vLLM/.agents/skills/train-inference-consistency/SKILL.md

name, description

name	description
train-inference-consistency	Build and validate train-inference consistency (true on-policy RL), where vLLM rollout and the training forward give bitwise identical logprobs at high throughput. Use when aligning a training stack with vLLM batch-invariant inference, validating forward, backward and quantization contracts, or diagnosing logprob mismatch, layout and performance regressions between training and rollout.

Train-inference consistency skill

Make the training side and the inference side in RL produce bitwise identical logprobs for the same token (diff = 0), and, while keeping 0 diff, bring performance back close to the high-performance baseline.

How to enable batch-invariant inference in vLLM is in batch invariance; this skill covers making the training side match it.

Start here: train-inference consistency workflow (C0–C10). It answers "what to do next, and what counts as passing"; acceptance details are in validation.md, and the report template is report-template.md.

Find a guide by problem:

Guide	Questions it answers
Fault localization and mismatch bisect	Find the side first, bisect, classify before changing precision, save and compare protocol
Training forward	Module inventory and route (mLite / MCore / FSDP2), reproducing the inference forward by calling its kernels, layout and length, framework integration
Backward validation	When to reuse backward, gradient contract, module-level VJP comparison, proxy SFT three-way control
BI kernels	Which kernel each operator uses in BI mode and what it costs
Precision and quantization	The conversion chain master -> training forward weights -> exported bytes -> vLLM bytes; QAT/STE; known pitfalls
Rollout engine and sampling	Sampling RNG, prefix cache semantics, MTP constraints, R3 constraints, rollout replay
Performance and profiling	Workload and measurement basis, fair A/B, breakdown and ablation
Checkpoint and resume	Exact resume, derived state after restore, memory budget for colocated training and rollout
Environment provenance	Which version is actually running now; immutable delivery image
Recipe and acceptance	BI=1 / BI=0 arms in one recipe and one image; evaluation contract; paired acceptance

Process rules: evidence and experiments, code and delivery.

Note: this skill comes from the DeepSeek-V4 and Nemotron-H train-inference consistency work and covers train-inference consistency only. Day-0 model support, per-model case records and cluster-specific notes are not included.