eeshansrivastava89

Running Qwen3.6 with Multi-Token Prediction in llama.cpp

Accurate as of May 18, 2026.

Multi-Token Prediction (MTP) uses the model's built-in prediction heads to draft multiple tokens in parallel, then verifies them against the main model. For Qwen3.6, this yields ~1.5–2× faster generation with no accuracy loss.

This guide covers the Qwen3.6 27B and Qwen3.6 35B-A3B (MoE) models. As of May 2026, MTP support is merged into llama.cpp — no fork required.

---

eeshansrivastava89

name	context-restore
description	Restore session context from previous session
disable-model-invocation	false

Context Restore

Resume work from where the last session ended. Read actual codebase along with this file.

eeshansrivastava89

name	context-save
description	Save session context for resuming later
disable-model-invocation	false

Context Save

Capture session state so the next session can resume seamlessly.

eeshansrivastava89

CLAUDE.md

Global instructions for Claude Code.

Guardrails

• NEVER modify CLAUDE.md — Read-only for Claude
• NEVER delete original content — Only reorganize or add
• NEVER make irreversible changes without approval
• NEVER commit unless explicitly asked