Skip to content
Language models

PocketLLM Model Collection

local LLMs curated for Android

2 min readupdated

This repository is an official collection of compact language models optimized to run directly on Android smartphones without an internet connection or cloud servers. Developer Amaresh Hebbar put this bundle together specifically for the PocketLLM mobile app, though the files also work with other local inference engines. Inside, you will find popular open models (Qwen 2.5, Gemma 1.1 and 3, Llama 3.2, SmolLM2, Phi-3.5 Mini) ranging from 0.5B to 4B parameters. Unlike desktop releases, all weights are pre-converted into lightweight GGUF (Q4_K_M quantization for llama.cpp) and MediaPipe (.bin with CPU and GPU support) formats, allowing them to run even on budget phones with 2 GB of RAM.

Key features

  • Three hardware tiers: ranges from ultra-light models under 1 GB (Qwen 2.5 0.5B, Gemma 3 1B) for low-end devices to larger Phi-3.5 Mini and Gemma 3 4B options for phones with 5–6+ GB of RAM.
  • Two mobile formats: includes .gguf files for llama.cpp (llama.rn) and .bin binaries for Google MediaPipe LLM Inference with CPU or GPU acceleration (Vulkan/OpenCL).
  • Ready-to-use selection guide: the model card provides a compatibility table matching models to device memory, from 2 to 8 GB of RAM.
  • Upcoming fine-tunes: the author announced upcoming custom fine-tunes of Llama 3.2 1B and Gemma 3 1B tailored for short chats and a 2,048-token context window.

What you can do

  • Run a chatbot in PocketLLM: download your chosen weights straight from the built-in Model Store and chat with the AI model offline.
  • Integrate into mobile projects: embed ready-made .gguf or .bin files into Android apps built on llama.rn or react-native-llm-mediapipe.
  • Match the model to your task: pick Qwen 2.5 0.5B for maximum speed on an older phone, or go with Phi-3.5 Mini for coding tasks on an 8 GB flagship.

Example

To download any file from the collection directly, use a standard link formatted like this:

[https://huggingface.co/AmareshHebbar/pocketllm-models/resolve/main/](https://huggingface.co/AmareshHebbar/pocketllm-models/resolve/main/);
Read moreDetails
File sizes: from 0.4 GB (Qwen 2.5 0.5B) to 2.8 GB (Gemma 3 4B).
Smartphone RAM requirements: at least 2 GB for entry-level models, 4+ GB for mid-range, and 5–6+ GB for larger models.
Licenses vary by base model: Apache 2.0 (Qwen, SmolLM2), MIT (Phi-3.5), plus specific terms for Google Gemma and Meta Llama 3.2.
Discuss Telegram 0 76

Will it run?

Runs in GGUF Q4_K_M

The most precise of the ready quantised files fits.

  • GGUF Q4_K_Mweights 9.4 GB · ready files11 GB

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: GGUF Q4_K_M — from RTX 2080 Ti

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

A ready script for the machine we advise above: paste it and the model downloads and starts.

  1. 1. Rent A30

    $0.26 an hour with the coupon. Pick the Ubuntu image when you create the machine.

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # kayradk/pocketllm-models — A30, GGUF Q4_K_M
    # NVIDIA driver: CUDA 13 needs 580+. If older or missing, the current nvidia-open is installed (keep yours: KEEP_DRIVER=1)
    DRV=$(nvidia-smi --query-gpu=driver_version --format=csv,noheader 2>/dev/null | head -1 | cut -d. -f1)
    case "$DRV" in ""|*[!0-9]*) DRV=0 ;; esac
    if [ "$DRV" -lt 580 ] && [ -z "${KEEP_DRIVER:-}" ]; then
      . /etc/os-release && \
      curl -fsSLO "https://developer.download.nvidia.com/compute/cuda/repos/ubuntu${VERSION_ID//./}/x86_64/cuda-keyring_1.1-1_all.deb" && \
      sudo dpkg -i cuda-keyring_1.1-1_all.deb && rm -f cuda-keyring_1.1-1_all.deb && \
      sudo apt-get update && \
      sudo DEBIAN_FRONTEND=noninteractive apt-get install -y nvidia-open || \
      { echo "Could not install the NVIDIA driver: see the error above. To keep the current one, run with KEEP_DRIVER=1."; exit 1; }
      echo "NVIDIA driver updated. Reboot the machine (sudo reboot) and run this same script again."
      exit 0
    fi
    curl -fsSL https://ollama.com/install.sh | sh
    ollama run hf.co/kayradk/pocketllm-models:Q4_K_M
    Add more LoRAs to pocketllm-models
  3. 3. Use it

    The chat opens right in the terminal. Leave with /bye, start again with the same ollama run command.

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

  • MTP (multi-token prediction)0.34.0from November 16, 2026new release

    1.5–2× generation speed

    If the model has MTP layers (DeepSeek-V3/R1, Qwen3-Next, GLM, MiMo and others), it predicts several tokens per step itself.

    How to enable: vLLM / SGLang: --speculative-config '{"method": "mtp", "num_speculative_tokens": 1}'. Check the model card.

    Earlier generations (2)
    • EAGLE-3 — Speculative decoding with a small draft head. For models without their own MTP layers.
    • EAGLE-2 / Medusa — Earlier draft-head generations. Still work if that is all you find.
  • FP8 KV cache0.34.0from November 16, 2026new release

    2× more context in the same memory

    Keeps the attention cache in 8 bits, so long contexts fit where they did not.

    How to enable: vLLM: --kv-cache-dtype fp8; llama.cpp: -ctk q8_0 -ctv q8_0.

  • vLLM / SGLang0.34.0from November 16, 2026new release

    many times more tokens/s on a server

    Server engines: batching, FlashAttention, FP8 and speculative decoding out of the box.

    How to enable: vllm serve <model> — an OpenAI-compatible API.

    Earlier generations (1)
    • llama.cpp — GGUF models on any GPU, Mac or even CPU; some layers can stay in RAM.

Model family

New posts about versions of pocketllm-models (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

kayradk/pocketllm-models
Author
kayradk
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

760

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Language models

Emotion-S1

Emotion-S1