Skills DirectorySkills Directory
SkillsLearnSecurityCategoriesDocsCommunityBlog
Sign InSubmit Skill
Skills Directory

Security-tested agent skills for Claude, coding agents, and AI workflows.

Directory

  • Browse Skills
  • All Skills A–Z
  • Claude Skills
  • Claude Code Skills
  • Agent Skills
  • Categories
  • Submit a Skill

Learn

  • Learn Hub
  • Install Claude Skills
  • Write SKILL.md
  • Skills vs MCP
  • Directories Compared

Security

  • Security
  • Methodology
  • Secure Claude Skills
  • Security Badges

Company

  • About
  • Community
  • Blog
  • API Docs
  • Advertise

2026 Skills Directory. All rights reserved.

Back to skills

Dflash Mlx Speculative Decoding

ASecurity

Lossless DFlash speculative decoding for MLX on Apple Silicon — 1.7–4x faster LLM inference using block diffusion drafting with target model verification.

81 stars
0 votes
0 copies
0 views
Added 9/19/2026
ai-agentspythongobashapi

Works with

cliapi

Security Analysis

A96/100
mediumInstalls packages at runtime which could introduce malicious dependencies

Scanned 9/19/2026

Install to Claude Code

$npx -y skills add reason-machines/trending-skills --skill dflash-mlx-speculative-decoding --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dflash Mlx Speculative Decoding?

Add the live security badge to your README — it updates automatically with every re-scan.

Security grade badge for Dflash Mlx Speculative Decoding
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/reason-machines-dflash-mlx-speculative-decoding/badge)](https://www.skillsdirectory.com/skills/reason-machines-dflash-mlx-speculative-decoding)

More formats (shields.io, HTML) on the badges page.

Download Zip
Files
SKILL.md
---
name: dflash-mlx-speculative-decoding
description: Lossless DFlash speculative decoding for MLX on Apple Silicon — 1.7–4x faster LLM inference using block diffusion drafting with target model verification.
triggers:
  - use dflash for faster inference
  - speed up mlx generation with speculative decoding
  - dflash speculative decoding apple silicon
  - run dflash-serve openai compatible server
  - benchmark dflash vs baseline mlx
  - set up dflash with qwen model
  - dflash draft model auto resolution
  - faster token generation on apple silicon mlx
---

# dflash-mlx Speculative Decoding

> Skill by [ara.so](https://ara.so) — Daily 2026 Skills collection.

DFlash implements lossless speculative decoding for MLX on Apple Silicon. A small draft model (~1B params) generates 16 tokens in parallel using block diffusion; the target model verifies all 16 in a single forward pass. Tokens are only emitted after target verification — output is lossless (every token is the target model's greedy argmax).

**Typical speedups**: 1.7x–4.1x over baseline `mlx_lm` depending on model size and context length. Acceptance rates hover around 87–90% for Qwen3.5 models.

## Installation

```bash
pip install dflash-mlx

# or isolated install
pipx install dflash-mlx
```

Requires Python 3.10+, MLX 0.31.1+, Apple Silicon Mac.

## Key CLI Commands

### Generate text

```bash
# Auto-resolve draft model from registry
dflash --model Qwen/Qwen3.5-9B --prompt "Explain backpropagation"

# Explicit draft model
dflash --model Qwen/Qwen3.5-9B \
       --draft z-lab/Qwen3.5-9B-DFlash \
       --prompt "Explain backpropagation"

# Disable EOS (useful for benchmarking fixed token counts)
dflash --model Qwen/Qwen3.5-9B --prompt "..." --max-tokens 1024 --no-eos
```

### OpenAI-compatible server

```bash
# Basic server
dflash-serve --model Qwen/Qwen3.5-9B --port 8000

# With explicit draft
dflash-serve --model Qwen/Qwen3.5-9B \
             --draft z-lab/Qwen3.5-9B-DFlash \
             --port 8000

# Disable thinking/reasoning tokens (Qwen3.5 thinking models)
dflash-serve --model Qwen/Qwen3.5-9B --port 8000 \
  --chat-template-args '{"enable_thinking": false}'

# Raise fallback threshold for longer prompts (large models)
dflash-serve --model mlx-community/Qwen3.5-35B-A3B-4bit --port 8000 \
  --chat-template-args '{"enable_thinking": false}' \
  --dflash-max-ctx 16384
```

### Benchmark

```bash
dflash-benchmark \
  --model Qwen/Qwen3.5-9B \
  --draft z-lab/Qwen3.5-9B-DFlash \
  --prompt "The function f satisfies..." \
  --max-tokens 1024 \
  --repeat 3 \
  --no-eos
```

Outputs per-run JSON reports with tok/s, acceptance rate, and speedup vs baseline.

