Trelis/SmolLM-135M-Instruct-layer-pruned-90M-raw
Partial fine-tune (fewer layers) vs HuggingFaceTB/SmolLM-135M-Instruct. Trelis/SmolLM-135M-Instruct-layer-pruned-90M-raw matches part of HuggingFaceTB/SmolLM-135M-Instruct (182 of its 272 tensors; the rest of HuggingFaceTB/SmolLM-135M-Instruct has no counterpart). On the shared part: Partial change: 4% of the weights changed (attention, mlp, norms), in 1 of 20 layers.
Tensors changed: Embeddings 0/1, Norms 2/41, MLP 3/60, Attention 4/80
2 derivatives checked
- mradermacher/SmolLM-135M-Instruct-layer-pruned-90M-raw-i1-GGUF: Quantization
- mradermacher/SmolLM-135M-Instruct-layer-pruned-90M-raw-GGUF: Same weights, converted
2 other repos hold byte-for-byte identical weight files.