mesolitica/Malaysian-Qwen2.5-7B-Dialect-Reasoning-GRPO
Full fine-tune vs mesolitica/Malaysian-Qwen2.5-7B-Reasoning-SFT. Fine-tune: 93% of the weights changed, across 28 of 28 layers.
Tensors changed: Embeddings 0/1, Norms 50/57, MLP 84/84, Attention 196/196, lm_head 1/1
2 derivatives checked
- mradermacher/Malaysian-Qwen2.5-7B-Dialect-Reasoning-GRPO-i1-GGUF: Quantization
- mradermacher/Malaysian-Qwen2.5-7B-Dialect-Reasoning-GRPO-GGUF: Same weights, converted