togethercomputer/Tev1-0.8B-experimental
LoRA-style vs Qwen/Qwen3.5-0.8B. Looks like a merged LoRA (or a fine-tune with frozen norms): only the weight matrices of down_proj, gate_proj, in_proj_a, in_proj_b, in_proj_qkv, in_proj_z, k_proj, o_proj, out_proj, q_proj, up_proj, v_proj changed, in 24 of 24 layers; norms, biases and embeddings are untouched.
Tensors changed: Attention 114/220, Norms 0/135, Embeddings 0/5, MLP 72/127, other 0/1
3 derivatives checked
- DreamBlooms/Tev1-0.8B-experimental-GGUF: Same weights, converted
- SirSahOl/Tev1-0.8B-experimental-chat-mlx-16bit: Partial fine-tune (fewer layers)
- gavinmroy/Tev1-0.8B-experimental-mlx: Partial fine-tune (fewer layers)