win10/GPT-OSS-30B-Preview
Partial fine-tune (fewer layers) vs unsloth/gpt-oss-20b-BF16. unsloth/gpt-oss-20b-BF16 matches part of win10/GPT-OSS-30B-Preview (411 of its 615 tensors; the rest of win10/GPT-OSS-30B-Preview has no counterpart). On the shared part: Partial change: 63% of the weights changed (attention, mlp, norms), in 16 of 24 layers.
Tensors changed: Embeddings 0/1, Norms 32/49, MLP 96/144, Attention 128/216, lm_head 0/1
2 derivatives checked
- mradermacher/GPT-OSS-30B-Preview-i1-GGUF: Quantization
- mradermacher/GPT-OSS-30B-Preview-GGUF: Quantization