DEV Community

Patrick Hughes profile picture

Patrick Hughes

404 bio not found

Joined Joined on  github website
A 32 GB GPU Still Needs Host RAM Headroom
pat9000 profile

A 32 GB GPU Still Needs Host RAM Headroom

#localai #localllm #ollama #rtx5090
Comments
5 min read

Want to connect with Patrick Hughes?

Create an account to connect with Patrick Hughes. You can also sign in below to proceed if you already have an account.

Create Account
Already have an account? Sign in
Your Benchmark Row Never Saved the Driver Version
pat9000 profile

Your Benchmark Row Never Saved the Driver Version

#localai #localllm #benchmarking #rtx5090
Comments
5 min read
Your Local LLM CSV Needs a Schema Version
pat9000 profile

Your Local LLM CSV Needs a Schema Version

#localai #localllm #benchmarking #rtx5090
Comments
5 min read
Preflight Local AI Before You Benchmark a Model
pat9000 profile

Preflight Local AI Before You Benchmark a Model

#localai #localllm #inference #rtx5090
Comments
5 min read
49W Average Hid a 338W Burst on Gemma 26B
pat9000 profile

49W Average Hid a 338W Burst on Gemma 26B

#localllm #gguf #ollama #rtx5090
Comments
5 min read
How to Calculate Local LLM Energy per Token
pat9000 profile

How to Calculate Local LLM Energy per Token

#localllm #modelevaluation #ollama #rtx5090
Comments
5 min read
Why a Failed Local LLM Benchmark Row Still Matters
pat9000 profile

Why a Failed Local LLM Benchmark Row Still Matters

#localllm #modelevaluation #ollama #rtx5090
Comments
4 min read
My local LLM eval hid four token caps
pat9000 profile

My local LLM eval hid four token caps

#localllm #ollama #rtx5090 #5090reports
Comments
4 min read
Mojo Is Open Source. What Local AI Builders Need to Know
pat9000 profile

Mojo Is Open Source. What Local AI Builders Need to Know

#localai #mojo #gpuprogramming #opensource
Comments
5 min read
A GPU Driver Is Not a Local LLM Benchmark
pat9000 profile

A GPU Driver Is Not a Local LLM Benchmark

#localllm #rtx5090 #llmbenchmarks #nvidiadriver
Comments
4 min read
The 26B Model Hit the Cap. The 8B Finished.
pat9000 profile

The 26B Model Hit the Cap. The 8B Finished.

#localllm #ollama #benchmarking #rtx5090
Comments
4 min read
My Local LLM Writer Failed Its Own Word-Count Gate
pat9000 profile

My Local LLM Writer Failed Its Own Word-Count Gate

#localllm #ollama #gemma #evalgates
Comments
4 min read
A 27B Model Fit on an 8 GB GPU. It Was Slow.
pat9000 profile

A 27B Model Fit on an 8 GB GPU. It Was Slow.

#localllm #ollama #qwen #vram
Comments
5 min read
Unload Local LLMs After Every Test
pat9000 profile

Unload Local LLMs After Every Test

#localllm #ollama #rtx3070 #5090reports
Comments
4 min read
How I Test a 30B Local Model Before I Load It
pat9000 profile

How I Test a 30B Local Model Before I Load It

#localai #localllm #rtx5090 #ollama
Comments
5 min read
How I Benchmark Local LLMs Before I Trust Them
pat9000 profile

How I Benchmark Local LLMs Before I Trust Them

#localllm #benchmarking #rtx5090 #5090reports
Comments
4 min read
When a 4B Local LLM Beats 26B on One Task
pat9000 profile

When a 4B Local LLM Beats 26B on One Task

#localllm #modelselection #rtx5090 #ollama
Comments
5 min read
Does Ollama Include That New llama.cpp Feature?
pat9000 profile

Does Ollama Include That New llama.cpp Feature?

#localllm #ollama #llamacpp #modeltesting
Comments
4 min read
My local models refused zero of 50 security tasks
pat9000 profile

My local models refused zero of 50 security tasks

#localllm #aisecurity #openweights #modelevaluation
Comments
5 min read
Prime Agent hit 95.5% on ARC-AGI-3. I did not install it.
pat9000 profile

Prime Agent hit 95.5% on ARC-AGI-3. I did not install it.

