tlg/prompts
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Neue Prompt-Dateien und neue Prompts in bestehenden Dateien

Browse Source
This commit is contained in:
rdms
2026-09-28 09:24:55 +02:00
parent d97414542b
commit 906760fd54
9 changed files with 3004 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Dirty-Prompts.md
+4
View File
@@ -1058,3 +1058,7 @@ Ich bedanke mich für Ihr Verständnis und Ihre Kulanz.
Mit besten Grüßen
Thomas Langer
---
The article `https://x.com/yume_arasaki/status/2102395182184534404` states EXL3 has a relatively good quality. I know that common 3-bit quantization has a severe impact on LLM inference while 4-bit quantization is acceptable. Is EXL3 really a game changer? Please explain; in case EXL3 is too new for your knowledge (your training data might be just outdated) do a research. Also find examples for recent LLMs like Qwen3.8-27B which compare intelligence of 16-bit, 8-bit, 4-bit and EXL3 quantizations.