## Supported Model Pairs

| Target Model | Draft Model |
|---|---|
| `Qwen/Qwen3.5-4B` | `z-lab/Qwen3.5-4B-DFlash` |
| `Qwen/Qwen3.5-9B` | `z-lab/Qwen3.5-9B-DFlash` |
| `mlx-community/Qwen3.5-27B-4bit` | `z-lab/Qwen3.5-27B-DFlash` |
| `mlx-community/Qwen3.5-35B-A3B-4bit` | `z-lab/Qwen3.5-35B-A3B-DFlash` |

Draft models are auto-resolved from a registry — no `--draft` flag needed for listed pairs. Models without a matching draft are rejected at startup.

## Python API Usage

### Streaming generation

```python
from dflash_mlx import DFlashRuntime

runtime = DFlashRuntime.from_pretrained(
    model="Qwen/Qwen3.5-9B",
    draft="z-lab/Qwen3.5-9B-DFlash",  # optional, auto-resolved
)

prompt = "Explain the Pythagorean theorem step by step."

for token_text in runtime.stream_generate(
    prompt=prompt,
    max_tokens=512,
    use_chat_template=True,
):
    print(token_text, end="", flush=True)
print()
```

### Full generation with stats

```python
from dflash_mlx import DFlashRuntime

runtime = DFlashRuntime.from_pretrained(model="Qwen/Qwen3.5-9B")

result = runtime.generate(
    prompt="What is speculative decoding?",
    max_tokens=256,
    use_chat_template=True,
)

print(result.text)
print(f"Tokens/sec: {result.tokens_per_second:.2f}")
print(f"Acceptance rate: {result.acceptance_rate:.2%}")
print(f"Total tokens: {result.total_tokens}")
```

### Custom draft block size and context

```python
from dflash_mlx import DFlashRuntime, DFlashConfig

config = DFlashConfig(
    draft_block_size=16,      # tokens drafted per speculative step
    max_ctx=8192,             # max context length before fallback
    enable_tape_replay=True,  # GatedDeltaNet recurrent rollback
    jit_sdpa=True,            # custom Metal SDPA for long contexts
)

runtime = DFlashRuntime.from_pretrained(
    model="mlx-community/Qwen3.5-27B-4bit",
    config=config,
)
```

### OpenAI client against dflash-serve

```python
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="not-needed",  # dflash-serve does not require auth by default
)

# Non-streaming
response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=[
        {"role": "user", "content": "Explain gradient descent."}
    ],
    max_tokens=512,
)
print(response.choices[0].message.content)

# Streaming
stream = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=[{"role": "user", "content": "Write a haiku about silicon."}],
    max_tokens=128,
    stream=True,
)
for chunk in stream:
    delta = chunk.choices[0].delta.content
    if delta:
        print(delta, end="", flush=True)
print()
```

### Tool calling (via dflash-serve)

```python
import json
from openai import OpenAI

client = OpenAI(base_url="http://localhost:8000/v1", api_key="not-needed")

tools = [
    {
        "type": "function",
        "function": {
            "name": "get_weather",
            "description": "Get current weather for a city",
            "parameters": {
                "type": "object",
                "properties": {
                    "city": {"type": "string", "description": "City name"},
                },
                "required": ["city"],
            },
        },
    }
]

response = client.chat.completions.create(
    model="Qwen/Qwen3.5-9B",
    messages=[{"role": "user", "content": "What's the weather in Tokyo?"}],
    tools=tools,
    tool_choice="auto",
)

tool_call = response.choices[0].message.tool_calls[0]
print(f"Function: {tool_call.function.name}")
print(f"Args: {json.loads(tool_call.function.arguments)}")
```

## Common Patterns

### Side-by-side demo (baseline vs DFlash)

```bash
PYTHONPATH=. python3 -m examples.demo --mode dflash \
  --target-model Qwen/Qwen3.5-9B \
  --draft-model z-lab/Qwen3.5-9B-DFlash \
  --prompt "Solve: f(x) + f(y) = f(x+y) - xy - 1" \
  --max-tokens 2048 \
  --no-eos
```

### Integrating with Open WebUI

1. Start `dflash-serve --model Qwen/Qwen3.5-9B --port 8000`
2. In Open WebUI settings → Connections → add OpenAI API with URL `http://localhost:8000/v1`
3. Select model `Qwen/Qwen3.5-9B` in the chat UI

Works the same for Continue, aider, OpenCode, and any OpenAI-compatible client.

