tianzl66/Llama-3.1-8B-Instruct-CommonSense170K-Spectral-Surgery-AllModules-8Plus2

Llama-3.1-8B-Instruct + Commonsense170K — Spectral Surgery (All Modules 8+2)

Spectral Surgery adapter for Llama-3.1-8B-Instruct fine-tuned on Commonsense170K.

Post-hoc Spectral Surgery applies Hybrid Newton-Schulz (HNS) editing directly to the LoRA adapter. No additional gradient-based training is performed during Spectral Surgery.

Adapter

• —Dataset: Commonsense170K
• —Training epochs: 2
• —LoRA rank: 16
• —LoRA alpha: 32
• —Target modules: qproj, kproj, vproj, oproj, gateproj, upproj, down_proj
• —HNS target modules: all_modules
• —Fast HNS steps: 8
• —Stable HNS steps: 2
• —See spectral_edit_meta.json for exact edit metadata.

Evaluation

Evaluation uses the Llama-3.1-Instruct tokenizer chat template, greedy decoding, max_new_tokens=8, the vLLM backend, max model length 2048, and seed 42.

Files

• —adapter_model.safetensors: PEFT LoRA weights
• —adapter_config.json: PEFT configuration
• —eval-commonsense8/summary.json: eight-task aggregate metrics
• —eval-commonsense8/summary.csv: compact task metrics
• —spectral_edit_meta.json: exact HNS edit metadata