dampflokfreund

• Issues with Gemma 4 tool calling - abrupt gen ending despite the model telling me it wants to do X.

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 22 comments

• Enough with perplexity and KLD! BenchLocal benchmarks real use cases and is easy to use for everyone

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 0 comments

• Bartowski vs Unsloth for Gemma 4

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 79 comments

• Notice Qwen 3.5 reprocessing the prompt every time, taking long to answer for long prompts? That's actually because of its architecture.

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 48 comments

• Qwen 3.5 MXFP4 quants are coming - confirmed by Junyang Lin

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 70 comments

• In my experience, the QAT Gemma 3 quants by stduhpf still perform the best.

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 28 comments

• 5060 chokes on VRAM just by playing Cyberpunk with raytracing enabled at 1080p with DLSS!

  Posted by dampflokfreund@reddit | hardware | View on Reddit | 195 comments

• One of the most annoying symptoms of low VRAM is not talking about nearly enough

  Posted by dampflokfreund@reddit | hardware | View on Reddit | 4 comments

• Honestly, THUDM might be the new star on the horizon (creators of GLM-4)

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 68 comments

• PSA: Gemma 3 QAT gguf models have some wrongly configured tokens

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 50 comments

• Every time

  Posted by dampflokfreund@reddit | LocalLLaMA | View on Reddit | 0 comments

• TechPowerUp-Benchmark: 5700 XT starts Alan Wake 2, but it is significantly slower than even a RTX 3050 due to lack of Mesh Shading.

  Posted by dampflokfreund@reddit | hardware | View on Reddit | 149 comments