Joakimpalm-Zen/Qwen3-4B-Runner-ToolUse-Q4_K_M
16.5M • Updated • 499
Xyntetik Research: LoRA Training on Quantized Weights
Adapters trained through the quantized GGUF that serves them, byte-reproducible, with the finding that a 4-bit merge erases the fine-tune.
Note The adapter, its merged Q8_0 and the three precision-study adapters.
Joakimpalm-Zen/Qwen3-4B-ToolUse-LoRA-training-record
Updated • 130 • 1Note Training data, provenance records, every held-out eval, the merge study and the logprob study.
Joakimpalm-Zen/Cloudflare-clef-IQ3_S-recovered-GGUF
27B • Updated • 1.61kNote Two LoRA recovery adapters trained with Runner on the project's own IQ3_S Clef backbone; house bar not met, stated on the card.