Kwai-Keye/Thyme-RL

Full fine-tune vs Qwen/Qwen2.5-VL-7B-Instruct. Fine-tune: 92% of the weights changed, across 28 of 32 layers.

Tensors changed: Embeddings 1/2, Norms 55/122, MLP 88/280, Attention 196/324, lm_head 1/1

2 derivatives checked

• mradermacher/Thyme-RL-i1-GGUF: Quantization (fewer layers)
• mradermacher/Thyme-RL-GGUF: Same weights, converted (fewer layers)

2 other repos hold byte-for-byte identical weight files.

On the Hub · Downloads on Model Pulse