Skip to content
Language models

LFM-3-Coder-Claude-Distill

a compact model for coding and logic

2 min readupdated

This lightweight 1.73B-parameter language model is fine-tuned for programming tasks and step-by-step reasoning. Built on LiquidAI/LFM3-1.7B, it was distilled (transferring capabilities from a larger model to a smaller one) using reasoning traces and responses from the Claude model family — Sonnet, Opus, and Fable. Unlike the base model, this fine-tune specifically learns to build a chain of thought before outputting code. It is a solid fit for developers who need a fast local assistant for basic coding on modest GPUs.

Key features

  • Reasoning-focused: trained to break down problems step by step, mimicking Claude's analytical and response-structuring style.
  • High inference speed: thanks to its compact 1.73B size, it can hit over 200 tokens per second on an RTX 4060.
  • Ready-to-use weight formats: the repository includes both full FP16/BF16 safetensors weights and a quantized GGUF version (Q4_K_M) for easy local deployment.
  • Permissive license: released under Apache-2.0, allowing free use in both personal and commercial projects.

What you can do

  • Load the GGUF version (Q4_K_M) into a local client for tasks like code autocompletion or explaining short English code snippets on a laptop or PC with an entry-level GPU.
  • Integrate the model into a local script or side project via transformers or Unsloth to generate code along with step-by-step logic breakdowns.
  • Run it in 4-bit quantization to quickly prototype pipelines that demand high token generation speeds with minimal VRAM usage.

Example

from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Schnuckade/LFM-3-Coder-Claude-Distill",
    max_seq_length = 4096,
    load_in_4bit = True,
)
Read moreDetails
Base model: LiquidAI/LFM3-1.7B (1.73B parameters); training framework: Unsloth.
Generation speed: over 200 tokens/s on an RTX 4060.
Language: English (en); max context length in the example: 4096 tokens.
No benchmarks are listed on the model card.
Discuss Telegram 1 371

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    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.

    # ITLL/LFM-3-Coder-Claude-Distill
    # ===== Tokens: paste yours between the quotes =====
    export HF_TOKEN=""        # huggingface.co/settings/tokens, for gated models
    # ==========================================
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download ITLL/LFM-3-Coder-Claude-Distill --local-dir ~/models/LFM-3-Coder-Claude-Distill
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

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 LFM-3-Coder-Claude-Distill (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

ITLL/LFM-3-Coder-Claude-Distill
Author
ITLL
Category
Language models
Released
Posted

On NeyroLearn

Hugging Face

3711

Ask AI about this model

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

Comments