### Override draft for unsupported models

```bash
# Force a custom draft — bypasses registry check
dflash --model my-org/MyCustomModel \
       --draft my-org/MyCustomModel-DFlash \
       --prompt "Hello"
```

### Disable thinking tokens for Qwen3.5

```bash
# CLI
dflash --model Qwen/Qwen3.5-9B \
       --chat-template-args '{"enable_thinking": false}' \
       --prompt "What is 2+2?"

# Server
dflash-serve --model Qwen/Qwen3.5-9B \
             --chat-template-args '{"enable_thinking": false}' \
             --port 8000
```

## Architecture Notes

- **Tape-replay rollback**: For hybrid GatedDeltaNet + attention models (Qwen3.5), dflash records an innovation tape during verify and replays only accepted steps via a custom Metal kernel — avoids full state snapshots.
- **JIT SDPA 2-pass**: For contexts ≥ 1024 tokens, a custom Metal attention kernel maintains numerical alignment with stock MLX attention.
- **Greedy acceptance**: Keeps the longest correct prefix from the 16 drafted tokens, rejects the rest. No temperature/sampling on verification — strictly lossless.
- **Qwen3 (pure attention)** models work but don't benefit from tape-replay rollback (that's GatedDeltaNet-specific).

## Troubleshooting

**Model rejected at startup**
```
Error: No DFlash draft found for model 'org/ModelName'
```
→ Pass `--draft org/ModelName-DFlash` explicitly, or use a model from the supported pairs table.

**Low acceptance rate (< 80%)**
- Usually caused by very long context (4096+). Try `--dflash-max-ctx 8192` to extend the fallback threshold.
- Qwen3 (non-3.5) models have lower acceptance than Qwen3.5 hybrid models.

**Numerical divergence / output differs from pure AR**
- Expected behavior: "Output can still differ from pure AR because of MLX dispatch divergence, but no unverified token is ever emitted."
- If outputs seem wrong (not just different), ensure MLX 0.31.1+ is installed: `python -c "import mlx; print(mlx.__version__)"`

**Server not accepting connections**
```bash
# Check port is not in use
lsof -i :8000

# Bind to all interfaces for network access
dflash-serve --model Qwen/Qwen3.5-9B --port 8000 --host 0.0.0.0
```

**Out of memory with large models**
- Use 4-bit quantized variants: `mlx-community/Qwen3.5-27B-4bit` instead of the full model.
- The draft model loads alongside the target — budget ~1–2GB extra for the draft.

**Benchmark results JSON location**
```bash
ls benchmark/results/
# Per-run JSON with tok/s, acceptance rate, repeat measurements
```

Attribution

reason-machinesreason-machines
View sourceMore from reason-machines →
SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments (0)

No comments yet. Be the first to comment!

SSkills DirectorySkills Directory

Ship a skill? Prove it's safe.

Free 120-pattern security scan, letter grade, and an embeddable README badge.

Submit a skill

Related Skills

Caveman

Ultra-compressed communication mode. Cuts token usage ~75% by speaking like caveman while keeping full technical accuracy. Supports intensity levels: lite, full (default), ultra, wenyan-lite, wenyan-full, wenyan-ultra. Use when user says "caveman mode", "talk like caveman", "use caveman", "less tokens", "be brief", or invokes /caveman. Also auto-triggers when token efficiency is requested.

1023331 votes

Hyperplan

Adversarial multi-agent planning skill. Self-orchestrates 5 hostile category members (unspecified-low, unspecified-high, deep, ultrabrain, artistry) via team-mode for ruthless cross-critique debate, distills only the defensible insights, then MANDATORILY hands the distilled insight bundle to the `plan` agent for executable plan formalization. Use when planning needs maximum rigor and surfacing of weak assumptions, blind spots, and over-engineering. Triggers: 'hyperplan', 'hpp', '/hyperplan', ...

686011 votes

Mcp Code Execution

Routes multi-tool workflows through MCP servers for large datasets and pipelines. Use when Bash tool overhead is limiting throughput on data-heavy tasks.

3331 votes

catchup

Recovers prior coding-agent session context by running `catchup <agent> --since-compact`, which extracts a clean summary of a previous Codex, Claude Code, Antigravity, OpenCode, or Pi Agent session. Use when the user says "catch up", "what did the last session do", "get me up to speed", "I switched agents", or asks to recover/summarize a previous session before continuing. Do NOT use for the current conversation, git history, or any non-agent log.

611 votes

math-skill

A comprehensive mathematical reasoning skill for AI assistants — handles arithmetic to research-level problems with rigorous step-by-step reasoning, systematic verification, and transparent uncertainty handling

381 votes
View all in ai-agents →