model

name	model
description	Algorithm/model development and fine-tuning skill. Use for tasks like dataset design/cleaning, supervised fine-tuning (SFT), preference optimization (DPO/RLHF concepts), LoRA/QLoRA, training configs, evaluation (offline/online), safety checks, deployment packaging, and cost/performance trade-offs.

model

Use this skill for 算法/模型开发/模型微调：从数据到训练到评测再到上线。

Defaults / assumptions to confirm

• Goal: improve quality, reduce cost/latency, add domain knowledge, safety alignment?
• Base model and license constraints
• Hardware: local GPU / multi-GPU / cloud
• Target inference stack (vLLM, TGI, llama.cpp, etc.)

Workflow

1. Define the objective and success metrics

• Task definition and input/output format.
• Primary metrics (task-specific) + guardrails (safety, latency, cost).
• Failure analysis categories (hallucination, format errors, refusal, toxicity).

2. Data strategy (most important)

• Collect/curate dataset; define labeling guidelines.
• Remove duplicates, leakage, PII, and near-duplicates.
• Balance by scenario; ensure coverage of edge cases.
• Split train/val/test with strict leakage prevention.

3. Choose training approach

• SFT for instruction following and domain formatting.
• LoRA/QLoRA for efficient fine-tuning (default for most cases).
• DPO/Preference tuning when “style/quality preference” is the target.
• Avoid fine-tuning when RAG or prompting solves it cheaper.

4. Training setup

• Pick tokenizer/model family compatibility.
• Hyperparameters: LR, batch size, sequence length, warmup, weight decay.
• Checkpoints and resume strategy; deterministic seeds.
• Track experiments (configs, metrics, artifacts).

5. Evaluation

• Offline eval set: small but representative; include hard negatives.
• Automatic metrics where meaningful; human eval for subjective qualities.
• Regression tests: keep a fixed “golden set” across iterations.

6. Safety & compliance

• Filter sensitive data; define refusal policy and tests.
• Measure unsafe outputs; create adversarial eval prompts.

7. Deployment

• Export adapters/merged weights; document inference requirements.
• Quantization plan if needed; benchmark latency and throughput.
• Monitor in production: quality signals, drift, safety incidents.

Outputs

• Data spec: sources, schema, labeling rules, splits.
• Training plan: method (SFT/LoRA/DPO), configs, compute estimate.
• Eval plan: datasets, metrics, sampling, acceptance thresholds.
• Deployment plan: packaging, quantization, benchmarks, monitoring.