Tweeties/tweety-7b-tatar-v24a
Partial fine-tune vs mistralai/Mistral-7B-Instruct-v0.2. Partial change: 16% of the weights changed (attention, embeddings, lm_head, mlp, norms), in 4 of 32 layers.
Tensors changed: Embeddings 1/1, Norms 8/65, MLP 12/96, Attention 16/128, lm_head 1/1
1 derivatives checked
- mradermacher/tweety-7b-tatar-v24a-GGUF: Same weights, converted