Tobi Lütke 102ff861d3 fix: use Qwen3 recommended sampling params to prevent repetition loops ... - Changed temperature from 0/0.1 to 0.7 (Qwen3 non-thinking mode default) - Added topK=20, topP=0.8 per Qwen3 docs - Added repeatPenalty with presencePenalty=0.5 for query expansion - Fixes infinite loop on acronyms like DHH, BFCM Qwen3 docs explicitly warn: 'DO NOT use greedy decoding, as it can lead to performance degradation and endless repetitions'		2026-02-01 03:24:20 +00:00
..
__init__.py	lots of training stuff	2026-01-31 23:02:23 +00:00
best_prompt_glm.txt	fix: use Qwen3 recommended sampling params to prevent repetition loops	2026-02-01 03:24:20 +00:00
best_prompt.txt	lots of training stuff	2026-01-31 23:02:23 +00:00
dspy_gepa.py	fix: use Qwen3 recommended sampling params to prevent repetition loops	2026-02-01 03:24:20 +00:00
example.py	lots of training stuff	2026-01-31 23:02:23 +00:00
generate.py	lots of training stuff	2026-01-31 23:02:23 +00:00
gepa_outputs_glm.jsonl	fix: use Qwen3 recommended sampling params to prevent repetition loops	2026-02-01 03:24:20 +00:00
gepa_outputs.jsonl	lots of training stuff	2026-01-31 23:02:23 +00:00
model.json	lots of training stuff	2026-01-31 23:02:23 +00:00
optimizer.py	lots of training stuff	2026-01-31 23:02:23 +00:00
score.py	lots of training stuff	2026-01-31 23:02:23 +00:00