Embarcadero - Kai 'Summer' Wave

The music at 20:00 is super loud, but only for a couple of minutes.

Yes, sorry about that. The video itself that I played had background music but it seems that even though I had broadcast platform’s volume slide down to less than 30% it was still insanely loud. In the end I muted the sound on the video playback tool.

I’ll be raising a ticket with Re:stream because that’s a new change they have made (the pros and cons of a web interface) and it’s pretty dumb. :sob:

1 Like

@ianbarker I feel honoured.

Thank you for chasing down the rabbit hole I partially excavated :slight_smile: (From 50:00)

It was maybe 1am when you started … and while I was listening to the webinar … I drifted down the path to the land of nod. :sleeping_face:

Yesterday I made a video, that needs re-doing (because Ollama was rearranging my list of models on me).
I started trying out all the local models I had to hand on :

  • first, questions about dependency injection :

    What is the smallest, simplest, lightest, beginner example of (maybe Spring4D) dependency injection you can create? The language is Delphi. label any code sections as pascal for syntax highlighting.

  • then, when it showed up as a weakness :

    how is an interface delegated to a property in delphi? (it uses the implements keyword)

I turned off internet access to the models, so they had to answer from their own “knowledge”.

i5-11400, 32GB ram, RTX 3070 (8GB) & GTX1070 (8GB)

I arranged the screen to show load on the two GPUs, the CPUs, the memory usage & model distribution across CPU/GPU.

Currently uploading : https://youtu.be/ABl9tsoOqpA

codellama:13b 9f438cb9cd58 7.4 GB 47 minutes ago
deepcoder:14b 12bdda054d23 9.0 GB 55 minutes ago
deepseek-coder-v2:latest 63fb193b3a9b 8.9 GB 2 days ago
deepseek-coder:33b acec7c0b0fd9 18 GB 37 minutes ago
deepseek-coder:6.7b ce298d984115 3.8 GB 2 days ago
deepseek-r1:8b 6995872bfe4c 5.2 GB 9 months ago
gemma3:12b f4031aab637d 8.1 GB 9 months ago
gemma4:12b 4eb23ef187e2 7.6 GB 16 hours ago
gemma4:26b 5571076f3d70 17 GB 21 hours ago
gemma4:31b 6316f0629137 19 GB 21 hours ago
gemma4:e4b c6eb396dbd59 9.6 GB 16 hours ago
llama3.1:8b 46e0c10c039e 4.9 GB 8 months ago
mistral-nemo:latest e7e06d107c6c 7.1 GB 59 minutes ago
qwen2.5-coder:14b 9ec8897f747e 9.0 GB 4 weeks ago
qwen3-coder:30b 06c1097efce0 18 GB 30 minutes ago
qwen3.5:9b 6488c96fa5fa 6.6 GB 17 hours ago
qwen3.6:latest 07d35212591f 23 GB 17 hours ago
qwen3.8:27b 22130167c4c2 17 GB 7 minutes ago
qwen3:30b ad815644918f 18 GB 27 hours ago
qwen3:4B-Instruct-2507-q4_k_m 0edcdef34593 2.5 GB 4 weeks ago
qwen3:4b 359d7dd4bcda 2.5 GB 4 weeks ago
qwen3:8b 500a1f067a9f 5.2 GB 4 weeks ago

Claude, Claude, Claude … :face_with_raised_eyebrow:

Paul wants to see thinking vs answering time and I want to see you juggling jelly. Win!

1 Like

Your wish is my command… :zany_face:

2 Likes

I asked what memory is taken up by a 16k context window, for an 8b 14b 22b model …