based on the issue author’s enabled-versus-disabled comparison, not a reproduction across builds. The issue author characterizes the implicated path as generic and the particular model pairing as incidental, but that does not independently establish the same behavior in every model, speculative mode, or release.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Fix the driver behind crashes, sound loss and screen glitches3Clear out junk files and repair common Windows errorsShould you use these values for evaluation?
If your build shows the reported signature, do not use those probability fields for perplexity, confidence thresholds, model rankings, or other likelihood-based scoring until you have verified the implementation or confirmed correct results for that exact build. A response can be syntactically well-formed and still contain placeholder values.
Rank #3
- System Compatibility Note: This 2-slot card measures 271 x 112 x 39 mm and requires a single 12V-2x6-pin power connector. Please verify chassis and PSU compatibility before purchase.
- Dedicated Support: Please contact us directly through Amazon for any product questions or assistance you may require.
- Professional Intel Arc Pro B70 GPU: Built on the Intel Xe2-HPG architecture, it features 32 Xe cores and 256 XMX engines, designed to accelerate AI, rendering, and complex visualization workloads.
- Massive 32GB GDDR6 VRAM: Equipped with 32GB of high-speed GDDR6 memory on a 256-bit bus, running at 19 Gbps, which allows for handling large AI models and complex datasets locally.
- High-Performance Engine Clock: Delivers an engine clock of 2540 MHz, providing the compute power needed for demanding professional applications and AI inference.
Issue #27972 was opened on August 29, 2026. The issue page showed it as open, without an assignee or linked branch or pull request when checked; that page status does not establish whether a fix has since merged or whether a later or unreleased build is affected. Verify the behavior against the exact commit you deploy rather than assuming either that all current builds are broken or that the issue is resolved. Issue #27972
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
