LFM2.5-350M

LFM2.5-350M (Apple Silicon)

TheStage Apple SDK

Original model: LiquidAI/LFM2.5-350M (Liquid AI)

On-device chat LLM (Liquid LFM2.5 350M) compressed and packaged by TheStage AI for Apple Silicon. Strong quality/speed trade-off for assistants that need more capacity than 230M.

Overview

---

This repo hosts CoreML / MLX engine bundles for the TheStage Apple SDK. Point engines_path at the repo id; the SDK downloads, caches, and loads the engines automatically. Chat template, EOS / stop tokens, and KV-cache horizon all come from the bundle.

System Requirements

---

Simulator is not supported — run on real Apple Silicon hardware.

TheStage AI Access Token

---

Create a token at app.thestage.ai and pass it to the SDK:

try await TheStageAI.shared.initialize(apiToken: "th_…")

await TheStageFlutterSDK.initialize(api_token: 'th_…');

Token is checked online in initialize (once per app process when reachable). Inference runs fully on-device. Offline initialize fails — reconnect and call initialize again.

TheStage Apple SDK

---

Docs: TheStage Apple SDK · LLM · Repo: TheStage Apple SDK

Installation (SwiftPM)

In Xcode: File → Add Package Dependencies…, paste https://github.com/TheStageAI/AppleSDK.git, and add the TheStageSDK product. Or in Package.swift:

.package(
    url: "https://github.com/TheStageAI/AppleSDK.git",
    exact: Version(1, 1, 0)
)

Installation (Flutter / iOS)

# pubspec.yaml
dependencies:
  thestage_apple_sdk:
    git:
      url: https://github.com/TheStageAI/AppleSDK.git
      path: plugin/thestage_apple_sdk
      ref: 1.1.0

Swift — batch

import TheStageSDK

try await TheStageAI.shared.initialize(apiToken: "th_…")

let llm = try await TheStageLLM(
    engines_path: "TheStageAI/LFM2.5-350M"
)

let result = llm.infer(
    prompt: "What is 2+2?",
    system_prompt: "You are a helpful assistant.",
    max_new_tokens: 64
)
print(result.text)

For sampling control, start from the bundle preset and override only what you need:

var config = llm.generation_defaults
config.max_new_tokens = 256
config.temperature = 0.7
config.top_p = 0.8
config.repetition_penalty = 1.1

let result = llm.infer(prompt: "List 5 facts about London.", config: config)

Swift — streaming

for await chunk in llm.infer_stream(
    prompt: "Tell me a short story.",
    max_new_tokens: 512
) {
    if chunk.is_final {
        print("\n--- \(chunk.tokens_per_second ?? 0) tok/s ---")
    } else {
        print(chunk.text, terminator: "")
    }
}

Flutter (iOS)

import 'package:thestage_apple_sdk/thestage_apple_sdk.dart';

await TheStageFlutterSDK.initialize(api_token: 'th_…');

await TheStageFlutterSDK.start_model(
  model_name: 'llm',
  engines_path: 'TheStageAI/LFM2.5-350M',
);

final result = await TheStageFlutterSDK.infer(
  model_name: 'llm',
  input_json: {
    'prompt': 'What is 2+2?',
    'system_prompt': 'You are a helpful assistant.',
    'max_new_tokens': 64,
  },
);
print(result[0]['text']);

Streaming:

final stream = TheStageFlutterSDK.infer_stream(
  model_name: 'llm',
  input_json: {'prompt': 'Tell me a short story.', 'max_new_tokens': 512},
);
await for (final chunk in stream) {
  if (chunk['is_final'] == true) break;
  final delta = chunk['delta'] as String?;
  if (delta != null) stdout.write(delta);
}

Inputs / outputs

On-device latency

Release build on Apple M2 Max (NPU / ANE), macOS 26.2 (guidance, not an SLA). Warmed decode, 256 new tokens:

Always measure release builds on device.

Acknowledgments

---

This work builds on LFM2.5-350M by Liquid AI: LiquidAI/LFM2.5-350M.

This Apple Silicon package is produced by TheStage AI; it is not an official Liquid AI release.

Links

---

• Original model: LiquidAI/LFM2.5-350M
• Platform: app.thestage.ai
• TheStage Apple SDK: github.com/TheStageAI/AppleSDK
• Docs: docs.thestage.ai
• Subscribe for updates: TheStageAI X
• Contact email: [email protected]