[bmdpat]
Download BMD free

Field journal

Blog archive

Every published note, with the newest work first. Use the topic links to stay inside one search intent.

  1. Local or API? Test the task before routing it

    Compare a local model with an API on the same task. Record settings, failed attempts, review time and cost before you decide where the workload belongs.

    #local-llm#hardware#ai-infrastructure
  2. Log Local LLM Fallbacks Before You Score the Output

    My local writer hit a host RAM guard, then Sonnet produced a draft that failed QA. Here is how I separate availability, model identity, and useful output.

    #local-llm#ollama#hardware
  3. Agent Memory: Test the Answer After a Correction

    Our agent-memory replay checks words in a file. I show what that proves, what it misses, and how to test whether an expert correction changes an answer.

    #ai-agents#agent-memory#local-llm
  4. My local daily brief used 0 model calls

    On September 4, 2026 my llama3.1:8b daily brief was ready with 0 model calls and 0 seconds. Preview-only file ranking is not a model run.

    #local-llm#ollama#hardware
  5. My agent wrote 126 empty pages and every gate passed

    One commit wrote 130 knowledge pages. 126 were the same four sentences with the title swapped. Schema checks, link checks and orphan checks all passed.

    #ai-agents#agent-management#verification
  6. Long Context Cost Me VRAM, Not Tokens a Second

    A 94x bigger prompt cost llama3.1:8b only 9.7% of its generation rate. VRAM rose 656 MiB, but num_ctx doubled in the same step. Budget memory first.

    #local-llm#ollama#hardware
  7. 3 Tests Before a GGUF Quant Runs Your Coding Agent

    A GGUF file fitting in VRAM does not prove it can run your coding agent. Use this local acceptance test for tools, patches, and repeat runs.

    #local-llm#quantization#ai-agents
  8. One model logged 14.5 and 6,178 prompt tokens a second

    gemma4:26b read 53 prompt tokens at 14.5 tok/s and 4,470 at 6,178.9 on the same RTX 5090 run set. Prompt speed is not one number.

    #local-llm#ollama#benchmarking
  9. Split Local LLM Prose and JSON Jobs by Model

    My local AI stack routes prose to Gemma and bounded JSON to Qwen. The split keeps model choice tied to the output contract, not one leaderboard.

    #local-llm#ollama#model-routing
  10. Ollama Load Time Can Hide a Fast Local LLM

    My RTX 5090 run spent 43.48 seconds on model load and 0.17 seconds on output. Split wall time before you judge local LLM speed.

    #local-llm#ollama#hardware
  11. A 32 GB GPU Still Needs Host RAM Headroom

    My RTX 5090 had 32 GB of VRAM ready, but my local policy stopped Ollama at 7.02 GiB of free host RAM. Here is my preflight before model load.

    #local-llm#ollama#hardware
  12. Your Local LLM CSV Needs a Schema Version

    I audited seven RTX 5090 benchmark CSVs and found four header widths, three power schemas, and no schema version. Here is the row contract I use now.

    #local-llm#benchmarking#hardware
  13. Your Benchmark Row Never Saved the Driver Version

    I audited seven RTX 5090 benchmark files on 2026-08-27. Not one records the GPU driver version. Here are the four fields every local LLM row needs.

    #local-llm#benchmarking#hardware
  14. Preflight Local AI Before You Benchmark a Model

    My local AI test stopped before model load because the runtime could not build. This preflight keeps setup failures out of benchmark results.

    #local-llm#inference#hardware
  15. 49W Average Hid a 338W Burst on Gemma 26B

    Gemma 4 26B Q4_K_M averaged 49 W on a long RTX 5090 run and peaked at 338 W. Keep both watt numbers before you compute energy per token.

    #local-llm#quantization#ollama
  16. How to Calculate Local LLM Energy per Token

    My RTX 5090 test shows how watts and output rate become joules per token, and why the faster of two matched settings can waste energy.

    #local-llm#model-evaluation#ollama
  17. Why a Failed Local LLM Benchmark Row Still Matters

    A failed local LLM row marks the test boundary. My RTX 5090 report shows why quality, speed, and settings belong in one receipt.

    #local-llm#model-evaluation#ollama
  18. Mojo Is Open Source. What Local AI Builders Need to Know

    Modular put Mojo under Apache 2.0 with LLVM exceptions on August 18, 2026. Here is what the release changes for local AI builders, and what it does not.

    #local-llm#open-source#ai-infrastructure
  19. My local LLM eval hid four token caps

    My 5090-rig eval uses four output caps: 50, 180, 192, and 256 tokens. A score without the cap is not a model result. Record the cap on every row.

    #local-llm#ollama#hardware
  20. A GPU Driver Is Not a Local LLM Benchmark

    My 2026-08-20 RTX 5090 report lists driver 610.88 and 17 benchmark rows. That does not show the driver changed model speed.

    #local-llm#hardware#llm-benchmarks
  21. The 26B Model Hit the Cap. The 8B Finished.

    Same code task, same 512-token cap. Ollama's done_reason showed one local model finished and one got cut off. Tokens per second hid it.

    #local-llm#ollama#benchmarking
  22. A 27B Model Fit on an 8 GB GPU. It Was Slow.

    Qwen3.8-27B Q3_K_S loaded on an RTX 3070. VRAM used 7,435 of 8,192 MiB. Decode ran 2.07 tok/s. Fit on 8 GB is not a usable rate.

    #local-llm#ollama#hardware
  23. My Local LLM Writer Failed Its Own Word-Count Gate

    Gemma 4 26B wrote a 535-word draft against a 550-word floor. The gate blocked it, both rescue engines were down, and no bad post shipped.

    #local-llm#ollama#hardware
  24. Unload Local LLMs After Every Test

    A local model test is not over when text appears. I unload the model, read idle VRAM, and record the result before I start another run.

    #local-llm#ollama#5090-reports