Tag: local-llm
- Strata on RX 6800: faster Qwen3.8-Flash-Next prompt processing
With an OCuLink-connected RX 6800 and 32 GB of RAM, Strata's Qwen3.8-Flash-Next Coder fixed all nine game bugs faster than Qwen3.8-27B. An FP32 SGEMM change raised prompt processing from 312 to 452 tokens per second.
- Coding on the RX 6800 while the Radeon 780M generates sprites
Qwen3.8-27B decoded about three times faster on the RX 6800 than on the Radeon 780M. I tested coding helpers on the 780M, then used it to generate sprites during a game build.
- Qwen3.8-27B on Radeon 780M: inference speed and game repair
Testing Qwen3.8-27B on a Radeon 780M with 24K context. MTP raised decode speed by 67%; a 20K-token input took about six minutes before output began.
- A Local LLM Harness for Building and Testing a Browser Game on a Radeon 780M
A local model pipeline on a Radeon 780M planned, built, tested, and repaired a mobile tower-defense game, exposing why executable browser evidence matters more than model self-review.
- A local 27B coding agent on the 780M: why it froze, and what unstuck it
Testing Qwen3.8-27B with DeepSeek Harness on a Radeon 780M. Large tasks stalled in reasoning; the goal tool let the agent split the work and continue.
- Minus Zero: running a spec-and-test codegen loop entirely on local models
A local codegen loop, escalating from qwen3:8b to qwen3.6:27b, reproduces its own past output, hits one real capability wall in a signed-zero bug, and ships eight tested functions into a working game.