#aiagents #selfimprovingagents #agentsecurity #agentcostcontrol
Comments
6 min read
My Local LLM Got Faster After It Passed the Tests
pat9000 profile

My Local LLM Got Faster After It Passed the Tests

#localllm #rtx5090 #ollama #modelevaluation
Comments
4 min read
Chunk Size Is a Reliability Setting
pat9000 profile

Chunk Size Is a Reliability Setting

#localllm #reliability #docker #offlineai
Comments
5 min read
The faster local model run took 83x longer
pat9000 profile

The faster local model run took 83x longer

#localllm #llmbenchmarks #rtx5090 #ollama
Comments
5 min read
VRAM Fit Is Not Runtime Support
pat9000 profile

VRAM Fit Is Not Runtime Support

#localllm #5090reports #ollama #modelselection
Comments
4 min read
Why Local LLM Benchmarks Need Power Data
pat9000 profile

Why Local LLM Benchmarks Need Power Data

#localllm #gpubenchmarks #ollama #5090reports
Comments
4 min read
My 5090 benchmark was missing the field I needed most
pat9000 profile

My 5090 benchmark was missing the field I needed most

#localllm #llmbenchmarks #rtx5090 #ollama
Comments
4 min read
Search Old Results Before Publishing an LLM Test
pat9000 profile

Search Old Results Before Publishing an LLM Test

#localllm #modeltesting #ollama #rtx5090
Comments
5 min read
Build Local LLM Eval Data From Real Failures
pat9000 profile

Build Local LLM Eval Data From Real Failures

#localllm #modelevaluation #codingagents #ownedhardware
Comments
5 min read
How I Keep LLM Results Valid After a Driver Update
pat9000 profile

How I Keep LLM Results Valid After a Driver Update

#localllm #llmbenchmarks #rtx5090 #nvidiadriver
Comments
4 min read
How I Budget VRAM for Shared Local AI Workloads
pat9000 profile

How I Budget VRAM for Shared Local AI Workloads

#localllm #vram #modelsizing #rtx5090
Comments
5 min read
Why I Benchmark Local LLM Input and Output Separately
pat9000 profile

Why I Benchmark Local LLM Input and Output Separately

#localllm #benchmarking #rtx5090 #ollama
Comments
5 min read
Why I Test 3 Workloads Before Sizing a Local LLM
pat9000 profile

Why I Test 3 Workloads Before Sizing a Local LLM

#localllm #rtx5090 #modelsizing #ollama
Comments
5 min read
Incident response needs a local model you already trust
pat9000 profile

Incident response needs a local model you already trust

#aisecurity #incidentresponse #localllm #openweights
Comments
6 min read
Local open-model agents just became a product category
pat9000 profile

Local open-model agents just became a product category

#localllm #openweights #aiagents #lmstudio
Comments
6 min read
Your local LLM benchmark is probably lying to you
pat9000 profile

Your local LLM benchmark is probably lying to you

#localllm #evals #llmbenchmarks #aiagents
Comments
5 min read
Test Retrieval Before Your Local LLM Writes
pat9000 profile

Test Retrieval Before Your Local LLM Writes

#localllm #retrieval #rtx5090 #5090reports
Comments
4 min read
Why I Did Not Promote My Smaller Local Model
pat9000 profile

Why I Did Not Promote My Smaller Local Model

#localllm #modelevaluation #rtx5090 #5090reports
Comments
4 min read
Build an AI Research Workbench on Your Own GPU
pat9000 profile

Build an AI Research Workbench on Your Own GPU

#localllm #rtx5090 #5090reports #localai
Comments
5 min read
Ollama num_batch: 256 Was My RTX 5090 Sweet Spot
pat9000 profile

Ollama num_batch: 256 Was My RTX 5090 Sweet Spot

#localai #ollama #localllm #rtx5090
Comments
4 min read
A 7 GB 27B Model Lost to My 17 GB Default
pat9000 profile

A 7 GB 27B Model Lost to My 17 GB Default

#localai #localllm #gguf #modelselection
1
Comments
5 min read
My Agents Have to Prove What They Did
pat9000 profile

My Agents Have to Prove What They Did

#aiagents #agentinfrastructure #showwork #localai
Comments 1
4 min read
Your Agent's Audit Trail Cannot Be Retrofitted
pat9000 profile

Your Agent's Audit Trail Cannot Be Retrofitted

#aiagents #agentverification #showwork #euaiact
Comments
5 min read
Ollama Raised $65M. What Builders Get
pat9000 profile

Ollama Raised $65M. What Builders Get

#localai #localllm #ollama #ownedhardware
1
Comments
5 min read
Why production AI is moving to open weights
pat9000 profile

Why production AI is moving to open weights

#localllm #openmodels #rtx5090 #aiinfrastructure
Comments
4 min read
Devlog 2026-07-13: local drive git corruption stalls the q
pat9000 profile

Devlog 2026-07-13: local drive git corruption stalls the q

#devlog #daily #bmdhodl
Comments
3 min read
My 8B Model Failed a 400-Word Task
pat9000 profile

My 8B Model Failed a 400-Word Task

#localllm #modelselection #ollama #rtx5090
Comments
5 min read
Q4km vs Q5km: Q4_K_M vs Q5_K_M
pat9000 profile

Q4km vs Q5km: Q4_K_M vs Q5_K_M

#localllm #gguf #quantization #llamacpp
Comments
3 min read
I Rebuilt 5,383 Embeddings After a Dimension Change
pat9000 profile

I Rebuilt 5,383 Embeddings After a Dimension Change

#localllm #embeddings #vectorsearch #ollama
Comments
5 min read
I Raised the QA Bar on Blog Infographics
pat9000 profile

I Raised the QA Bar on Blog Infographics

#localllm #5090reports #blog #matplotlib
Comments
4 min read
Pin Your Context Window or Pay the Reload Tax
pat9000 profile

Pin Your Context Window or Pay the Reload Tax

#localllm #ollama #numctx #localaiagents
Comments
5 min read
Preview Retrieval Before Your Local LLM Runs
pat9000 profile

Preview Retrieval Before Your Local LLM Runs

#localllm #retrieval #localaiagents #privateai
Comments
5 min read
Pin Your Local LLM Context Size Before You Build a Router
pat9000 profile

Pin Your Local LLM Context Size Before You Build a Router

#localllm #modelrouting #ollama #rtx5090
Comments
4 min read
When Claude hits a weekly limit, your agent fleet still needs a third CLI
pat9000 profile

When Claude hits a weekly limit, your agent fleet still needs a third CLI

#localllm #aiagents #gemini #agentops
Comments
4 min read
I built a self-improving code model on one RTX 5090. Here is what actually worked.
pat9000 profile

I built a self-improving code model on one RTX 5090. Here is what actually worked.

#localllm #aiagents #finetuning #rtx5090
Comments
5 min read
Will That Local Model Fit? Do the VRAM Math First
pat9000 profile

Will That Local Model Fit? Do the VRAM Math First

#localllm #vram #quantization #llamacpp
Comments
5 min read
Local LLMs Need a Timeout Before They Need a Bigger Model
pat9000 profile

Local LLMs Need a Timeout Before They Need a Bigger Model

#localai #ollama #5090reports #aiagents
Comments
5 min read
Your local LLM is not a worse Claude. It is a different tool.
pat9000 profile

Your local LLM is not a worse Claude. It is a different tool.

#localllm #openweights #aiagents #agentguard
Comments
5 min read
How I Gate a Local Coding Model Before I Trust It
pat9000 profile

How I Gate a Local Coding Model Before I Trust It

#localai #5090reports #ollama #codingagents
Comments
5 min read
A verifier loop beats a faster local model
pat9000 profile

A verifier loop beats a faster local model

#localai #localllm #5090reports #agentguard
Comments
4 min read
How I Make Local Model Runs Fail Safely On A 5090
pat9000 profile

How I Make Local Model Runs Fail Safely On A 5090

#localai #rtx5090 #qlora #localllm
Comments
5 min read
loading...